0:00:04 Hi everyone, welcome to the A6NC podcast.
0:00:07 Today’s episode features Mark Andreessen interviewing Bobby Kotick,
0:00:12 CEO of Fortune 500 company Activision Blizzard, the largest game network in the world,
0:00:17 responsible for popular entertainment franchises such as Call of Duty, Candy Crush and World of
0:00:23 Warcraft. The discussion originally took place at our most recent annual innovation summit
0:00:28 and covers everything from the evolution of games in the 80s to the mergers and acquisitions that
0:00:33 created the company he runs today to trends in gaming, including touching on esports.
0:00:40 You can also find other podcasts and videos from this event at asixnz.com/summit.
0:00:45 Please note that the content here is for informational purposes only, should not be taken as legal
0:00:51 business tax or investment advice, or be used to evaluate any investment or security.
0:00:55 For more details, please also see asixnz.com/disclosures.
0:00:59 So, Bobby, it is really fun as a long-time video game aficionado. Really fun to have the chance
0:01:05 to talk to you today. I would love to start with your origin story, as they say in the superhero
0:01:09 business. So, part of your origin story, if I recall correctly, is that you started writing
0:01:14 software for the Apple II while you were still in college. So, my college roommate and I started
0:01:19 a company. He worked at Apple Computer in France. He was French and his summer internship,
0:01:25 he was working for a guy called Jean-Louis Gasset, and they had a prototype of the Lisa,
0:01:30 and the Lisa was the Mac before the Mac. It was the $10,000 version of the Mac.
0:01:38 So, he saw this prototype of the Lisa and thought for $10,000 this would be too expensive to turn
0:01:44 into a consumer product. So, he came back from his internship and said, “We should make Lisa-like
0:01:50 software and a mouse for the Apple II.” And we were in another sort of technology-related
0:01:54 business in our dorm room at the time, but we thought we were making hardware. I thought this
0:01:59 would be a better business, is make software. And we really thought if Steve Jobs is going to
0:02:04 appropriate all the great technology from Xerox Palo Alto Research Center, we should do the same,
0:02:11 but do it on a broader scale. So, we designed this mouse and a word processor and a spreadsheet
0:02:17 and a database all for the Apple II with a graphical user interface about a year before
0:02:21 the Macintosh was released. You also, I believe, credit, if I’m correct, Steve Jobs with convincing
0:02:25 you to drop out of college? I don’t know if it’s credit, but Steve Jobs heard about us.
0:02:29 Because when you started, you were in college. I was in college. I was making Apple II software,
0:02:36 and he heard about the software, and he called me and said, you know, the lady said, “Steve Jobs
0:02:40 calling.” I was like, “Okay, it’s one of the kids I grew up with from Rosalind Long Island, and this
0:02:46 is not even that good a joke.” Steve is like super famous, like Cover Time magazine. Yeah, this is
0:02:54 like 1983, late 1983. ’83, ’83, ’83. Right. Super famous. Yeah, and I’m like, “Yeah, right, whatever.”
0:02:59 So, I pick up the phone, he’s like, “Hi.” I said, “Hi.” He said, “This is Steve Jobs.” I said,
0:03:03 “Yeah.” Sure. And all I’m thinking is like, I wanted to say like, “I’m Oscar Robertson.” I was
0:03:09 a big Knicks fan. I thought like, so he starts like telling me, “You need to come to Cupertino,
0:03:14 and I really want to talk to you about this Jane thing that you made.” And so I go, and he shows
0:03:20 his prototype of the Macintosh, and like, I’ll never forget this moment. He unzips this little blue
0:03:24 bag from off of his table, and he takes it out, and he turns it on, and you see the hell out come
0:03:28 up on the screen, and I thought, “Wow, this is unbelievable. This is going to change computing.”
0:03:33 And I like, I still get the goosebumps of just thinking about it. And then he said, “Okay, now
0:03:40 show me yours.” And so like, I’ve had lugged this Apple II with like the 64K floppy drive,
0:03:44 and this mouse that we designed, and we put it on, and I show him the mouse, and he looks at it,
0:03:51 he’s like, “This is a piece of shit.” And he throws it on the floor, and he says, “You’re going to use
0:03:54 our mouse, and don’t ever think of using a two-button mouse. You’re going to use a one-button
0:03:59 mouse.” And then I show it to him, and the first thing he says is, “Wait a second. You select the
0:04:03 text, and then you select boldface?” I’m like, “Yeah.” He’s like, “No. We’re going to think about
0:04:08 verb noun versus noun verb in the way you actually boldface type.” And for 45 minutes, we had this
0:04:13 huge debate about how you boldface type. And at the end of it, he’s like, “You’re going to make
0:04:17 this for a new computer that we’re going to build called the Apple II GS, and we’re going to tell
0:04:21 you all about it, but you’re going to make this software for the Apple II. And we’re going to
0:04:27 give you a contract.” He goes, “This is a contract.” And then he comes to visit us in Ann Arbor,
0:04:35 Michigan, in our office above a Burger King. And the first thing he says is, “How the hell do you
0:04:43 work here? The smell of burgers comes up the elevator.” He’s like, “It was cheap rent. Nobody
0:04:49 else wanted to rent on top of a Burger King.” So at the end of the meeting, he says, “Do you have
0:04:56 any vegetarian restaurants here that we could go to for dinner?” And I said, “Yeah, I’m sure we do,
0:05:01 but I can’t go to dinner. I have a class.” He’s, “What do you mean?” I said, “I have a class.” He
0:05:06 said, “In what?” I said, “It’s a history of art makeup class.” He said, “What are you making up for?”
0:05:11 I said, “Well, I didn’t go to the class.” He’s like, “Well, why do you need to go to this?” I said,
0:05:17 “Well, I’m in college.” And he looks at me. He’s like, “What are you talking about? You have a
0:05:21 contract with Apple Computer, and we have a deadline for the Apple II GS. You can’t be in
0:05:25 college. You have employees. You have to work full-time.” He’s like, “Get out of college.”
0:05:30 And I said, “I can. I promise my parents I would finish college.” He said,
0:05:36 “No, you’re not finishing college. I will rip,” and I can’t say the word,
0:05:41 “I will rip this f-ing contract up right now if you don’t quit college.” So I quit the next day.
0:05:47 Did your parents believe the story? I didn’t tell them for about eight months. I felt like,
0:05:51 all right, I needed to get more progress in the business before my father would say,
0:05:56 “You’re an idiot. Now, you took money from this gambling guy and you quit college. You’re just a
0:06:04 loser.” So then fast forward a few years. So 1990, you bought 25% stake in Activision,
0:06:09 became CEO in 1991. So Activision, people may not know, Activision was a storied brand in video
0:06:13 games. Activision, I believe this is correct, was the first third-party developer of video games
0:06:19 in 1980. So it was a spin-off from Atari. It was for the top people at Atari. They got in
0:06:22 conflict with management. And this is a very big deal, number one, just because, you know, to quit
0:06:25 starting a company is a big deal. But also it’s just a big deal because like literally there had
0:06:29 not been a business of making video games for somebody else’s platform. Atari made all the
0:06:33 games for Atari. And now at this company, Activision, they had a run of hits, I guess in the 80s,
0:06:37 early 80s, you know, they made a bunch of the top games for Atari systems. And then they went
0:06:41 bankrupt or they were about to go bankrupt. They had a decade of pathetic performance and then
0:06:46 finally went bankrupt. Okay, got it. So as long as a whimper, not a bang. And so how did you get
0:06:53 from building mice in your above the Burger King to buying Activision? So in 1987, there were no
0:06:59 real video game hardware companies. And I played a lot of video games as a kid and I loved Activision
0:07:05 games. So they made the original Atari 2600 games that made the company so successful. Games like
0:07:12 Pitfall and River 8 and Kaboom. And these are just great games. And so in 1987, I thought there’s a
0:07:18 great opportunity to make video game hardware. And nobody is making video game hardware games are
0:07:24 just played on personal computers. Nintendo was just coming to the US. So there was no
0:07:31 dedicated video game hardware. My best friend from growing up had just started a hedge fund
0:07:36 with this guy from Texas named Richard Rainwater. And they were looking for investments.
0:07:42 And I went to him and I said, I have this idea. And in 1987, in October, the market had crashed.
0:07:46 It was like a 500 point market crash was Black Monday. Black Monday was one of the
0:07:53 biggest crashes in the market history since the Great Depression. And I had been working
0:07:58 making software for, among other companies, Apple and Commodore. And Commodore was this
0:08:06 $900 million revenues company at the time with $150 million market value. And so I went to
0:08:10 my best friend and I said, we should buy Commodore. They have this computer called the Amiga.
0:08:14 And it has a keyboard and a disk drive, but we should pull the keyboard and the disk drive out of
0:08:20 it. And it would be the first 16 bit video game system. And it was designed by Exitari engineers
0:08:25 and was made basically as a video game. It was a big leap forward graphics performance at the time.
0:08:31 68,000 microprocessor and dedicated graphics processors. And it was like a really innovative
0:08:39 idea. And my best friend at the time said, yeah, let’s try and do this. And so we tried. And
0:08:45 ultimately couldn’t persuade the chairman of Commodore to do the deal. But I just became then
0:08:53 fixated with being in the video game business. And my best friend at the time was Eddie Lampert,
0:08:57 who then went on instead of buying Commodore to buy Sears and Kmart. And he was a customer for a
0:09:04 little while. We stopped extending credit to him. I think four years ago, though. So I thought, okay,
0:09:10 we have to be in the video game business. And I had a little side business that was a licensing
0:09:18 company. We licensed characters. So one of our licensing partners was Nintendo. And we were
0:09:25 licensing Nintendo characters for bedsheets and lunchboxes. And I knew the Nintendo people. And
0:09:30 one day I was having a Nintendo meeting. And they said, have you ever thought about
0:09:35 Activision? And I said, yeah, I know the company well, I played all the games. And they said,
0:09:39 they’re not in really good shape. And they’re about to lose a patent infringement judgment that
0:09:47 will probably make them bankrupt. So you should consider buying Activision. So I bought a 25%
0:09:53 stake in Activision for $440,000. And I became the largest shareholder.
0:10:00 And it was insolvent. But I tried to get the CEO on the phone to tell him I was his new largest
0:10:04 shareholder. It’s a public company. It’s a public company with a market cap of $1.6 million. Yeah,
0:10:12 $1.6 million. And a patent infringement judgment that made it insolvent. And so I couldn’t… Here’s
0:10:17 this company that’s doing horribly, has all these great franchises. And the CEO wouldn’t return my
0:10:22 phone call. And so I kept calling, kept calling. And finally I just thought, I’ll just go to the lobby
0:10:29 of the building and tell him I’m there and see if he’ll see me. So I go and I’m waiting in the lobby.
0:10:34 And finally after three hours, he says, okay, I’ll come talk to you. And I talked to him and I
0:10:37 said, you know, we’re your largest shareholder. I have some really great ideas for you. I have
0:10:42 some game ideas I’d like to make. And, you know, I love some of the old properties. Maybe we can
0:10:47 figure out how to really get some of those properties back to being games. And he’s like,
0:10:52 well, thank you very much for visiting and nice to meet you. And I said, no, no, I’m like, when is
0:10:58 the next meeting? And he said, there’s not a next meeting, but we’re very happy to have you as
0:11:05 shareholders. And so I thought, well, how does that work? I own 25% of the company. I’m the
0:11:10 largest shareholder. We’re really a quarter of the furniture in the lobby. And he didn’t return my
0:11:14 phone calls for a little while. And then he agreed to have breakfast with me at this consumer
0:11:19 electronic show in Las Vegas. Obviously, Activision was then, you know, tremendously successful.
0:11:23 You then did one other really, really big deal that was transformative for the company, which was
0:11:27 and I don’t quite know how to describe it, but I think it’s a merger with Vivendi games
0:11:31 that resulted in Activision Blizzard, because Vivendi owned Blizzard, Blizzard, obviously,
0:11:35 World of Warcraft and all these other amazing properties. And then ultimately that partnership
0:11:38 on one, maybe you could tell us like that was a very, very big deal for you at the time.
0:11:44 How did that deal come about? So it was the spring of 2008. And I had a lot of anxiety about the
0:11:50 public markets and financial crisis is brewing. Yeah. And you could see there was a lot of
0:11:55 volatility instability. We were nervous, but we thought, you know, if we could buy something,
0:11:59 we had a big market value at the time. And I thought if we could buy something great,
0:12:04 that we really love, this would be a good time to do it. And you guys weren’t yet doing the massive
0:12:08 multiplayer. You guys weren’t yet doing like there was this big World of Warcraft had been a big hit
0:12:14 at that point. Huge hit. And we had explored doing a massively multiplayer persistent game.
0:12:19 There were a couple of other games before World of Warcraft that were massively multiplayer games,
0:12:25 but nothing that had had the success of WoW. And we looked and said, if we even could figure out
0:12:31 how to do it, it would take us five, six, seven years and a billion dollars. And the likelihood is
0:12:36 we wouldn’t do a good job of it. But it was a, and I knew the Blizzard team because we had worked
0:12:42 with them on, they were a contract development company in the early 1990s. And I knew the team
0:12:48 very well and really liked them. And I tried to recruit them out of the company. And I knew that
0:12:52 they had so much of a love and a passion for the company that no matter who owned it, they probably
0:12:57 wouldn’t leave. And I called the guys at Blizzard a bunch of times and said, we should truly try and
0:13:03 work this out. And they were stuck as a division of Vivendi Games, which was owned by Vivendi.
0:13:10 And you might describe what Vivendi was at this point. It was a big mess. And it was a former,
0:13:15 if I think it’s correct, it was a former public utility. Well, it was the water company of France
0:13:20 that this very, and by water and other things that flow through pipes. Yeah, water, other things that
0:13:26 flow through pipes. It was a collection of more industrial businesses that the, a man had taken
0:13:32 it over and decided he was going to become the media mogul of earth and bought universal studios.
0:13:38 And anything he could actually buy, he just bought. And then it got disassembled because
0:13:44 it was insolvent. And they ended up with a couple of businesses. The best one for them at the time
0:13:51 was probably Universal Music, which I think Lucine is here somewhere. So they had Universal Music.
0:13:58 They owned Morocco Telecom, Upstake. They owned SFR, which is a French mobile company.
0:14:04 They own Kennel Police. And somehow they managed as a part of a bunch of things to own
0:14:10 this games business, which included Blizzard. And I asked them to sell us Blizzard. And we didn’t
0:14:14 have any interest in the rest of their games business, but we wanted Blizzard. And they said,
0:14:20 no, repeatedly. And we offered them $4 billion and then $5 billion and then $6 billion and then $7
0:14:25 billion. They kept saying, no, they like video games. So we came up with this idea and we said,
0:14:31 how about this? We’ll stay a public company. You sell us Blizzard or give us Blizzard and $2
0:14:38 billion of cash, and we’ll give you 51% of the company. And my view was in 2008, even the biggest
0:14:45 institutional investors no longer were really long-term holders. And all the big institutional
0:14:49 investors were trading in and out of the stocks like they were hedge funds. So if we could get
0:14:56 Vivendi to own 51% of our company, we got Blizzard as a partner, we would have a great business and
0:15:01 a stable shareholder who would never sell our stock and would be enthusiastic about investing,
0:15:07 at least what they told us, investing with us for the future. So we went back and forth for a
0:15:13 long time, finally negotiated a deal where they would do that deal. And I almost blew the deal
0:15:19 in the worst way too. They had this beautiful headquarters in France, like the nicest building
0:15:23 in France. And the guy who had put the original Vivendi together, there were lots of these
0:15:29 beautiful French offices that had gardens on the top of their roof. He built a park
0:15:36 and like mature trees on the top of the Vivendi building. It was like a park with trees that
0:15:41 were all over the place. And one room was like a wine cellar and one room was this magnificent
0:15:46 dining room. And so we’re standing on the top of this roof overlooking the Arc de Triomphe
0:15:54 and the Eiffel Tower. And the chairman of the Vivendi says to me, “Bobby, this building,
0:15:59 it will be your home. This will be your place. You can do your business in France and you should
0:16:03 treat this building like your home. You can do anything you want with this building, but it will
0:16:08 be your place for Paris. You can make a business here. And it’s the most beautiful building in
0:16:13 Paris. It’s the most beautiful view in Paris and it’s for you. You can do anything you want with
0:16:19 it.” And I said, “Anything? You can do anything.” And I said, “Can I build 20 stories of condominiums?”
0:16:26 And he turned white and I could see the like, “Oh my God, who are we getting in business with love?”
0:16:31 I was going to say, “He hadn’t been briefed.” And I said, “No, no, Jean-Marie, I’m just kidding.
0:16:37 We would only build 10 stories.” But he still went through with the deal. And so we had five
0:16:45 wonderful years with them as our 51% shareholder until they were forced to sell our stake. But it
0:16:49 was the thing that actually allowed us to acquire a business. So it’s a big deal. I mean, it’s really
0:16:53 uncharacteristic for a company like this with a founder really of the modern business and then a
0:16:57 CEO like you to be willing to sign over control. So that story to me makes a lot of sense if you’d
0:17:05 said 49%. What was it about that deal that made you willing to literally sign up? Because it
0:17:09 worked out well with that. Was it a pretty big risk at the time or not? I didn’t really think I
0:17:16 was selling control. I think I was selling 51%, but I thought they really know nothing about video
0:17:22 games. What are they going to do? And I don’t think they’re going to interfere all that much. I was
0:17:26 actually wrong. They didn’t really have, you know, it’s like a corporate holding company. So they
0:17:32 were always trying to justify their value as a corporate holding company. And you know, we had
0:17:38 like, I remember the, and Lucien who’s here will attest to this is exactly what happened. But
0:17:43 they said we’re having a synergy meeting and all the business unit heads need to come together
0:17:50 for the synergy meeting. Now they own a stake in Morocco telecom. We didn’t do business in Morocco.
0:17:56 They own SFR, the French telephone company and mobile games wasn’t really a thing at that time.
0:18:01 They own Canal Plus. So a French TV network didn’t really have any applicability and
0:18:06 universal music where we did license some music for Guitar Hero, but other than that we had no
0:18:11 relationship and a broadband company in Brazil. So we all get together and have this big synergy
0:18:15 meeting. And then we had to go around the room and say the synergies that we identified between
0:18:23 each other. And they got to me and I said that Morocco telecom, we went to their cafeteria
0:18:30 and they have tagine. And we got the tagine recipe for our cafeteria,
0:18:42 which I thought was a great synergy because I like tagine, but there wasn’t really that much
0:18:48 synergy. So then they had to sell us back their stake. And then it was completely unwinding,
0:18:53 ultimately. And I guess they completely unwound in their own way. And everybody did well.
0:18:57 So it turned out. So give us a kind of a state of the video games industry today,
0:19:03 gigantic global phenomenon, lots and lots of change and flux, lots of potential controversy.
0:19:09 So I would say of the 28 years I’ve been doing Activision, 30-some-odd years I’ve been doing
0:19:16 software, I’ve never seen more opportunities than exist today. Markets that are opening. And you
0:19:22 think just 10 years ago, if you wanted to play video games, you either needed $1,000 PC or $300
0:19:27 or $400 video game console. And there really weren’t any other ways to play video games. But
0:19:34 phones have ushered in this whole new opportunity. And like for years, for most of the tenure that
0:19:41 I’ve had as CEO, we sold in developed countries to middle-class consumers on expensive devices.
0:19:47 Today, we sell in 196 countries around the world. We have 400 million customers.
0:19:56 And anyone in any socioeconomic strat can actually play games. I think that was the biggest shift
0:20:01 that took place is now you truly have a global market. The second thing that then happened is
0:20:08 when you started to see the games become more social experiences. I can use a headset. I can
0:20:12 talk to the person I’m playing with. I can play with somebody from anywhere around the world.
0:20:19 The introduction of the social experience was the true transformation to me of the opportunity.
0:20:24 And so where you look out in the world today, you have a global audience. You have this ability
0:20:31 to create this true social experience. And I remember years ago hearing this, and I’ll paraphrase,
0:20:36 but Mandela had this definition of sport, that it was the great equalizer. And it was this thing
0:20:42 that allowed you to actually break racial barriers and religious barriers and economic barriers in
0:20:48 order to foster competition and that the great competitors in sport could come from anywhere.
0:20:53 And everybody felt this ability to have a sense of belonging and purpose and meaning. And that
0:21:00 is what video games has become for so many people. And it’s hard to illustrate this for some people,
0:21:05 but I was at a panel not long ago with Alex Rodriguez, who actually owns one of our Overwatch
0:21:13 team franchises, and Roger Goodell, who is the NFL commissioner. And the moderator said, “Are eSports
0:21:20 sports?” And I said, “The same characteristics that Mandela described, what makes sport great
0:21:26 is what makes eSports so compelling and engaging.” And I said, “To Alex, stand up.”
0:21:34 And Alex stood up and I said, “Look at you. How many people in the world can play professional
0:21:38 baseball?” And he said, “Well, there are roughly 1,200 professional baseball players in the
0:21:45 Major League Baseball and about 3,000 capable of playing Major League Baseball.” And I said,
0:21:52 “Look at this guy. Like this is like the most fit athletic specimen of a human on the earth.
0:21:58 And there are only 3,000 of those people who can do what he does. Video games is the only
0:22:03 competitive medium that is going to give me that experience and that purpose and that sense of
0:22:07 belonging and that camaraderie that you get from sport.” And so, of course, it’s going to be
0:22:13 as popular as sport, if not more popular than sport. And that, I think, more than anything, is now
0:22:19 what we see as driving consumption and engagement and interest and passion. And we’re just scratching
0:22:24 the surface of opportunity. So I think people have obviously had a lot of great experiences
0:22:27 and a great faith in the video game industry for many years, based on the idea that everybody
0:22:32 can participate. This idea that people are going to voluntarily watch other people playing video
0:22:37 games is a new idea. And obviously it’s becoming a twitch and one of our companies and so forth.
0:22:42 Like this is going to be a very big phenomenon. It’s a key part of eSports is the ability to
0:22:45 fill an arena with people watching other people play video games. Like a few years ago,
0:22:50 that just sounded wildly implausible. What was the point? Like, when did you figure that out?
0:22:55 Well, I think probably when we launched. I didn’t own Blizzard at the time, but when Starcraft
0:23:00 launched, this was a game that in Korea, I think at the height of its popularity. Rob’s here,
0:23:05 so he’ll know the exact number. But I think, you know, this is more than a decade ago, but the
0:23:11 height of its popularity in Korea, Starcraft had 5 million registered players. Now this is a country
0:23:18 of 60 million people. The game is primarily a male game experience. And so you think about 20% of
0:23:24 the population actually of the male population played, or was a registered player of Starcraft.
0:23:30 And we saw arenas getting filled with spectators. There were three dedicated cable channels in
0:23:34 South Korea that just broadcast Starcraft competition. There were sponsors. There were
0:23:40 professional players making $100,000 or more. So this is an amazing phenomena that took place that
0:23:45 we looked at as purely marketing. You know, the people are enthusiastic. We saw the box. There was
0:23:51 nothing more to it than that. And we managed to do every single thing wrong in commercializing the
0:23:58 eSport of Starcraft. But it was the first time where I really thought, you know, there’s something
0:24:04 that could even be bigger than the games themselves that would relate to the spectator experience.
0:24:10 Now, even with games like Overwatch, which is probably our most successful eSport initiative,
0:24:16 it’s more like golf. So if you’re a spectator of Overwatch, it’s likely you’re a player of
0:24:24 Overwatch. Fortnite, I think, was the first game where people would spectate and it would actually
0:24:30 be a catalyst for them to play. And so I think what’s happened is it’s more of a social experience
0:24:35 in a lot of respects than it is just a game. But I think that what you’re now starting to see is
0:24:41 that games have so infiltrated the popular culture of the world that it’s exciting for people to watch
0:24:45 their heroes who compete against each other in the same way as sport.
0:24:50 Right. So many of the most successful games that people watch or that are now actually
0:24:53 formerly eSports, correct me if I’m wrong, they were not originally designed for this
0:24:56 period. They were designed to be games that people just played and they’ve been kind of repurposed
0:25:00 into this kind of broader public phenomenon. Maybe that’s to play untrue, but I guess my question
0:25:06 is kind of how will video games be designed going forward for eSports and for people watching it in
0:25:10 addition to playing that is different than how video games have been designed up to this point?
0:25:14 Yeah, that’s a great question. So a lot of the games today were not specifically designed for
0:25:21 spectating, which is why you end up with that phenomenon of the players are the spectators.
0:25:29 In order to have a more broad appeal spectator experience, the games need to be designed in a
0:25:33 way that you actually want to watch them, whether you do or don’t play them. And I would say that
0:25:40 the Overwatch team spent a lot of time early on trying to construct a game from the ground up
0:25:45 that would be a fun spectator experience. But I think what you will see is that people are now
0:25:51 paying more attention to in game design, the idea that the games may be spectated by people who
0:25:56 aren’t players. I don’t think anytime soon that’s going to be the primary consideration. We’ll still
0:26:05 be principally focused on gameplay, but things like camera angles and commentating and making sure
0:26:10 like when we organized the Overwatch League, one of the organizing principles was the reason why
0:26:17 sports are so successful is tribalism and that having a local affiliation was so crucial to the
0:26:22 success of sport, whether it’s a country affiliation or city affiliation. So we created a structure
0:26:29 that allowed for 28 independent cities to field teams. And I think that as you start to take those
0:26:34 considerations into play when you’re thinking about the design of the games or the leagues or the
0:26:39 competitive experience, that they will have more of the characteristics of traditional sport.
0:26:42 Right. Would you venture a guess as to when video games will be in the Olympics?
0:26:46 Which is a logical implication of what we’re discussing, right?
0:26:51 I don’t think so. I actually don’t think like the Olympics has never been about
0:26:57 a commercial enterprise. And so if you think of the analog, right, there’s not like if you had to
0:27:04 pick a game, you’re now endorsing someone’s commercial enterprise. You know, there’s not any,
0:27:11 there’s they don’t have that analog today. And so I don’t know that you could see the logical
0:27:16 jump to the Olympics. Okay. The video game industry seems to have a particularly cute version of a
0:27:21 dynamic that you see with, you know, let’s just say consumer properties that inspire an avid fandom.
0:27:24 So it’s the enthusiastic early adopters, right? And so you see this with movies,
0:27:28 you see this with TV shows, you see this with basically things that really occupy the popular
0:27:33 imagination. You know, for sure see it with video games. So you’ve got this kind of leading edge,
0:27:38 you might say early adopters slash super enthusiastic user base, and they start to develop
0:27:42 opinions and they start to develop opinions that maybe the people who make the games are quite
0:27:46 doing what they want. And then when things, you know, really go sideways, there can be, you know,
0:27:50 protestant boycoss and all kinds of like, you can end up with the inmates running the asylum or at
0:27:54 least looking like they’re certainly trying to, how do you throw the needle as somebody who make
0:27:58 an overseas letters? How do you how do you throw the needle for the early adopter base as opposed to
0:28:02 the mainstream? And how much is the early adopter base an asset? How much is the early adopter base
0:28:08 a challenge, a problem? So I think that the difference between film or television, you know,
0:28:14 a great film, you’re going to spend two hours of your life watching. You know, great TV shows,
0:28:20 it’s going to be 13 episodes or 22 episodes a season. You know, it’s going to be 13 or 22 hours.
0:28:28 Video game, our average duration of gameplay that includes games like Candy Crush is an hour per
0:28:36 person per day. Games like Call of Duty or World of Warcraft are hours a day. So the interest in
0:28:41 the engagement and the commitment that you’re making to that form of media is so different
0:28:50 than film and television. In my view, you have the right to have a strong opinion and voice your
0:28:56 opinion in exchange for making that hour plus commitment a day, which becomes more of a lifestyle.
0:29:02 And so instead, you know, I think some companies run and hide and don’t really engage their user
0:29:08 base. But I think we have users and players who will and audience members who will tell you
0:29:14 and give you really good insight into how you can modify and adapt your game. So you listen to
0:29:20 them and they’re not always right. But oftentimes they’re pretty and especially when you hear the
0:29:27 sort of the mass view, they’re pretty right. The beauty of our business though is that if you
0:29:31 can get out in front of it early, let people actually have an experience with the game,
0:29:36 get the feedback and you’re willing to take that feedback and enhance and improve and modify the
0:29:40 games. It’s a great roadmap for innovation. So I’d like to ask you about two games that I think
0:29:44 are arguably transformative from a conceptual standpoint for gaming. And you tell me what
0:29:48 you think or maybe describe what you think are the structural significance of each of these.
0:29:52 And so the first is Fortnite. And you alluded to one of the dramatic changes. So maybe describe
0:29:57 like what is the significance of Fortnite to the industry? So for starters, there’s this perception
0:30:02 that Fortnite is an overnight success. It’s not epic. The company that made it has been in the
0:30:07 video game. Tim has been in the game business almost as long as I have been. And they are excellent
0:30:15 at making games. And what they did was to really spend the time in a very focused, determined way
0:30:20 taking the Unreal Engine and turning it to something that was going to be a broad appeal,
0:30:25 very compelling social experience. And I think the aspiration was build the social network that’s
0:30:32 anchored in a game conceit. And they made it cross-platform in its playability. They made it
0:30:39 very accessible. They changed the way that they deliver content to season. So moved from the
0:30:45 feature film model to the serialized television model. And it’s really fun. And they managed to
0:30:52 do what you probably can’t do intentionally, but capture the popular cultural zeitgeist.
0:30:58 And the other thing is that it doesn’t require you to make a two hour a day investment. You know,
0:31:04 you can have a 20 minute experience that is really satisfying. But this is not accidental. You know,
0:31:08 these guys have been doing this for a very long time. And I think what it has started to do is
0:31:14 broaden the appeal of games to people who might never have played games before.
0:31:17 And then what is the significance of Pokemon Go? And by the way, for people who don’t actually
0:31:21 just saw this yesterday, Pokemon Go, third party report. But the report was revenue last month was
0:31:26 still in the order of $75 million. So it’s still something like a billion dollar a year revenue
0:31:30 business today. And this is where the game is now what, two years old or something like that.
0:31:31 Almost three years.
0:31:34 At least rumor has it that they have new stuff coming. But it’s been a giant hit.
0:31:38 Again, it’s like, where’s the innovation there? But it’s like, you know,
0:31:44 Nintendo is really great at innovations that are very physical in their nature. So the Wii,
0:31:50 that moment I was describing to you about the Macintosh, when I first saw the prototype of the
0:31:57 Wii, it was like that equivalent goose bump moment. I was in Kyoto and I went into his room
0:32:04 and there was a TV and it wasn’t like an LED. It was an actual tube TV. And there was a pond that
0:32:10 was on the screen, like a little cartoon pond with little bubbles popping up from every once in a
0:32:15 while. And the head of Nintendo at the time was a guy named Iwata-san. And he gave me the controller.
0:32:19 And I held the controller and I just started going like this. And all of a sudden you could feel
0:32:26 the tension of the controller and the motion control of the controller. And I started to like
0:32:31 fish around and I grabbed the fish and I pulled the fish out. And I thought that video games
0:32:38 will be completely transformed. Nothing had really taken the physical experience in video games to
0:32:45 that level. And I think that what Pokemon Go did is something very similar. But it created this
0:32:52 physical experience that I think it was the first time AR had been executed on a broad scale.
0:32:59 And so I haven’t seen anything in gaming that I would tell you has really captured the imagination
0:33:06 of people on a broad scale using AR besides Pokemon Go. But I would say when you look at
0:33:09 where some of the next big innovations in gaming, including what we’re working on,
0:33:15 AR is going to have more near-term impact than VR. Amazing. Okay, good. And then final closing
0:33:19 question. What’s the one Activision game that has come out this year that people really have to
0:33:24 play? And then I know it may pain you, but I’m going to ask you, what’s the best non-activision game
0:33:29 that has come out this year that people really have to play? So I would say Call of Duty Blackhouse
0:33:35 4, which we just released, the blackout mode of Call of Duty Blackhouse 4 is so incredibly fun to
0:33:40 play. And what is that? What is that mode? It’s like a PUBG mode. It’s like a battle royale mode.
0:33:47 And it’s super fun to play. And you can do it in like small 25 minute increments, but very accessible,
0:33:53 very fun to play. I haven’t played Red Dead Redemption yet, but I want to. And I would say of the things
0:33:59 that have come out this year, I think it looks like Westerns are very hard to do because they’re
0:34:04 very American in their field. But the game looks fantastic. And everybody they know has played
0:34:08 it. Is that a lot of fun playing it? Yeah, fantastic. And I think we have party favorites. We brought
0:34:11 Call of Duty for everybody. Call of Duty. So I believe everybody’s going to have a copy of Call of
0:34:18 Duty. Bobby, thank you so much. Mark, thank you very much.

0:00:06 Hi everyone, welcome to the A6NZ Podcast. I’m Sonal. Today is part of our ongoing series
0:00:12 on consumer tech trends. We’re talking all about the category of consumer packaged goods
0:00:18 or CPG and where this fits with what’s going on in online and offline commerce trends overall,
0:00:23 including when it comes to the grocery business. We also talk about the trends of DTC or direct
0:00:29 to consumer and the concept of micro or emerging brands as well. Joining us to have this conversation,
0:00:34 we have A6NZ General Partner Jeff Jordan, who’s written a lot about competing with Amazon and
0:00:38 the future of e-commerce and marketplaces and has been on the front lines of this area both as
0:00:43 an operator and investor, witnessing firsthand from many angles a lot of changes in the industry.
0:00:48 And then we also have Ryan Kahlbeck of Circle Lab, an investment platform powered by data and
0:00:53 technology. He also talks a lot about innovation in the CPG space and more, which is why we
0:00:58 invited him to join this discussion. Please note that the content here is for informational purposes
0:01:03 only should not be taken as legal business tax or investment advice or be used to evaluate
0:01:09 any investment or security. It’s not directed at any investors or potential investors in any fund.
0:01:15 For more details, please also see A6NZ.com/disclosures. The broader question that we covered throughout
0:01:21 this episode is how does technology and especially the internet change and in some ways not change
0:01:26 the way we do things. Why do some of the traditional businesses in these industries, though full of
0:01:32 some of the smartest people, have trouble innovating and can tech really help? In fact, that’s where
0:01:37 we begin the conversation with the assertion that it’s been hard to bring tech to CPG, even though
0:01:45 there seem to be a lot of products trying out there. As a user, I have a hard time perceiving this
0:01:50 because I am constantly bombarded with consumer products that are techie, whether it’s like
0:01:54 a thousand beauty product lines, there’s a thousand even generics lines that I’m getting hit with.
0:01:58 So I have a hard time grokking that this is a reality. Tell me what’s happening there.
0:02:05 It’s a great point. So I think, first, one of the reasons that you see that is the D2C movement,
0:02:11 which is direct to consumers. A handful of tech VC firms, they’re putting a fair amount of money
0:02:17 into consumer product companies that sell just direct to consumer or predominantly direct to
0:02:23 consumer. That money is usually not used for innovation, for product innovation. It’s usually
0:02:28 used for marketing. I agree with Ryan that they’re largely marketing companies. It’s actually
0:02:32 derivative of Amazon, unfortunately. If you want to play any e-commerce in a post-Amazon
0:02:37 era, you cannot sell what Amazon sells successfully. I mean, early on, we tried a number of concepts
0:02:43 that some are around anymore that said, okay, I can kind of sell the same thing as Amazon with
0:02:47 a different distribution twist with something like this, but you can’t. So then the next wave
0:02:52 of companies was, okay, if I can’t sell, I can’t compete with Amazon directly skew by skew,
0:02:58 let me get proprietary skews. And so they’re formulating different products with different
0:03:05 branding, and they typically aren’t on Amazon or on other retail. But that is where a lot of the
0:03:11 direct to consumer stuff is happening. And frankly, some success, not a ton of success
0:03:14 in doing it. Why is that, by the way? At some point, you’re still competing with
0:03:21 legacy players. And what’s worse is they’re also competing with each other. So there are a half
0:03:29 dozen, a dozen brands all trying to find you. Early on, they find you at $20 acquisition cost
0:03:34 when 10 of them go at each other. They’re competing for you, 10 companies trying to grow,
0:03:39 competing for you. And then that also caps their ability to price. So one of the big
0:03:44 secrets in e-commerce is virtually no companies get big and make money in the United States on
0:03:51 e-commerce. I can name five e-commerce companies in the US after two decades of investing and
0:03:56 God knows how many tens of billions of dollars who are a billion dollars in sales and profitable.
0:04:01 Yeah, I mean, you’re making the point, particularly around the rise in CAC. We see a lot of
0:04:07 DTC companies raise money with a CAC, let’s call it $10, $30. And then over the course of
0:04:13 subsequent two years, it raises literally sometimes by an order of magnitude. And that is just killing
0:04:17 their profitability. And by that, just to quickly summarize, you mean CAC is on customer acquisition
0:04:20 costs and they’re using all that marketing dollars to acquire their customers.
0:04:26 Exactly. I think when DTC directed consumer businesses, I think a lot of folks look at it
0:04:31 from the outside and say, okay, we’re stripping out the middleman, the retailer, the offline
0:04:36 retailer. And because of that, we can put that margin that the retailer gets into our business.
0:04:40 And this is going to be a wonderful business. The problem is that you need to attract people to
0:04:47 your site, right? To get people to come to your startup, to come to your URL and buy something
0:04:51 is a really difficult thing. And so that’s what’s driving these customer acquisition costs up.
0:04:55 Now, there are some DTC companies, Native Deodorant was bought. They only raised
0:04:59 two or three million bucks in total, sold for $100 million to Unilever.
0:05:03 Why did Unilever want to buy that company? Why wouldn’t they just make their own in-house brand,
0:05:09 another brand? The CPG companies, large ones, have lost the ability to innovate. They innovate
0:05:13 through acquisition now and they buy early. So Unilever bought Dollar Shave Club. Unilever
0:05:18 doesn’t have a shaving business. I don’t know Unilever Deodorant offerings, but I’m guessing
0:05:24 they saw something in there. So it and Biotech are kind of the same. The big pharma companies are
0:05:28 buying their innovation. The big CPG companies are buying their innovation. Historically,
0:05:34 large CPG has been able to rely on their brands. Brands that have existed literally in some cases
0:05:40 for 50 or 100 years. Meg, Scott Cook, and Brian, Sweetie, and Steve Ballmer, we’re all at P&G
0:05:45 together. Right after I graduated from college, Meg was the brand manager, I believe, on Crest.
0:05:52 Wow. That’s awesome. But can you think of that cadre of talent in all analysts in Cincinnati?
0:05:56 CPG seems like, because I have a lot of friends that go through their career starting at Clorox.
0:05:59 And it seems like sort of a training, there’s always like a mafia and a training ground that
0:06:02 people get some fundamental skills in these companies. So what are they actually getting
0:06:06 out of these companies? There’s a quote in one of the Apple movies, if you can sell sugar water,
0:06:12 you can certainly sell. That was Steve John’s quote, like he tried recruiting John Scully.
0:06:16 I think there’s probably something to that, right? To be able to sell literally the same product
0:06:19 for 30 years in a row. Think about trying to do that in the technology space. That’d be
0:06:26 impossible. There’s a fun book by a Harvard dean called Different. And it basically says,
0:06:31 almost all consumer goods are these small incremental improvements that everyone copies
0:06:38 because that’s the improvement that wider mouths on the Crest toothpaste because that you go through
0:06:44 it faster. Oh, so Colgate will do wider mouths and Spearman and whitening and versus brands that
0:06:50 kind of just throw it out and just do it differently. She uses Harley. She uses Apple. She uses Red Bull.
0:06:56 And they just came to market with completely, we are going to be different. And it’s so interesting.
0:07:02 I love the book because 99% is exactly the same. And then the 1% is different.
0:07:05 That gets back to the point of like why these large brands are losing market share.
0:07:11 The large brands have not had to innovate. They’ve been able to rely on their brands on,
0:07:16 whether it’s Coke or Pepsi or whatever it is, the same product, just their brand name for decades.
0:07:22 Now these smaller upstart brands, these emerging brands are able to develop the innovation in
0:07:27 house and deliver that to the consumer. The consumer, on the other hand, for the first time
0:07:32 in history is saying, I demand products that meet my unique needs. Maybe the three of us used to
0:07:37 eat the same breakfast cereal. Now the three of us are eating different things. We’re demanding
0:07:43 products that meet our unique needs. That customer consumer demand is pulling forward the innovation
0:07:46 and giving it a market. That’s why though, I just want to push a little bit more on this DTC topic
0:07:51 though, because when I think of the theory of the internet, which, you know, it disintermediates this
0:07:56 intermediary, you can go direct to people, you create movements on the internet. I mean,
0:08:00 talk about meme culture. Like if a meme can monetize, why the hell can’t a product?
0:08:03 So I’m still having a hard time buying why DTC is so hard.
0:08:07 The internet enables a long tail. So all of a sudden everything’s discoverable,
0:08:11 where it used to be, there were three television networks and 10 magazines that mattered.
0:08:18 The world is different. It used to have, you know, very narrow portals to work, which was great for
0:08:25 big companies that have stable product lines. So you couldn’t do TV on a $5 million revenue line.
0:08:29 You can do internet on a $5 million revenue line. Right. But that’s exactly my point,
0:08:33 is we have a world where we do have a long tail discoverability. You have the ability to access,
0:08:38 find your niche audience. You also have a world where you can still make hits on that long tail.
0:08:42 So you just describe why there’s a proliferation of CPG companies.
0:08:45 But why does DTC not work? That’s a part I don’t get.
0:08:49 When we say it doesn’t work, I think it is an incredible channel for iterating on a product.
0:08:54 I think it drives innovation. I can find my consumer base and I can test a new product.
0:08:57 Next month, I can test a different product. That’s very different than the offline world,
0:09:00 where if I’m selling into Safeway or Whole Foods or Costco down the street,
0:09:05 it’s hard for me to switch the product out if it’s not working, to tweak a package, whatever it is.
0:09:09 People in the technology space take that kind of A/B testing for granted. In CPG, it’s much,
0:09:12 much harder because you literally… Because it’s physical product and offline space.
0:09:14 Yeah, you have atoms, you need to move, not just bits.
0:09:18 And so if you run a chain of stores, actually a couple of my… All the companies that are
0:09:23 getting challenged growing, continuing to grow DTC sales are turning to offline retail in different
0:09:27 forms. Every company I’ve worked with is trying to do that because incremental sales…
0:09:29 That’s actually rather counterintuitive.
0:09:33 It is. But the way store management work, I used to be CFO of the Disney store, is
0:09:40 you try products. Typically, you do two or four sets a year and you try products in a
0:09:44 significant subset of your stores because if it’s a bomb, you don’t want to buy
0:09:48 deep in it. So you’ll put it in five stores a test for the first six months.
0:09:52 And then they say, “That worked well. Let’s go to 50 stores for the next six months.”
0:09:56 And then by the time you get chain-wide, it literally is a couple of years later.
0:10:00 And so it’s really hard to grow your business.
0:10:03 And that’s for Disney stores where you have your own product in the store.
0:10:09 So if you think of the CPG products, they’re selling through a third party, the retailer.
0:10:14 Now, they have to make a much bigger bet. They can’t usually just start in five safe ways.
0:10:18 So they have to start in 200. Now when you start in 200, flipping that and changing that product,
0:10:21 if it didn’t work, it can kill the company.
0:10:26 When I was at Xerox PARC, one of the companies we partnered with was a large CPG company.
0:10:30 And their number one challenge was trying to figure out what happens after the consumer
0:10:33 buys the product. They had zero insights.
0:10:35 Oh, actually, it’s even worse. They have zero insight.
0:10:40 They know what they sell into the retail chains. They don’t know what’s sold where to who.
0:10:41 Right. Seriously?
0:10:41 No. It’s incredible.
0:10:42 It’s absolutely incredible.
0:10:44 That’s great. That’s crazy. How is that not possible?
0:10:47 So people who are listening to this will respond, “Well, what about credit card data?”
0:10:50 The problem with credit card data is it can tell you what the retailer is selling,
0:10:51 not what was bought at the retailer.
0:10:54 So I can see what Nordstrom.com sales for last month.
0:10:56 I can’t see what pairs of jeans were bought there.
0:10:58 Why not? Doesn’t it say?
0:10:58 They don’t sell.
0:11:00 Isn’t that the whole point of a skew?
0:11:03 No. Well, yeah, but these credit card companies don’t sell it.
0:11:08 And more importantly, back to Jeff’s point, they don’t tie those products to individual people.
0:11:16 Some retailers have loyalty card data, but there’s privacy issues with them selling that.
0:11:22 $84.51, a division of Kroger sells part of it, but not in the way that we’re talking about here.
0:11:26 They can say, “This person who I can, by the way, I can match Sonals,
0:11:29 what her purchasing power to your Instagram account, that is not done anywhere.”
0:11:33 Typically, CPG marketing is you buy a circular in the Sunday paper.
0:11:35 You buy an end cap, and you have no idea–
0:11:37 An end cap being the little display at the end of the grocery store.
0:11:38 At the end of the aisle in the grocery store.
0:11:42 And billions of dollars goes into those, and who knows if it works.
0:11:48 And they certainly don’t have any data beyond, “Oh, our total sales went up a little bit there.”
0:11:53 One of the key things that was compelling about Instacart to us is that they have a revenue
0:11:56 stream from the consumer, a revenue stream from the grocery retail partner,
0:12:00 and a revenue stream from CPG companies who are interested in accessing the consumer.
0:12:04 Why the CPG companies were so interested in that is,
0:12:06 it’s the first performance marketing they’ve ever seen.
0:12:09 On something like Instacart, they know everything I’ve ever bought.
0:12:15 And if they know I love Heineken beer and the product manager of Stella wants to try to convert me,
0:12:19 they can give me a Stella coupon, a Stella samples, a Stella, everything else.
0:12:21 And then see, did I change my behavior over time?
0:12:23 That’s the holy grail of marketing.
0:12:24 The holy grail for marketing.
0:12:26 I mean, performance marketing, like that type of data,
0:12:29 like closing that feedback loop is a big F-deal.
0:12:34 They also over time have the ability to kind of literally move the product on the page, right?
0:12:36 Which if you think about offline retail, you can’t do.
0:12:38 What do you mean by on the page?
0:12:43 Meaning when you’re looking at Instacart, what product you’re looking for can be moved physically.
0:12:47 Oh, like kind of personalized rearrange to your needs.
0:12:47 Exactly.
0:12:48 Which you cannot do in a physical way.
0:12:50 And it’s completely obvious in tech.
0:12:52 But for people that are in CPG, it’s a game changer.
0:12:57 So I was in a Safeway half a mile from here on Sand Hill six months ago,
0:12:59 and was buying some steak.
0:13:01 Steak was in the refrigerator.
0:13:04 Next to the steak in the refrigerator were wood chips.
0:13:06 The wood chips have no need to be refrigerated.
0:13:08 They’re literally just wood chips to put in a barbecue.
0:13:11 They’re in the refrigerator because they’re on sale.
0:13:13 And they want to put them next to the steak.
0:13:14 That’s it.
0:13:16 And so you’re refrigerating literally wood.
0:13:17 That’s crazy.
0:13:19 Because back to Jeff’s point, if I bought an end cap,
0:13:21 the end cap would have been 30 feet away.
0:13:23 Then you would have lost the consumers.
0:13:24 And they need to have that proximity.
0:13:26 That’s a crazy awesome example.
0:13:27 I love that.
0:13:30 Let’s talk about grocery for a little bit as an interesting category.
0:13:35 So first of all, is grocery going to go the way of malls and other retail?
0:13:37 So I’ve got a not very popular opinion on this.
0:13:38 Let’s hear it.
0:13:39 The more popular, the better.
0:13:43 I think grocery stores are here for the very, very long term.
0:13:46 Very long term, meaning next 20 years at least.
0:13:50 So taking a step back, D2C and e-commerce has been around for 20 plus years.
0:13:52 This is not a recent phenomenon.
0:13:55 Food is still, Jeff, you could tell me, I think 5% of sales.
0:13:56 Of digital?
0:13:56 Yeah.
0:13:57 Way under.
0:13:58 Yeah.
0:14:02 I don’t foresee a world in which everyone is just going to go online, A.
0:14:07 And B, when that happens or when it gets to be a much higher proportion of where you buy
0:14:13 your groceries, I still think that that grocery is going to be delivered locally.
0:14:18 But the core point is grocery today, 2%, 3% net margin business.
0:14:21 That’s hard to strip out a lot of costs from that.
0:14:21 There’s not a lot of room left to go.
0:14:23 You can’t do any cost cutting at all.
0:14:23 That’s right.
0:14:29 So you think about where technology has been particularly successful at killing other industries.
0:14:32 They tend to be industries where there’s a fair amount of profit, right?
0:14:35 It’s also like Bommel’s cost disease, too, if you think about it, eventually penetrating
0:14:39 health care and education, like things that are way more expensive than they need to be.
0:14:41 And technology is a vector to just cut right through that.
0:14:42 Absolutely.
0:14:46 But the safety down the street has a 1.5% net margin.
0:14:48 So what are we going to do to rip the cost of that?
0:14:52 Because that thing is already delivering a pretty good product to the people that live within a
0:14:53 mile and a half of here.
0:14:57 I mean, there was a set of companies that was trying to help physical retail compete
0:14:58 with the digital commerce.
0:14:59 And it was so interesting.
0:15:04 We didn’t invest in many of them at all because our internal reference was we’re shorting the
0:15:05 future.
0:15:08 You know, just so it’s not a long-term winning proposition and their margins are going to
0:15:10 get squeezed, so they’re going to squeed their vendors.
0:15:13 Then we made the Instacart investment in the belief that the grocery stores are there.
0:15:15 It’s distributed.
0:15:17 They have distributed little warehouses.
0:15:18 They provide service.
0:15:19 And it was interesting.
0:15:22 Fred Smith, the founder of FedEx, when the internet first came up said,
0:15:26 “I think a whole lot of goods are going to be delivered by FedEx.
0:15:27 Groceries ain’t going to be one of them.”
0:15:32 Because the concept that you’re going to load a truck in the morning and having
0:15:36 bouncing down the streets and this and that and deliver at 8.
0:15:40 It’s something at 8 p.m. that was put in the truck at 4.30 a.m.
0:15:42 And he just said, “I don’t think it works.”
0:15:47 Building up the shopper network is so hard to replicate for someone else going forward.
0:15:50 It’s not as simple as, “Let’s just hire FedEx to go do this thing.”
0:15:54 Oh, and the operational intensities observe.
0:15:58 We think in grocery, we think effectively there are a couple different dimensions
0:15:59 that grocery chains need to compete on.
0:16:01 We think it’s assortment.
0:16:03 We think it is convenience.
0:16:05 We think it is pricing and experience.
0:16:07 Pricing is a really, really hard place to compete on.
0:16:10 You’re competing with Amazon to a point that Jeff made earlier.
0:16:12 You’re competing with Walmart.
0:16:14 I think that’s just a really, really hard place to win.
0:16:15 Yeah, you can’t win on price.
0:16:18 Especially when you already have this 1% to 2% margin.
0:16:20 Exactly. The margins are already so low.
0:16:22 So let’s put pricing aside for a second.
0:16:28 I also think that grocery is, in some ways, an experience for some people.
0:16:33 An experience for the consumer to go in, perhaps with the family,
0:16:36 perhaps just buy a couple of things here and there.
0:16:38 They want the immediacy of that.
0:16:41 We don’t see that going away anytime soon.
0:16:43 I think it’s going to be hard to win, again,
0:16:47 because how do you invest into an experience if you have such low margins?
0:16:49 I will say one quick sidebar on experience,
0:16:50 such as if Connie were in this room,
0:16:52 she talks a lot about what happens in China.
0:16:55 And she talks about this incredibly fascinating phenomenon
0:16:59 where grocery stores in China have become destinations themselves
0:17:01 because there’s literally restaurants inside.
0:17:04 They’re doing all kinds of neat food chef things, etc.
0:17:06 Cooking on site, doing all these interesting things.
0:17:07 I mean, what should you take on that?
0:17:10 I was with the CEO of a large grocery chain about a month ago
0:17:12 who made that same point.
0:17:14 They are starting to experiment by putting restaurants,
0:17:18 effectively restaurants, inside their grocery chain.
0:17:22 The local Asian supermarkets almost all have restaurants forever.
0:17:23 Indian ones, too.
0:17:25 They have like samosas for $2.
0:17:25 Yeah.
0:17:30 I would say I think it is nice to have not a requirement to be successful.
0:17:33 I’m not convinced that that’s going to be the core differentiator going forward.
0:17:35 So what do you think is going to be the core differentiator?
0:17:35 You had one more dimension.
0:17:38 Well, convenience and assortment, those are the last two.
0:17:40 So convenience, to me, is delivery.
0:17:41 Delivery or in-store pickup.
0:17:45 I think that that will end up being table stakes, not a nice to have.
0:17:47 Meaning you need to have convenience.
0:17:50 Well, my question on this is why bother even having a grocery store
0:17:51 if you only need to deliver?
0:17:53 Why not keep warehouses then?
0:17:54 If experience isn’t going to win the thing.
0:17:56 Like why not just have a bunch of warehouses
0:17:57 that deliver food if delivery is the thing?
0:17:58 Yeah.
0:18:01 So to be frank, that could be the 50-year vision.
0:18:02 That really could.
0:18:04 In the UK, which for whatever cultural reason
0:18:06 has been doing grocery delivery for a long time
0:18:09 and it’s a much deeper state of penetration,
0:18:11 they’re starting to have what’s called dark stores,
0:18:14 where you basically, they look a lot like supermarkets.
0:18:16 They’re locally distributed.
0:18:17 They have inventory assortment,
0:18:19 but they’re not in high traffic parts of town,
0:18:22 which means the rents are much, much lower.
0:18:26 And so, as a result, you don’t have to make it pretty.
0:18:28 You don’t have to have the lights, lots of lights.
0:18:29 You don’t have to have high labor.
0:18:30 All you have to do is pick in them.
0:18:33 And so that is how grocery has been developing
0:18:36 in one of the most advanced digital grocery markets.
0:18:39 So it’ll be interesting if whether that happens in the US,
0:18:43 every argument that I can make on why grocery stores won’t
0:18:46 just be warehouses where it gets delivered from,
0:18:49 every argument that I can make is a short-term argument.
0:18:51 I can’t make a 50-year argument for that.
0:18:53 So I think that that could happen.
0:18:55 Over the next 10 years, I don’t think customer adoption
0:18:56 will be big enough to get rid of these stores.
0:18:59 These stores, these aren’t little mom and pops.
0:19:02 They’re Fortune 500 companies that would not go down
0:19:03 without a really big fight.
0:19:04 No, and I think there are destinations.
0:19:06 I mean, I see like families on the weekends all the time.
0:19:08 It’s like, it’s an outing to go to the grocery store,
0:19:10 which if they had like a daycare,
0:19:12 I think that would be a huge win, frankly.
0:19:14 That alone would be a big differentiator.
0:19:15 I would love that.
0:19:16 I know, I think everybody would take that.
0:19:18 I exclusively shop through Instagram.
0:19:20 I do our family’s grocery because I can’t bring the kids
0:19:23 to the grocery store because it’s too much of a disaster.
0:19:26 So I believe that convenience will be table stakes.
0:19:29 So then it comes down to what is the way grocery wins?
0:19:30 What are the way grocery store competes?
0:19:32 To me, it is just about assortment.
0:19:33 That’s your fourth dimension.
0:19:33 This is the fourth dimension.
0:19:36 After price, experience, convenience, and now assortment.
0:19:37 That’s right.
0:19:40 Assortment is what are the products that are on the shelf?
0:19:43 So if you think the last 80 years,
0:19:46 it has been, well, what’s Pepsi gonna give us this month?
0:19:49 Coke, General Mills, Unilever, Procter & Gamble, et cetera.
0:19:50 Today, as we’ve talked about,
0:19:53 consumers want brands and products to meet their unique needs.
0:19:57 The problem with that is that the buyer at the retailer,
0:20:00 meaning the person that selects, let’s say, the chocolate company,
0:20:03 the buyer at the retailer is still the same buyer
0:20:06 that lived there 20 years ago in some cases.
0:20:08 They don’t use a lot of data.
0:20:10 They’re literally trying chocolate bars
0:20:11 to decide what goes on the shelf.
0:20:13 That’s a really hard place to be in a market
0:20:16 that is trillions of dollars
0:20:19 to decide which of these emerging brands they want to work with.
0:20:23 So the default is, gosh, maybe I just rely on the big brands
0:20:24 to tell me who to work with.
0:20:25 That’s a real example.
0:20:27 Sometimes the big brands literally say,
0:20:29 work with these emerging brands.
0:20:30 Maybe I work with the distributor,
0:20:33 who, by the way, has paid more than by the larger brands.
0:20:35 But there isn’t really a good solution
0:20:38 for a lot of these Fortune 500 retailers
0:20:41 on how to optimize their assortment.
0:20:45 The data that exists for these grocery chains is pretty poor.
0:20:48 You’ve got a couple retail-level sales providers
0:20:51 that on average track 20,000, 30,000 products,
0:20:53 or companies, rather, each.
0:20:56 There’s about a million and a half out there.
0:20:58 A million and a half, and they cover less than 5% of them.
0:21:00 And even the grocery and food shows, aren’t they?
0:21:03 Sort of like, I’m thinking of the equivalent in fashion
0:21:03 and boutiques.
0:21:05 You essentially have a fashion show
0:21:07 to curate all the brands, the emerging brands,
0:21:09 the existing brands, new product lines.
0:21:11 Why isn’t there an anthropology of grocery stores?
0:21:13 Because when I think of anthropology as a retailer,
0:21:16 they have assortment because they go around the world
0:21:17 to find a variety of designs.
0:21:20 So it works very well for a certain demographic of women,
0:21:22 like in their 20s, 30s, et cetera.
0:21:25 Then you have to say, which of these will my consumers
0:21:27 respond to at the rate price?
0:21:29 So you get exposed to the million,
0:21:33 but then I can only carry X,000 in my store.
0:21:34 How do I figure that out?
0:21:36 Online might actually help,
0:21:38 because you can put infinite selection online
0:21:39 and see what’s selling.
0:21:42 So if you’re a grocer, I’d be looking at my online sales
0:21:44 just to inform my offline sales.
0:21:46 Like, oh, wow, that’s a breakout hit online.
0:21:48 Why wouldn’t it be a breakout hit offline?
0:21:51 Interestingly, then, you might come back to a customer
0:21:53 acquisition challenge for the CPG companies,
0:21:56 which is if we have a ton of products online,
0:21:57 how do we then stand out?
0:21:59 That comes back into advertising.
0:22:00 This kind of vicious circle.
0:22:01 Or virtuous.
0:22:02 Or virtuous.
0:22:03 It’s a skewed sample, too.
0:22:05 Like what happens online and offline sometimes.
0:22:05 It’s definitely skewed.
0:22:08 But if you’re in a complete information vacuum,
0:22:09 which essentially these guys are, at least you have something.
0:22:11 Even, by the way, loyalty cards are a really skewed sample.
0:22:14 That’s a very self-selected, self-interested group.
0:22:16 You’re not really getting the huge untapped space
0:22:16 of what people want.
0:22:19 You’re not, but you’re actually finally getting data
0:22:21 on a per person basis to kind of understand what’s there.
0:22:22 Exactly.
0:22:26 It’s hard to understate how blind most buyers
0:22:29 at most physical retailers are right now.
0:22:31 Yeah, they’re just kind of, it’s instinct.
0:22:33 It’s, you optimize the current assortment.
0:22:36 But then how do you layer in new things is,
0:22:38 and there’s a plethora of new things.
0:22:39 Or just everything that you don’t know
0:22:41 people have the taste for.
0:22:41 Exactly.
0:22:42 So you think about that you’ve got a buyer
0:22:44 for the chocolate category, right?
0:22:47 And that could be an older, candidly white male
0:22:49 who’s making a decision on the entire category.
0:22:50 For a very diverse audience.
0:22:53 By the way, not diverse racially by age,
0:22:57 a certain gender, demographic, location.
0:22:59 So should we be selling the same thing in LA
0:23:00 that we do in Vermont?
0:23:02 So how are people going to get this data?
0:23:03 Because right now Jeff is describing
0:23:04 that they’re desperate for data.
0:23:06 So they have to rely on these,
0:23:07 they have to get data somewhere.
0:23:09 But that’s still adverse selection type data.
0:23:11 It’s not like the true opportunity space of data.
0:23:12 So where does the data come?
0:23:15 Broadly speaking, in consumer,
0:23:16 there’s a really beautiful thing
0:23:17 that a lot of people that don’t live and breathe
0:23:19 this space, they don’t recognize,
0:23:21 which is there is a tremendous amount of data
0:23:22 that’s out there in the world.
0:23:26 Meaning I can already see where a product is sold,
0:23:28 how many SKUs a company has,
0:23:30 meaning how many products that company sells,
0:23:32 what the price points of the products are,
0:23:33 what the end users think of the product.
0:23:34 And if I’m tracking it,
0:23:37 I can see how all those things change every single month
0:23:39 and how they compare to every other company in the category.
0:23:43 Those factors have been shown to be predictive of success.
0:23:46 People can aggregate those.
0:23:47 Now the challenge,
0:23:48 and this is where it gets really tricky.
0:23:51 The challenge is that you’re consolidating information
0:23:53 across literally hundreds of unstructured data sources.
0:23:55 It’s extremely intensive,
0:23:57 extremely intense and very difficult.
0:23:59 Sounds like a deal for an AI solution.
0:23:59 Yeah, it is.
0:24:00 It is.
0:24:01 But the data is out there is my point.
0:24:05 We think that that data is going to start getting consolidated
0:24:08 by data providers, technology companies,
0:24:11 that then sell it to the CBG companies,
0:24:13 or in this case, the grocery chains.
0:24:15 So the data opportunity, do you believe that?
0:24:15 Yeah, no.
0:24:17 We actually have seen a ton of companies trying to do it.
0:24:23 The interesting part is how hard they have to work to get the data.
0:24:24 There have been a whole bunch of them
0:24:28 that are trying to incent consumers to take pictures of their receipts
0:24:29 and submit the pictures.
0:24:32 And they do Optical Character Direct Edition on the picture
0:24:34 to try to reverse engineer.
0:24:36 What did Jeff buy?
0:24:37 We’ve seen multiple companies trying to do that.
0:24:39 There are other approaches too.
0:24:42 There was an on-demand,
0:24:45 they’d send armies of people with smartphones
0:24:46 to take pictures of shelves
0:24:48 so that they know the competitive pricing.
0:24:53 Oh, look, Crest is $2.29 and Colgate is $3.15.
0:24:54 What happened to sales?
0:24:57 But they don’t know what Colgate and Crest is,
0:24:58 and so they don’t know.
0:25:03 All they know is my philosophy in the southern region slowed down.
0:25:04 Or it’s sped up and you don’t know why,
0:25:05 which is equally problematic actually.
0:25:07 In the case of the receipts companies,
0:25:08 we’ve seen a lot of those too.
0:25:10 The challenge has been many of them tap out
0:25:13 at five, 10 million consumers, at least here in the US.
0:25:16 The problem with that level is that then,
0:25:18 that’s such a small portion of the overall population
0:25:21 that you’re not capturing the long tail of companies.
0:25:25 The chances that the $5 million popcorn company
0:25:29 is bought in a population that small is relatively low.
0:25:30 They’re going to be buying
0:25:31 the Procter & Gamble’s General Mills of the world,
0:25:33 but then you’re just getting more data
0:25:34 on the larger companies.
0:25:35 They need it on the smaller ones.
0:25:37 Exactly, they need data on the long tail.
0:25:40 Okay, so in this world, this long tail world,
0:25:42 this world of offline and online distribution,
0:25:44 and those are the two distribution broadly
0:25:45 that we’re talking about.
0:25:47 We are seeing a lot of microbrands,
0:25:48 and we didn’t talk about that
0:25:50 when we were talking about direct-to-consumer.
0:25:51 I mean, there’s a whole lot of media pieces
0:25:53 dissecting this phenomenon.
0:25:54 First of all, what is a microbrand
0:25:57 and why does it matter, or is it just a hypey thing?
0:25:59 Yeah, so let me just clarify.
0:26:02 By microbrand, you mean an emerging consumer product company
0:26:05 typically called less than $10 or $15 million in revenue.
0:26:08 So yes, but I think you’re the one who argues
0:26:10 that it denotes size, not channel,
0:26:12 because the way I’ve read it in the media,
0:26:13 it does talk about it as well.
0:26:15 Yeah, so the recent Economist article
0:26:16 kind of confused the two.
0:26:17 They started talking about microbrands,
0:26:19 and then they said, basically implied
0:26:21 that microbrands means D to C.
0:26:24 Micro to me, unless I’m missing something,
0:26:26 denotes size, not channel.
0:26:28 So I don’t really love that term to be candid with you.
0:26:29 Why didn’t you like it?
0:26:30 Well, for the same reason,
0:26:32 I would imagine that the entrepreneurs around here
0:26:35 wouldn’t like it if we called them micro technology companies.
0:26:36 They’re starting companies
0:26:37 because they want to build something big.
0:26:41 And I think that sometimes that comes across
0:26:43 as a little bit high and mighty, perhaps,
0:26:46 but especially if it’s just an ice cream company.
0:26:47 But look, I’ll be frank with you.
0:26:51 The ice cream company that’s able to strip out
0:26:53 fat and calories from the ice cream
0:26:54 and still deliver great products,
0:26:56 to me, is having a bigger impact on the world.
0:26:57 This is like Halo, right?
0:26:59 Halo Top, yeah, it’s exactly right.
0:27:03 So when we talk about them as emerging brands,
0:27:07 so emerging brands can be sold certainly offline or online.
0:27:09 But what’s happening,
0:27:12 these brands are growing very, very quickly right now.
0:27:14 In every single category in consumer,
0:27:17 large brands are losing share to emerging brands.
0:27:19 And the large brands are just terrified by this.
0:27:20 So why is it happening?
0:27:22 We think three primary reasons.
0:27:25 First is what we talked about
0:27:27 before the personalization of the consumer.
0:27:29 Consumers are demanding products that meet their unique needs.
0:27:31 Two other reasons that are pretty relevant
0:27:32 to this conversation.
0:27:36 One is decline in distribution costs.
0:27:39 Really, instead of just saying it’s a decline,
0:27:41 it’s really a shift from fixed costs
0:27:43 to variable costs.
0:27:44 And that’s because of the internet,
0:27:46 like the entry point to be able to buy,
0:27:47 get up a business up and running.
0:27:50 The internet is part of the driver,
0:27:52 but I don’t want to imply that that is the key driver.
0:27:55 In my view, it’s more that the offline retailers
0:27:58 are hungry to work with the small brands.
0:27:59 They’re struggling to figure out how to do it.
0:28:01 And so what we’re seeing is many of the offline retailers
0:28:04 are eliminating or lowering slotting fees.
0:28:07 Slotting fees are the cost to get your product on the shelf.
0:28:08 So if I want to launch a new chocolate bar
0:28:09 to get on the shelf of Safeway,
0:28:12 it is literally $50 to $100,000 just to get it on the shelf.
0:28:13 Think about that for a second.
0:28:15 If it’s the Apple App Store to launch an app,
0:28:18 it’s $100,000, it’d be a huge buried entry.
0:28:21 That fixed cost is declining rapidly
0:28:22 over the last five or 10 years.
0:28:23 Because these big brands are eager for–
0:28:25 The grocery stores are eager for these smaller brands.
0:28:27 More assortment, actually.
0:28:27 That’s right.
0:28:29 The third driver for why these emerging brands
0:28:32 have been so successful over the last five or 10 years
0:28:33 is marketing costs.
0:28:36 And that does get back to your point about the internet.
0:28:39 So marketing costs have also flipped from fixed to variable.
0:28:41 Fixed, meaning it used to be, let’s buy an ad in us weekly.
0:28:42 It cost me $100,000.
0:28:43 Or Jeff’s circular example.
0:28:44 Circular example is a great one.
0:28:47 Today, it is, let’s say, it’s the dollar-shaped flood YouTube ad.
0:28:50 And that’s an extreme example.
0:28:52 But you’re able to at least get your product out there
0:28:54 on a variable cost basis.
0:28:57 Because the flick fixed to variable transition,
0:29:00 these smaller brands, these emerging brands,
0:29:02 are able to grow much more effectively
0:29:03 than they could have 10 years ago.
0:29:05 So it literally is a variable cost.
0:29:07 When I was managing eBay, we were doing TV
0:29:09 and it was $1 million to produce an ad.
0:29:13 And then $10 million to distribute it with a frequency
0:29:15 that the brand people thought was efficient.
0:29:20 So $11 million was the ante to go on television for eBay at the time.
0:29:24 And so now you can go buy a $10,000 of Facebook ads
0:29:26 and try to reach your core audience.
0:29:28 So when you think about where this goes
0:29:30 in terms of innovation in the consumer space,
0:29:33 we think that over the next 10, 15, 20 years,
0:29:36 the brands that win, there will be more of them,
0:29:39 but they will get to a smaller level.
0:29:41 So what I mean is in the last 10 or 15 years,
0:29:44 the brands that won, Chabani, et cetera,
0:29:47 you can build multi-billion-dollar businesses.
0:29:51 I’m skeptical if that’s true in CPG going forward.
0:29:52 I mean, it’s almost math
0:29:56 because the total grocery market’s growing one or 2% a year.
0:29:58 If the number of brands proliferate,
0:30:00 the average revenue per brand should come down.
0:30:01 That’s exactly right.
0:30:04 But there’s no possibility for a player
0:30:08 like with Coke or Pepsi or white vitamin water or LaCroix.
0:30:10 So there’s certainly some very, very smart people
0:30:12 that disagree with me on this point.
0:30:17 I don’t see a world in which consumers want less options
0:30:19 where they’re saying, “Look, I want to go to the grocery store.
0:30:20 I want to go on Instacart.
0:30:22 And I only want to see one chocolate bar.
0:30:25 I don’t want to see something that’s different
0:30:27 for Sonal versus Ryan versus Jeff.
0:30:28 We don’t foresee that happening.”
0:30:31 Yeah, my view on that whole debate always comes down
0:30:34 to when people talk about more choice versus less choice.
0:30:36 Always comes down to actually what is the right choice for me.
0:30:38 And that does go to your point about personalization.
0:30:41 But the nuance I would say and data
0:30:42 is that it doesn’t necessarily have to do
0:30:43 with how many choices you have,
0:30:45 but that the right choices are presented to you.
0:30:45 Absolutely.
0:30:47 And that’s going to be a huge challenge.
0:30:49 Whoever the grocery chain or in this case,
0:30:50 or in the case of Instacart,
0:30:53 is to be able to present the right choices to you.
0:30:55 But we don’t think that the answer is going to be,
0:30:57 “Let’s strip out the choices altogether.”
0:30:57 Right.
0:31:01 And in 20 years ago, the choice was Coke or Pepsi.
0:31:03 But if now you go into the beverage aisle
0:31:06 and there’s things that I don’t even recognize where I am.
0:31:06 Right.
0:31:07 You know, just saying you’re just like–
0:31:08 I really don’t even understand
0:31:09 still why people love LaCroix so much.
0:31:12 It’s just like water with like a little hint of a taste.
0:31:12 I still don’t get it.
0:31:13 It is passion fruit.
0:31:16 I’m so confused over this.
0:31:17 Well, let’s talk about data then.
0:31:19 So we’ve been scooting around this topic of data for a while.
0:31:20 The biggest thing I’ve heard so far
0:31:23 as we talk about online to offline
0:31:25 is that in the offline world,
0:31:27 it has been nearly impossible to get the data we want.
0:31:29 On one hand, I’ve heard you say, Ryan,
0:31:31 that there are many data sources out there
0:31:32 that are really good proxies
0:31:34 and that we can do a lot.
0:31:36 But Jeff, you’ve also said that people are starved for data,
0:31:38 that things are missing.
0:31:40 So tell me what the status is in the world of CPG and data.
0:31:43 Like where are we right now really on where–
0:31:46 how far data– what data can do in this world?
0:31:47 Well, so I think both are true.
0:31:49 Both that there is a lot of data
0:31:51 and that people are starved for data.
0:31:54 And the bridge there is that while there is an outrageous amount
0:31:56 of data in this industry,
0:31:58 the data is very hard to pull together.
0:32:01 And in some cases, impossible.
0:32:04 It’s just impossible to actually go get it.
0:32:08 So the data that’s out there around distribution,
0:32:09 around product uniqueness,
0:32:11 how unique a product is relative to its competitive set,
0:32:14 on what the user thinks of the product,
0:32:17 on the competitive set of that product, et cetera,
0:32:19 all that data is out there in the world.
0:32:20 And if you begin to kind of think about
0:32:22 where would I get that data?
0:32:24 Well, most brands are not trying to hide.
0:32:26 There’s no concept of living in stealth in consumers.
0:32:29 Yeah, you can’t really, like, hide your toothpaste.
0:32:32 That’s right. When I launch in my chocolate bar in the Whole Foods,
0:32:34 Whole Foods wants you to know that you can buy the Whole Foods,
0:32:36 and I want you to know you can buy the Whole Foods.
0:32:39 Similarly, I want you to know what’s in the product.
0:32:40 How you have an ingredient deck and nutritional panel
0:32:43 that’s mandated by the FDA, but it’s out there.
0:32:46 Then consumers talk about my chocolate bar.
0:32:47 That’s also key.
0:32:50 So people can begin to understand what people think of it.
0:32:53 They get the sentiment analysis of the data of the chocolate bar.
0:32:55 So if you think about, you know,
0:32:58 rewind to the first year slack was in existence, right?
0:33:00 So if you could see every customer that used slack,
0:33:03 what they paid for it, what the end users thought of the product,
0:33:05 and how all of those things changed every single month,
0:33:08 that’d be pretty valuable data for a lot of people to have.
0:33:12 You have that data on every single CPG company in existence.
0:33:13 It’s out there.
0:33:13 Who has it?
0:33:15 Well, that’s the challenge, right?
0:33:19 So the challenge is basically it’s out there living somewhere on Google.
0:33:22 Meaning if you Google the smallest brand you can find,
0:33:24 you can see everything that I just said.
0:33:29 The challenge, though, is how do you pull all that unstructured data together
0:33:30 and then normalize it?
0:33:32 So I see a brand on Amazon.
0:33:36 I see the same brand being sold and Whole Foods being talked about on Instagram, etc.
0:33:39 So a process that we call entity resolution.
0:33:42 I remember this from NLP where you have to essentially find the same entity
0:33:43 with different variations or names on it,
0:33:46 but be able to resolve that it is the same entity.
0:33:47 It’s an incredibly hard problem.
0:33:49 Right. It’s a lot harder than people think.
0:33:50 Yeah. And then on top of that,
0:33:55 you also have the problem of how do you know that that product is a chocolate bar
0:33:56 and not a pair of shoes?
0:33:57 That sounds like an easy problem.
0:34:00 I have eaten chocolate bars that taste like a bar of shoes.
0:34:03 You’re right. Chocolate bars do taste like shoes.
0:34:04 Not that I’ve actually eaten shoes.
0:34:06 I mean, have you eaten shoes?
0:34:07 You’re a basketball player.
0:34:08 I always put my foot in my mouth.
0:34:11 So that’s another challenge.
0:34:13 Is that now you think about the people that want to consume this data.
0:34:20 Grocery chains, large CPG companies, they’re not equipped to pull all that data
0:34:24 and structure data in, normalize it, make sense of it, match it together.
0:34:25 That’s not a core competency for them.
0:34:29 That’s why I strongly believe it’ll be a technology company that does that
0:34:31 and then sells that data to others.
0:34:34 Right. So that’s where you believe that CPG companies can compete
0:34:38 and where technology has a place to play by providing that data.
0:34:43 There is a ton of data, and yet people are running blind.
0:34:47 It’s not the data they need.
0:34:52 There’s data around. It’s not helping to make the important decisions
0:34:53 that are moving the needle.
0:34:56 In some cases, it’s not the data that they need.
0:34:58 In other cases, they’re not equipped to digest the data.
0:35:03 And you think of, okay, so at the grocery store, the chocolate buyer,
0:35:07 going back to that example, that’s trying to pick what chocolate bars are on their shelves,
0:35:11 they are used to, for their entire career,
0:35:14 basing it off of either A, their own taste, literally.
0:35:17 That’s how they try product or that’s how they decide products.
0:35:20 Or B, a small amount of retail-level sales,
0:35:23 bought from a very structured data source over and over again.
0:35:25 And that’s not competitive, right?
0:35:26 Because every company has access to that same data.
0:35:28 Every grocery store has access to that same data.
0:35:31 Everyone’s got this commoditized data.
0:35:32 Everyone’s using the exact same thing.
0:35:34 So now when you go to them and you say,
0:35:37 well, actually there’s this whole universe of data out there,
0:35:42 covers 30 to 50 times as many companies as your existing data source does.
0:35:42 Do you want to use it?
0:35:44 The answer is yes, but.
0:35:47 And the but is, well, how do I digest that?
0:35:51 I’m stuck in Excel, Windows 95.
0:35:53 That’s what I’m stuck using here.
0:35:55 How do I digest all of this data?
0:35:56 Who’s going to pull this together for me?
0:35:58 Because by the way, I’ve got literally one engineer
0:36:02 that works at the entire company, entire division that I’m in right now.
0:36:03 How am I going to do this?
0:36:05 That’s why we think it’s an outsource provider.
0:36:08 We think grocery stores and CBG companies
0:36:10 are either going to be buying it from a technology company
0:36:12 or they’ll have to buy that technology company,
0:36:13 bring it in-house.
0:36:16 It makes me think a little bit of automotive and traditional
0:36:19 Detroit car companies trying to become autonomous car companies.
0:36:22 And I think it’s fascinating because they could potentially
0:36:23 win on that front if the right competency is.
0:36:25 There are some really smart people in the grocery business.
0:36:26 Yeah, exactly.
0:36:27 Walmart’s diving in.
0:36:31 They’re also in top line sales right now in stores.
0:36:33 That said, it’s hard.
0:36:36 A lot of the best engineers might want to work for a grocery store
0:36:38 and might not want to live in Bentonville, Arkansas.
0:36:41 And then they’re incredibly low margin businesses.
0:36:44 So how do you hire a ton of really expensive engineers?
0:36:46 I give Walmart credit for making the bet.
0:36:47 They bought jet.
0:36:49 They bought a number of small–
0:36:50 They also have their own labs.
0:36:52 Like Walmart is known to be a little bit more innovative
0:36:54 with internal R&D than some of the other CBG players
0:36:55 that we’re talking about here.
0:36:58 But it’s not that the grocers aren’t capable
0:37:00 and don’t understand it.
0:37:02 They’re a bit of a prisoner’s dilemma.
0:37:05 There are extremely, extremely smart people working at both
0:37:08 large CBG companies and large grocery stores
0:37:10 that are in a really difficult position.
0:37:13 You think about what would you do if you were in their position?
0:37:14 Well, they’re out of business.
0:37:18 That’s, in the case of grocery, 2% net margin business.
0:37:20 How do they then say to their boss,
0:37:23 let’s go hire 150 engineers to go build this thing?
0:37:26 By the way, it’s not going to pay back for two or three years.
0:37:28 That’s a really, really hard proposition to make.
0:37:31 So I think incredibly talented smart people that work there,
0:37:33 it’s a hard position to be in.
0:37:35 One of the smartest people I’ve ever met in business
0:37:39 was the CEO of one of the largest consumer package companies.
0:37:42 And she came to Silicon Valley and wanted to sit down.
0:37:43 And I’m like, why are you here?
0:37:45 We believe software’s eating the world.
0:37:47 How is software eating snacks?
0:37:50 She came up with this very lucid argument of two things.
0:37:53 One is all of a sudden, there’s promotional transparency.
0:37:57 I used to move, gain share by going on special on knob,
0:38:00 Hills food one day and then special at Safeway.
0:38:02 Now the consumer knows exactly where I’m on special.
0:38:04 So I’ve lost the share moving thing,
0:38:08 all I’m doing is discounting and then nutritional transparency.
0:38:11 And so she says, I have to retool my company
0:38:15 because the two bedrocks that it was built on, software eight.
0:38:17 The challenge with that, and I, by the way,
0:38:19 I totally agree with that diagnosis.
0:38:22 The challenge then is that I don’t think that the CEOs
0:38:24 of these large CPG companies or large grocery chains
0:38:26 are given enough runway to go do that.
0:38:28 So there’s this, what we call 3G effect,
0:38:30 large financial institution down in South America
0:38:35 that has been investing into or buying public CPG companies,
0:38:36 basically stripping costs out.
0:38:39 And delivering shareholder value.
0:38:42 Problem with that is when you’re stripping the costs out,
0:38:44 one of the first things to go is R&D.
0:38:46 So you already had a problem for these large CPG companies.
0:38:48 What’s the percentage of R&D that they have?
0:38:51 So today, large CPG spends about 2% of sales in R&D.
0:38:52 Are you kidding?
0:38:55 Tech spends about 14% of sales in R&D.
0:38:57 CPG spends nothing on R&D.
0:39:01 And so- You have low margins, like 1.5 to 2% R&D.
0:39:01 That’s right.
0:39:03 So when you strip out costs,
0:39:06 I can deliver shareholder value and hit quarterly numbers
0:39:08 for the next, let’s say, year or two years.
0:39:10 But you look five, seven years out.
0:39:11 It’s not going to last long-term.
0:39:12 How far can you cut?
0:39:13 It’s like the activist investor problem, right?
0:39:14 That’s exactly right.
0:39:18 And so what we’re seeing is that many of these companies,
0:39:20 regardless of whether they’re doing that cost cutting or not,
0:39:23 they’re saying, look, we were already spending almost nothing on R&D.
0:39:24 How do we get the innovation?
0:39:26 And this goes back to a point Jeff made earlier.
0:39:27 That’s why CPG is beginning to look a lot
0:39:29 like Big Pharma has for the past 20 years.
0:39:31 So one last question then,
0:39:34 just to think about more on the data and innovation side.
0:39:36 All of us in this room are in the world of investing.
0:39:38 How does this change the investing game?
0:39:42 So in terms of how I think the investing landscape will change,
0:39:49 we have a pretty strong thesis that in CPG specifically,
0:39:52 there will be quantitative VC firms.
0:39:54 So what I mean by that is if you think of the public markets,
0:39:58 there are systematic quant funds that have historically,
0:40:00 over the last 20 years or so, in some cases 30 years,
0:40:05 invested into public companies basically just using technology.
0:40:07 It’s very different than a discretionary hedge fund
0:40:09 where you’ve got a team of really smart people
0:40:12 that research a stock for six months and make a decision.
0:40:14 In the case of a systematic quant fund,
0:40:17 they’ve got a lot of really, really smart engineers and data scientists
0:40:18 who are building algorithms to evaluate the company
0:40:20 and then make an investment decision.
0:40:23 We think that that is possible in some industries in the private markets.
0:40:26 We don’t think it’s possible in tech.
0:40:27 Why do you think that’s possible in CPG?
0:40:30 So there’s two main reasons that CPG is beautiful.
0:40:33 First, the business models are basically the same.
0:40:36 So what I mean is if I’m selling shampoo, dog food, or water,
0:40:37 the margins are different,
0:40:39 but I’m making a product and I’m selling the product.
0:40:40 It’s very different than tech.
0:40:42 In tech, I might give away the product for five years.
0:40:43 You might do a freemium, right?
0:40:44 I might do a SaaS business, et cetera.
0:40:45 Very different business models.
0:40:47 Because it’s the same business model,
0:40:49 it’s the same game of chess over and over again.
0:40:50 Interesting.
0:40:54 The second reason that there’s just an outrageous amount of data in this space,
0:40:59 but many of the dimensions that are predictive of success of a consumer company
0:41:01 are not data that you can get externally.
0:41:04 It’s hard to get information from afar.
0:41:05 In consumer, you can get it from afar
0:41:07 without ever talking to the company.
0:41:09 We have gone private financials on thousands of companies.
0:41:11 That acts as our training data.
0:41:13 The training data is private financials on thousands of companies.
0:41:15 And is that kind of providing the ground truth
0:41:16 in order to sort of compare the decisions again?
0:41:17 That’s exactly right.
0:41:18 Yeah, so it’s the ground truth.
0:41:23 So I need to know what success and failure looks like in order to predict it.
0:41:25 If you look at a SaaS company, a consumer company,
0:41:28 and an infrastructure company in tech, you can’t compare them.
0:41:31 Whereas if you’ve got two beverage companies,
0:41:38 and one is selling 20 units per Whole Foods per week at a 60% margin,
0:41:42 another selling 10 units per Whole Food per week at a 30% margin,
0:41:44 it’s pretty obvious which is the better company.
0:41:48 But also in CPG, I can find that company
0:41:50 just by using publicly available information.
0:41:52 So I can see that one of the two companies
0:41:55 started in one Whole Foods six months ago,
0:41:58 and now is in 400 Whole Foods and added 300 targets.
0:42:01 There isn’t really an equivalent in, let’s say, the SaaS world.
0:42:04 Going back to the SaaS world, while there might be structured metrics to look at,
0:42:06 and those metrics might be predictive of success,
0:42:08 almost all of them, not all, but almost all of them,
0:42:10 are only available once you start talking to the company
0:42:11 and you already know who to focus on.
0:42:14 I do wonder because when I watch a show like Billions,
0:42:16 you know, they have, they use data like empty parking lots
0:42:19 and aerial shots of an empty parking lot for retail
0:42:22 to be able to make a decision on whether to pull a trigger on a company or not.
0:42:24 The equivalent in SaaS might be developer heat,
0:42:28 like viral developer activity or the adoption rate among developers.
0:42:30 There’s various heat maps and sources.
0:42:32 The part that I have to ask on the big question
0:42:35 of quant investing in general, and especially in this case,
0:42:37 and even in our world, Jeff, which is more tech than CPG,
0:42:41 doesn’t it miss the outlier, the outsize winners?
0:42:42 It’s a great point.
0:42:43 It’s a great point.
0:42:43 Yeah.
0:42:46 So we, that’s another reason we struggle to believe
0:42:47 that it is possible in tech.
0:42:50 You think about many of the massive home runs in tech.
0:42:51 Exactly.
0:42:55 There weren’t prior examples that did something very similar.
0:42:56 No, they’re all, they’re all a priori.
0:42:57 Right. Yeah.
0:42:59 And so, so that’s why I really struggled to believe
0:43:00 that that’s possible in tech.
0:43:02 When you look at the winners and consumer,
0:43:04 the previous winners looked really similar.
0:43:06 That might have been a different,
0:43:07 might have been a different magnitude,
0:43:08 but they grew in really, really similar ways.
0:43:11 So this might be the one case where that phrase pattern recognition
0:43:13 that people throw around so wildly in the valley actually applies.
0:43:17 But who the winner was, wasn’t that still an irrational
0:43:20 behavioral thing versus something programmable?
0:43:24 So we’ve been able to show that there are some common themes
0:43:29 between both in A, why something wins and how it wins.
0:43:29 So here’s what I mean.
0:43:31 So you’re nuanceifying the two, which is really important.
0:43:31 That’s right.
0:43:36 So why typically the winners have, and look, to be clear,
0:43:37 this is not perfect.
0:43:39 There are holes, there are inaccuracies.
0:43:42 Typically the winners have brand intensity with the consumer.
0:43:44 So think of vitamin water 15 years ago,
0:43:49 kind bar 10 years ago, the brand really resonates with the consumer.
0:43:51 The second thing that is common amongst the winners in CPG
0:43:55 is the product has uniqueness that matters.
0:43:55 Interesting.
0:43:58 So kind bar is actually a pretty good example here.
0:43:59 When it first came out, everyone said,
0:44:00 “This is a really crowded category.”
0:44:03 By the way, everyone always says every CPG category is crowded.
0:44:06 Because they are, because they are, but they’re also massive.
0:44:08 Right, crowded but massive.
0:44:09 I like that distinction.
0:44:12 So kind bar, at the time though, what people missed was,
0:44:15 there’s Nature’s Valley, there’s Cliff Bar, a number of others.
0:44:18 When you look at those products, they didn’t look like real food.
0:44:21 Kind bar had an insight, which was,
0:44:23 “Let’s make a product that actually just looks like real food.”
0:44:24 That’s actually why I’m drawn to them.
0:44:26 I don’t feel like I’m eating a crappy processed bar.
0:44:28 Yep, and they actually show the food.
0:44:30 The biggest innovation for Guy Bar from my perspective is,
0:44:33 it’s see-through and you can see that there’s real food.
0:44:33 So you can see a whole lot to that.
0:44:35 And it’s not symmetrical.
0:44:35 Yeah, that’s exactly right.
0:44:36 Like it’s not this perfect rectangle.
0:44:38 It’s got like jagged edges where you can see it’s real.
0:44:39 It’s not processed.
0:44:41 Exactly, exactly, exactly.
0:44:45 So in that case, you know, you can build an algorithm
0:44:47 which evaluates the picture of the package.
0:44:49 And literally just says, “Is this different?”
0:44:51 So we’ve done that in the case of snack bars.
0:44:52 We track about 3,000 snack bars.
0:44:54 So you answer that question of, “Is it different?”
0:44:54 “Is it different?”
0:44:57 And then, now I could put fish in my snack bar
0:44:57 and it’d be different then,
0:44:59 but it may not resonate with the consumer.
0:44:59 Right, that would be great.
0:45:01 So those are two kind of orthogonal dimensions.
0:45:03 One is does it resonate with the consumer
0:45:05 and is the product unique?
0:45:08 The ones that have one tended to be high in both.
0:45:11 Then there’s the question of how it wins, right?
0:45:13 So why it wins and then how it wins.
0:45:16 The how it wins is typically distribution gains.
0:45:17 Distribution gains.
0:45:18 Meaning–
0:45:20 It’s because it’s not the one of Alex’s famous lines.
0:45:22 Like it’s always about distribution.
0:45:24 Yeah, yeah, it’s, and that is very true in CBG.
0:45:26 So there aren’t a lot of big winners
0:45:28 that have only been sold in five stores.
0:45:30 That’s not a thing.
0:45:31 So then if you think about,
0:45:34 okay, if it’s number of, it’s breadth and quality of doors,
0:45:35 meaning TG Max not as valuable
0:45:37 as let’s say Whole Foods or Costco.
0:45:40 So how do you measure breadth and quality?
0:45:41 That data is out there.
0:45:44 Where the product is being sold is out there in the world.
0:45:45 Whole Foods wants you to know it.
0:45:46 The brand wants you to know it.
0:45:48 Now it’s just aggregating that information
0:45:49 and making sense of it.
0:45:50 But that’s the how it wins.
0:45:52 And you can also get into price points.
0:45:53 You can get into skew count,
0:45:54 number of other things.
0:45:56 But marrying those two things together is the foundation.
0:45:58 But it’s basically one business model.
0:46:00 A couple of the differences in tech.
0:46:02 We’ve been trying to figure out,
0:46:04 how do you leverage data in the decision-making process?
0:46:05 Is this a holy grail?
0:46:09 Yes, we can probably define a dozen, two dozen,
0:46:12 three dozen different worlds of,
0:46:15 you know, universes of tech companies that are unique.
0:46:18 You have SaaS and open source and consumer.
0:46:20 And then within consumer, you have social and e-commerce.
0:46:23 So, you know, just by definition, comparison’s hard.
0:46:26 And then the other part you have is what you said,
0:46:28 most of the huge winners are lightning strikes.
0:46:30 Facebook was the second or third social network.
0:46:32 But, you know, no one was thinking social networks
0:46:35 when they talk about a trillion-dollar company.
0:46:40 eBay was collectibles online, weird, Facebook hot or not.
0:46:44 You know, these things, I mean, the Airbnb sleep on someone’s couch.
0:46:47 You know, these things don’t present as monster companies.
0:46:51 They start, hit a chord and are expanded
0:46:53 by their community of users.
0:46:54 And the thing I would add to that, by the way,
0:46:56 is that there’s a complexity math to this,
0:46:57 which is a little bit like the Brian Arthur,
0:46:59 we had Brian Arthur on the podcast.
0:47:00 And he’s the father of network effects
0:47:02 and network effects theory.
0:47:05 And the point of when that tipping point hits,
0:47:07 you don’t know when it tips.
0:47:07 That’s the hard one.
0:47:08 Yeah, yeah.
0:47:10 Like the when it seems like the really tough question.
0:47:11 Will it.
0:47:12 Then when, yeah, so it’s good.
0:47:14 And if you think of Jeff’s point,
0:47:15 like when Facebook hit,
0:47:17 there was two other social networks and maybe three.
0:47:17 Yeah.
0:47:19 There weren’t 300 or 3000.
0:47:20 Right.
0:47:21 There were very few examples.
0:47:24 You don’t have 40 or 400 examples
0:47:26 to begin to look at pattern recognition.
0:47:26 Right.
0:47:28 The n equals three or the CPG.
0:47:30 You’ve got massive data set.
0:47:33 So now I can compare certainly one ice cream company
0:47:34 to the many others that have hit.
0:47:36 But the difference between ice cream companies
0:47:37 and snack bars is not great enough
0:47:40 that you can’t compare the two and begin to see patterns.
0:47:42 Well, that was a fascinating discussion.
0:47:45 Ryan, Jeff, thank you for joining the ice cream company.
0:47:46 It’s a pleasure.
0:47:46 Thank you.
0:47:47 Thank you.
0:47:57 [BLANK_AUDIO]

0:00:02 Hi, and welcome to the A16Z podcast.
0:00:07 I’m Hannah, and this episode is a Valentine special where I talk with Kate Darling, researcher
0:00:12 at MIT Labs, all about our emotional relationships with robots.
0:00:16 We already know that we have an innate tendency to anthropomorphize robots, but as we begin
0:00:20 to share more and more spaces, both social and private, with these machines, what does
0:00:23 that actually mean for how we’ll interact with them?
0:00:28 From our lighter sides, from affection and love and emotional support to our darker sides,
0:00:33 what do these relationships teach us about ourselves, our tendencies, and our behaviors?
0:00:36 How will these relationships in turn change us?
0:00:40 And what models should we be thinking about as we develop these increasingly sophisticated
0:00:43 relationships with our robots?
0:00:47 Besides just that natural instinct that we have to anthropomorphize all sorts of things,
0:00:49 how is it different with robots?
0:00:54 Robots are just so fascinating because we know rationally that they’re machines, that
0:00:57 they’re not alive, but we treat them like living things.
0:01:04 With robots, I think they speak to our primal brain even more than a stuffed animal or some
0:01:10 other object that we might anthropomorphize because they combine movement and physicality
0:01:14 in this way that makes us automatically project intent onto them.
0:01:19 So that’s why we project more onto them than the Samsung monitor that I’m looking at back
0:01:23 here because it looks like it has agency in the world?
0:01:24 Yeah.
0:01:25 I think that tricks our brains.
0:01:28 There’s a lot of studies that show that we respond differently to something in our physical
0:01:31 space than just something on a screen.
0:01:35 So even though people will imbue everything with human-like qualities, robots take it
0:01:39 to a new level because of this physical movement.
0:01:44 People will do it even with very, very simple robots, just like the Roomba vacuum cleaner.
0:01:49 It’s not a very compelling anthropomorphic robot, and yet people will name them and feel
0:01:54 bad for them when they get stuck and insist on getting the same one back.
0:02:01 If it gets broken, and so if you take that and then you create something that is specifically
0:02:05 designed to push those buttons in us, then it gets really interesting.
0:02:10 So what is the reason why we should be aware of this tendency besides this cute attachment
0:02:15 to our Roombas or our vectors or whatever our pet robots are?
0:02:18 Why does it matter that we develop these relationships with them?
0:02:23 Well, I think it matters because right now, robots are really moving into these new shared
0:02:24 spaces.
0:02:30 I mean, we’ve had robots for decades, but they’ve been increasing efficiency in manufacturing
0:02:31 contexts.
0:02:34 They haven’t been sharing spaces with people.
0:02:39 And as we integrate these robots into these shared spaces, it’s really important to understand
0:02:43 that people treat them differently than other devices, and they don’t treat them like toasters.
0:02:49 They treat them subconsciously like living things, and that can lead to some almost comical
0:02:58 challenges as we try and figure out how to treat these things as tools in contexts where
0:03:03 they’re meant to be tools, but then at the same time kind of want to treat them differently.
0:03:08 When you talk about these giant manufacturing robots that do exist in plants and factories
0:03:11 on the floor, do we see that there?
0:03:17 So there’s this company in Japan that has this standard assembly line for manufacturing,
0:03:23 and a lot of companies, they have people working alongside the robots on the assembly line,
0:03:27 and so the people will come in and they do these aerobics in the morning to warm up
0:03:32 their bodies for the day, and they have the robots do the aerobics with their arms with
0:03:37 the people so that they’ll be perceived more like colleagues and less as machines.
0:03:40 People are more accepting of the technology.
0:03:45 People enjoy working with it more, and I think it’s really important to acknowledge this
0:03:48 emotional connection because you can harness it too.
0:03:50 So we know we have this tendency.
0:03:57 If we think about being aware of it and being able to foster it or diminish it, what are
0:04:00 some of the ways in which we negotiate those relationships?
0:04:05 My focus is trying to think about what that means ethically, what that means in terms
0:04:12 of maybe changing human behavior, what challenges we might want to anticipate.
0:04:18 We’re seeing a lot of interesting use cases for social robots that are specifically designed
0:04:22 to get people to treat them like living things, to develop an emotional connection.
0:04:29 One of my favorite current use cases for this technology is a replacement for animal therapy.
0:04:30 We have therapeutic robots.
0:04:35 They’re used as medical devices with dementia patients or in nursing homes.
0:04:39 I saw an article about that recently, specifically with people with dementia.
0:04:41 Yeah, there’s a baby seal that’s very popular.
0:04:42 That’s right.
0:04:43 That’s the one I read about.
0:04:45 A lot of people think it’s creepy.
0:04:50 We’re giving old people robots and having them nurture something that’s not alive.
0:04:57 But then you look at some of the advantages of having them have this experience when their
0:05:00 lives have been reduced to being taken care of by other people.
0:05:04 It’s actually an important psychological experience for them to have.
0:05:10 They’ve been able to use these robots as alternatives to medication for calming distressed patients.
0:05:15 This isn’t a replacement for human care and that’s also not how it’s being used.
0:05:20 It’s really being used as a replacement for animal therapy where we can’t use real animals
0:05:25 because people will consistently treat them more like a living thing than a device.
0:05:29 What is the initial interaction like when you hold something like that?
0:05:31 Is there a prelude that’s necessary?
0:05:36 Do you have to educate a little bit those patients or do they just put the robot seal
0:05:38 in their arms?
0:05:43 The most clever robotic design doesn’t require any prelude or anything because you will
0:05:46 automatically respond to the cues.
0:05:48 The baby seal is very simple.
0:05:53 It just makes these little cute sounds and movements and responds to your touch and will
0:05:56 purr a little bit.
0:06:02 It’s very intuitive and it’s also not trying to be a cat or anything that you would be
0:06:07 more intimately familiar with because no one has actually held a baby seal before.
0:06:11 It’s much easier to suspend your disbelief and just go with it.
0:06:18 What are some of the very broad umbrella concerns that we want to be thinking about as we’re
0:06:20 watching these interactions develop?
0:06:24 A lot of my work has been around empathy and violence towards robotic objects.
0:06:28 Are we already being violent towards them?
0:06:29 Sometimes.
0:06:35 There was this robot called Hitchbot that hitchhiked all the way across the entire country of Canada,
0:06:38 just relying on the kindness of strangers.
0:06:42 It was trying to do a road trip through the US and it made it to Philadelphia and then
0:06:44 it got vandalized beyond repair.
0:06:49 Of course, Philadelphia, by the way, because I’m from New Jersey.
0:06:55 As you’re telling me this story, I’m already imagining this alien life doing a little journey
0:06:56 through the world.
0:07:01 I’m completely projecting this narrative onto it and that was the interesting thing about
0:07:02 the story.
0:07:06 It wasn’t that the robot got beat up, but it was people’s response to that, that they
0:07:12 were empathizing with this robot that was just trying to hitchhike around and that it
0:07:15 got … People were so sad when this robot got …
0:07:17 It’s poor little stranger in a strange land.
0:07:18 Yeah.
0:07:22 There was news about this all over the world that hit international news.
0:07:23 What did we learn from that?
0:07:26 Why is it interesting that we empathize with them?
0:07:31 Even more interesting to me is the question, how does interacting with these very life-like
0:07:35 machines influence our behavior?
0:07:42 Could you use them therapeutically to help children or prisoners or help improve people’s
0:07:43 behavior?
0:07:49 Then the flip side of that question is, could it be desensitizing to people to be violent
0:07:53 towards robotic objects that behave in a really life-like way?
0:07:59 Is that a healthy outlet for people’s violent behavior to go and beat up robots that respond
0:08:04 in a really life-like way, or is that training our cruelty muscles?
0:08:09 Isn’t that a new version of almost the old video game argument?
0:08:11 How is it shifting?
0:08:17 It’s the exact same question, which by the way, I don’t think we’ve ever really resolved.
0:08:24 We mostly decided that people can probably compartmentalize, but children we’re not
0:08:29 sure about, and so we restrict very violent games to adults.
0:08:36 So we’ve decided that we might want to worry about the kids, but adults can probably handle
0:08:37 it.
0:08:44 Now, robots, I think, make us need to re-ask the question because they have this visceral
0:08:52 physicality that we know from research people respond differently to than things on a screen.
0:08:57 There’s a question of whether we can compartmentalize as well with robots, specifically because
0:09:01 they are so present in the world with us.
0:09:06 So do you think that’s because it’s almost a somatic relationship to them?
0:09:10 Will it matter the same way when we are immersed in, say, virtual reality?
0:09:15 I mean, as virtual reality gets more physical, I think that the two worlds merge.
0:09:23 And so even though the answer could very well be people can still distinguish between what’s
0:09:29 fake and what’s real, and just because they beat up their robot doesn’t mean that they’re
0:09:33 going to go and beat up a person or that their barrier to doing that is lower, but we don’t
0:09:34 know.
0:09:35 How do you start looking at that?
0:09:39 What are the details that start giving you an inkling one way or the other?
0:09:45 The way that I think we’re beginning to start to get at the question is just trying to figure
0:09:48 out what those relationships look like at first.
0:09:54 So I’ve done some work on how do people’s tendencies for empathy relate to their hesitation
0:09:57 to hit a robot?
0:10:01 Just to try and establish that people do empathize with the robots because we don’t…
0:10:02 We have to show that first.
0:10:04 Yeah, we have to show that first.
0:10:05 It’s so interesting.
0:10:10 We all know what that feeling is, but to show, to demonstrate, to model it, and then see
0:10:17 it and recognize it in our kind of research experimentation, how do you actually categorize
0:10:19 the response of empathy?
0:10:24 One of the things we did was have people come into the lab and smash robots with hammers
0:10:31 and time how long they hesitated to smash the robot when we told them to smash it.
0:10:36 Did you give them a framework around this experiment or just have them walk in and just
0:10:37 start?
0:10:38 Definitely.
0:10:42 They did not know that they were going to be asked to hit the robot and we did psychological
0:10:48 empathy testing with them to try and establish a baseline for how they scored on empathic
0:10:53 concern generally, but also we had a variety of conditions.
0:10:59 So what we were trying to look at was a difference, for example, in would people hesitate more
0:11:05 if the robot had a name and a backstory versus if it was introduced to them as an object?
0:11:08 Oh, well, presumably the name and the backstory, right?
0:11:09 Yeah.
0:11:15 Yeah, not a huge surprise that when the robot’s name is Frank, people hesitate more.
0:11:17 So sorry, Frank.
0:11:25 We actually tried measuring slight changes in the sweat on their skin to see if they
0:11:29 were more physically aroused.
0:11:33 Unfortunately, those sensors were really unreliable, so we couldn’t get reliable data from that.
0:11:39 We tried coding the facial expressions, which was also difficult.
0:11:43 That’s what I was wondering about, because as one human, reading another human, you do
0:11:46 have some sense, right?
0:11:52 And I have to say the videos of this experiment are much more compelling than just the hesitation
0:11:59 data because people really did, like one woman was looking at this robot, which was a very
0:12:05 simple, looked kind of like a cockroach, was just a thing that moved around like an insect.
0:12:10 And so this one woman is holding the mallet and stealing herself, and she’s muttering
0:12:11 to herself.
0:12:12 It’s just a bug.
0:12:13 It’s just a bug.
0:12:21 So the videos were compelling, but we just didn’t find it easy enough to code them in
0:12:25 a way that would be scientifically sound or reliable.
0:12:27 So we relied just on the timing of the hesitation.
0:12:33 Other studies have measured people’s brainwaves while they watch videos of robots being tortured.
0:12:36 So there are a bunch of different ways that people have tried to get at this.
0:12:42 So when we start learning about our capacity for violence towards robots, are you thinking
0:12:48 about that in terms of what it teaches us back about humans or about what going forward,
0:12:50 the reason we need to know this?
0:12:56 We are learning actually more about human psychology as we watch people interact with
0:13:02 these machines that don’t communicate back to us in an authentic way.
0:13:08 So that’s interesting, but I think that it’s mainly important because we’re already facing
0:13:12 some questions of regulating robots.
0:13:16 For example, there’s been a lot of moral panic around sex robots.
0:13:21 We already need to be answering the question, do we want to allow this type of technology
0:13:23 to exist and be used and be sold?
0:13:27 Do we want to only allow for it in therapeutic contexts?
0:13:29 Do we want to ban it completely?
0:13:34 And the fact is we have no evidence to guide us in what we should be doing.
0:13:39 So it’s all coming down to the same question of like, is this desensitizing or is this
0:13:40 enhancing basically?
0:13:41 Yeah.
0:13:48 Unfortunately, a lot of the discussions are just fueled by superstition or moral panic
0:13:53 or in this context, a lot of it is science fiction and pop culture.
0:14:00 And our constant tendency to compare robots to humans and look at them as human replacements
0:14:04 versus thinking a little bit more outside of the box and viewing them as something that’s
0:14:07 more supplemental to humans.
0:14:10 Do we have a model for what that even might be?
0:14:18 I’ve been trying to argue that animals might be the better analogy to these machines that
0:14:24 can sense and think and make autonomous decisions and learn and that we treat like they’re alive,
0:14:31 but we know that they’re not actually alive or feel anything or have emotions or can make
0:14:33 moral decisions.
0:14:36 They are still controlled by humans.
0:14:37 Property.
0:14:38 Property.
0:14:39 They’re property.
0:14:44 And throughout history, we’ve treated some animals as property, as tools.
0:14:50 Some animals we’ve turned into our companions and I think that that is how we’re going
0:14:52 to start integrating robotic technology as well.
0:14:56 We’re going to be treating a lot of it like products and tools and property.
0:15:01 And some of it we’re going to become emotionally attached to and we might integrate in different
0:15:02 ways.
0:15:08 But we definitely should stop thinking about robots as human replacements and start thinking
0:15:14 about how to harness them as a partner that has a different skill set.
0:15:18 So while you’re talking, I’m thinking about the incredibly fraught space of how we relate
0:15:20 to animals.
0:15:25 Some people might argue that since that’s such a gray area as it is and we’re always
0:15:30 feeling our way, and that model is always changing, it almost sounds like it just makes
0:15:33 it messier in a way.
0:15:38 And I also think there’s a way in which we have this primal instinct of how to relate
0:15:39 to animals.
0:15:44 Do you think we have the same kind of seed for a primal relationship with robots there?
0:15:45 I think we do.
0:15:51 I think that ironically, we’re learning more about our relationship to animals through
0:15:56 interacting with robots because we’re realizing that we’re complete hypocrites.
0:15:57 Oh.
0:15:58 Well, yeah.
0:16:07 I think we fancy ourselves as caring about the inner biological workings of the animals
0:16:09 and whether animals can suffer.
0:16:12 And we actually don’t care about any of that.
0:16:15 We care about what animals we relate to.
0:16:19 And a lot of that is cultural and emotional, and a lot of that is based on which animals
0:16:21 are cute.
0:16:26 For example, in the United States, we don’t eat horses.
0:16:31 It’s considered taboo, whereas in a lot of parts of Europe, people are like, well, horses
0:16:33 and cows are both delicious.
0:16:35 Why would you distinguish between the two?
0:16:38 There’s no inherent biological reason to distinguish.
0:16:39 Right.
0:16:40 And by the way, we boil them into glue.
0:16:46 And yet culturally, we feel this like bond with horses in the U.S. as this majestic beast,
0:16:50 and it seems so wrong to us to eat them.
0:16:53 The history of animal rights is full of stories like this.
0:17:01 Like, the Save the whales campaign didn’t start until people recorded whales singing.
0:17:05 Before that, people did not care about whales, but then once we heard that they can sing
0:17:09 and make this beautiful music, we were like, oh, we must save these beautiful creatures
0:17:12 that we can now suddenly relate to.
0:17:15 Because it needs to be about us kind of on some deep level.
0:17:21 The sad but important realization is that we relate to things that are like us and we
0:17:26 can build robots that are like that, and we are going to relate to those robots more than
0:17:27 to other robots.
0:17:33 So it’s a principle almost of design thinking, then, when you think about, well, I want this
0:17:40 robot to have a relationship to humans like cattle pulling a plow.
0:17:44 It gives you a sort of vision of a different spectrum of relationships for starters.
0:17:47 I mean, we’ve even tried to design animals accordingly.
0:17:53 We’ve bred dogs to look specific ways, so that we relate more to them.
0:17:57 And the interesting thing about robots is that we have even more freedom to design them
0:18:00 in compelling ways than we do with animals.
0:18:02 It takes a while to breed animals.
0:18:03 Yeah, generations.
0:18:11 Yeah, so I think we’re going to see the same types of manipulations of the robot breeds.
0:18:16 Why would you go down on that spectrum to the lesser relationships when it’s something
0:18:19 that is performing a service to humans?
0:18:25 If it’s not directly harmful to have people develop an emotional attachment, it’s probably
0:18:27 not a bad idea to do.
0:18:34 But a lot of the potential for robots right now is in taking over tasks that are dirty,
0:18:36 dull, and dangerous.
0:18:42 And so if we’re using robots as tools to go do the thing, it might make sense to design
0:18:45 them in a way that’s less compelling to people so that we don’t feel bad for them when they’re
0:18:48 doing the dirty, dull, dangerous work.
0:18:50 There are contexts where it can be harmful.
0:18:55 So for example, you have in the military, you have soldiers who become emotionally attached
0:18:57 to the robots that they work with.
0:19:01 And that can be anything from inefficient to dangerous because you don’t want them hesitating
0:19:06 for even a second to use these machines the way that they’re intended to be used.
0:19:07 Like police dogs.
0:19:09 That’s a great analogy.
0:19:16 If you become too attached to the thing that you’re working with, if it’s intended to
0:19:21 go into harm’s way in your place, for example, which is a lot of how we’re using robots
0:19:26 these days, bomb disposal units, stuff like that, you don’t want soldiers becoming emotionally
0:19:33 affected by sending the robot into harm’s way because they could risk their lives.
0:19:37 So it’s really important to understand that these emotional connections we form with these
0:19:40 machines can have real world consequences.
0:19:48 Another interesting area is responsibility for harm because it does get a lot of attention
0:19:51 from policymakers and from the general public.
0:19:56 With robots generally, there’s a lot of throwing up our hands, like how can we possibly hold
0:20:00 someone accountable for this harm if the robot did something no one could anticipate?
0:20:09 I think we’re forgetting that we have a ton of history with animals where we have things
0:20:14 that we’ve treated as property that can make autonomous unpredictable decisions that can
0:20:15 cause harm.
0:20:21 So there’s this whole body of legislation that we can look to, basically.
0:20:22 Yes.
0:20:26 The smorgasbord of different solutions we’ve had is really compelling.
0:20:34 The Romans even had rules around, if your oxen tramples the neighbor’s field, the neighbor
0:20:40 might actually be able to appropriate your oxen or even kill your oxen.
0:20:47 We’ve had animal trials, I talked about that in a podcast with Peter Leeson about the trials
0:20:49 of the rats for decimating crops.
0:20:56 There’s different ways even today that we like to assign responsibility for harm.
0:21:01 There’s the very pragmatic, okay, how do we compensate the victim of harm?
0:21:05 How do we hold the person who caused the harm accountable so that there’s an incentive to
0:21:07 not do it again?
0:21:11 And a lot of that is done through civil liability.
0:21:19 There’s also, however, criminal law that is kind of a primitive concept when you think
0:21:20 about it.
0:21:26 There was just this case in India where an old man was stoned to death with bricks by
0:21:29 monkeys who were intentionally flinging bricks.
0:21:35 And the family tried to get the police to do something about the monkeys and hold the
0:21:39 monkeys criminally accountable for what happened.
0:21:42 Just because of that human assigning of blame?
0:21:50 Yes, because it wasn’t enough to just have some sort of monetary compensation.
0:21:56 They really wanted these monkeys to suffer a punishment for what they had done.
0:21:59 And I know it seems silly, but we do sometimes have that tendency.
0:22:05 So it’s interesting to think about ways that we might actually want to hold machines themselves
0:22:08 accountable and ways that that’s problematic as well.
0:22:13 So can you illustrate what that would look like with robots when we think about those
0:22:15 different ways of assigning responsibility?
0:22:16 Yeah.
0:22:22 So for example, the way that we regulate pit bulls currently in some countries is really
0:22:23 interesting.
0:22:29 Austria has decided there are some breeds of dogs that we are going to place much stricter
0:22:35 requirements on than just the other dog breeds.
0:22:41 So you need to get what’s basically the equivalent of a driver’s license to walk these dogs.
0:22:46 They have to have special collars and they have to be registered.
0:22:50 And you could imagine for certain types of robots having a registry, having requirements,
0:22:57 having a different legal accountability, like strict liability versus, “Oh, did I intend
0:23:00 to cause this harm or did I cause it through neglect?”
0:23:05 The way that we distinguish, for example, between wild animals and pets.
0:23:10 If you have a tiger and the tiger kills the postal service worker, that’s going to be
0:23:14 your fault regardless of how careful you were with the tiger because we say having a tiger
0:23:16 is just inherently dangerous.
0:23:21 It’s almost, the model is sort of developing different ideas around certain categories
0:23:28 and groups of the way we relate to them, depending on whether those relationships are based on
0:23:35 then sort of our emotional narratives around it or evidence based becomes really important.
0:23:41 The heart of it is that we need to recognize that social robots could have an impact on
0:23:46 people’s behavior and that it’s something that we might actually need to regulate.
0:23:51 One of the interesting conversations that’s happening right now is around autonomous weapon
0:23:57 systems and accountability for harm in settings of war, where we have war crimes, but they
0:24:00 require intentionality.
0:24:06 And if a robot is committing a war crime, then there’s maybe not this moral accountability.
0:24:10 But wouldn’t it be an obvious whoever programmed and owns the robot?
0:24:16 Because you need someone to have intentionally caused this rather than accidentally.
0:24:22 The thing about robots is that they can actually now make decisions based on the data that
0:24:30 they gather that isn’t a glitch in the code, but is something that we didn’t foresee happening.
0:24:36 We’ve used autonomous unpredictable agents as weapons in war previously.
0:24:45 For example, the Soviets, they trained dogs to run under tanks, enemy tanks, and they
0:24:51 had explosives attached to them and they were meant to blow up the tanks.
0:24:53 And a bunch of things went wrong.
0:25:00 So first of all, they had trained the dogs on their own tanks, which means that the
0:25:09 dogs would sometimes blow up their own tanks instead of the enemy tanks.
0:25:15 They didn’t train the dogs to be able to deal with some of the noise on the battlefield
0:25:21 and the shooting, so the dogs got scared and would run back to their handlers with these
0:25:27 explosives attached to them, and the handlers had to end up shooting the dogs.
0:25:31 And we’re not perfect at programming robots either, so there’s a lot of things that can
0:25:37 go wrong that don’t necessarily, they’re not glitches in code.
0:25:40 It’s unanticipated consequences.
0:25:44 So when we’re thinking about regulating things, I think that’s a pretty good analogy to look
0:25:50 at the history of how we’ve handled these things in the past and who we’ve held accountable.
0:25:55 The interesting thing that occurs to me is how do we both acknowledge our human emotional
0:25:59 attachment and yet not let it direct us too much?
0:26:00 What’s that balance like?
0:26:02 Step one is probably awareness, right?
0:26:07 But is it something we can manage and navigate or is it kind of beyond our control?
0:26:14 I think we struggle with that culturally as well, because we have this Judeo-Christian
0:26:19 distinction, like we have this clear line between things that are alive and things that are
0:26:24 not alive, whereas in some other countries, they don’t necessarily make that distinction.
0:26:29 Like in Japan, they have this whole history of Shintoism and treating objects as things
0:26:30 with souls.
0:26:39 And so it’s, I think, easier for them to view robots as just another thing with a soul and
0:26:44 they don’t have this contradiction inside themselves of, “Oh, I’m treating this thing
0:26:46 like a living thing, but it’s just a machine.”
0:26:49 Oh, that’s so fascinating because I would have thought it would be the other way.
0:26:53 If you think everything has a soul, it’s sort of harder to disentangle, but you’re saying
0:26:57 you sort of are desensitized to it in a way.
0:27:02 Or you’re more used to viewing everything as connected but different.
0:27:07 And so, you know, you still face the same design challenges of how do you get people
0:27:12 to treat robots like tools in settings where you don’t want them to get emotionally attached
0:27:13 to them.
0:27:17 So those design challenges still exist, but I think as a society, you’re not also dealing
0:27:21 with this contradiction of, “I want to treat this thing like a machine, but I’m treating
0:27:22 it differently.”
0:27:27 Right, the sort of ethical wrappers around this that we need to be aware of when we’re
0:27:32 starting to introduce these different types of interactions as these relationships become
0:27:33 more sophisticated.
0:27:36 Thank you so much for joining us on the A16Z podcast.
0:27:36 Thanks for having me.

0:00:05 The content here is for informational purposes only, should not be taken as legal business
0:00:10 tax or investment advice or be used to evaluate any investment or security and is not directed
0:00:15 at any investors or potential investors in any A16Z fund. For more details, please see
0:00:21 a16z.com/disclosures. Hi everyone, welcome to the A6Z podcast. I’m
0:00:26 Sonal. Today we have another one of our podcast on the road episodes with guests from New
0:00:33 York City on the topic of crypto, but more broadly on crypto networks as emerging economies.
0:00:38 This conversation goes into the super interesting nuances of structuring these networks to avoid
0:00:43 some of the failings we’ve seen in the monetary and fiscal policies of traditional economies,
0:00:49 including debating how to empower users when it comes to risk, but also how to better distribute
0:00:54 access in terms of who captures value from networks. We then also discovered different
0:00:59 mindsets for the governance of these networks, which are really crypto economic systems,
0:01:04 especially as they evolve and grow more mainstream over time. Joining us to have this conversation,
0:01:09 we have two guests, our friends at Placeholder VC. Chris Berniske, who formerly led ARC Invest
0:01:13 Crypto Efforts, has written a lot about financial modeling influence frameworks for analyzing
0:01:18 crypto and co-wrote a book on crypto assets, and Joel Monegro, who before starting Placeholder
0:01:23 with Chris was an analyst at USV, where he helped develop their early blockchain theses
0:01:28 prior to that, he managed the Dominican Republic’s government office in charge of developing
0:01:32 the country’s national and digital economy technology agenda. I share all that as key
0:01:36 context for the conversation that follows. And last but not least, we have two of our
0:01:41 partners from A6 and Z Crypto, Jesse Walden and Dennis Nazarov, formerly co-founders
0:01:45 of MediaChain, which was acquired by Spotify, Jesse and Dennis interview our guests and
0:01:50 also add their perspectives in the discussion and debate. On that note, the content here
0:01:57 is for informational purposes only and should not be taken as legal business tax or investment
0:02:03 advice or be used to evaluate any investment or security. It is not directed at any investors
0:02:10 or potential investors in any fund. For more details, please also see a6nzcrypto.com/disclosures.
0:02:15 Before the discussion goes into what it takes to design such crypto economic systems at
0:02:20 scale from value capture to risk to governance, they first quickly begin with the fundamental
0:02:26 concept of layers in a stack of protocols and decentralized applications. The first
0:02:31 voice you’ll hear is Joel followed by Chris’s and then Jesse and Dennis joins in later.
0:02:35 You can think about it from an engineering point of view and how different kinds of
0:02:40 software are layered on top of each other. You can think about it also from a more social
0:02:46 point of view. Layer one is more machine work and layer two is more human work. And as you
0:02:51 transition from layer one to layer two, then you end up with lighter weight models because
0:02:55 you don’t have the capital cost of the machines to actually do the work, to store the files,
0:03:00 to mine the transactions. What we see at layer two are more abstract units of work that require
0:03:06 human judgment. And there the cost is harder to model because it’s harder to quantify what
0:03:10 is the value that goes into performing some unit of work. And that unit of work can be
0:03:16 anything from curating content to making a governance decision. All of those things are
0:03:21 very difficult to put in a spreadsheet model, but we have the aid of the invisible hand
0:03:26 in a way and that’s helping us figure out what human work is worth. I think the infatuation
0:03:31 with layer one comes from it being much easier to understand the cost and value relationship
0:03:35 between providing a service and consuming a service. It gets harder as you move up the
0:03:40 layers. Once you get into the realm of human work is when you start to really imagine that
0:03:44 there are different ways in which crypto can really change the way we do things.
0:03:48 I think as we go up higher in these layers we’re going to see different incentivization
0:03:54 and therefore value capture mechanisms. Layer one, the priority is security because that’s
0:03:59 basically our clearing and settlement layer. And so we’ve really seen Bitcoin prioritize
0:04:05 security or Ethereum. And because if we build all these layers on top of an insecure layer
0:04:11 one then we’re screwed, right? And then as we move up the stack it’s less about that
0:04:18 machine security say and more about how do you incentivize the economic actors to perform
0:04:23 the service that you’ve promised to provide. And the only way that that ends up being a
0:04:28 service that the end user uses is either if it’s cheaper than existing services that
0:04:32 you can get from the centralized model but on par in terms of user experience or it’s
0:04:38 a fundamentally new experience or service and this is the only place I can get it there’s
0:04:43 the access tokens where basically there’s demand of what I find interesting about an
0:04:47 access token is historically we’ve thought of tokens as needing to connect the supply
0:04:52 side and demand side. An access token really just focuses on the supply side and it can
0:04:56 have a fixed supply and there can be scarcity because basically if the supply side needs
0:05:01 that token to perform the service and performing the service is a profitable activity for them
0:05:06 then there will be a clamoring to get a hold of that token and that drives its own scarcity
0:05:11 and that’s slightly easier to model you can actually use discounted cash flow model to
0:05:17 approximate the value of that token and the demand side can pay in fiat if they want and
0:05:21 I think we’ll increasingly see this where the demand side is going to just pay in whatever
0:05:26 asset they want they don’t need to interface with crypto assets from a day to day perspective
0:05:31 but you can still have value in a work token that organizes the supply side and induces
0:05:34 a competition around being able to provide that service.
0:05:39 So in the model you described Chris we call it the taxi medallion model what’s interesting
0:05:44 about it is there’s one token that is the right to do the work and the other token which
0:05:49 is the payment token and so as you said it’s you can do sort of a cash flow analysis on
0:05:54 what the work token is going to earn and this differs from the base layer where you know
0:06:00 today most of the layer one blockchains use one token for both rewarding the supply side
0:06:05 and for the demand side to consume the service and I’m curious what do you guys think the
0:06:13 implications of moving to work token model are is there any sort of implications for
0:06:17 users and suppliers not being aligned around the same token.
0:06:23 Well the thing that’s really scary about this trend to create dual token systems where one
0:06:27 token gives you access to the supply side or a right to participate in the supply side
0:06:31 and another one into being the payment token whether it’s another token that it’s created
0:06:36 by the crypto network or something like ETH or any other one is that you were doing the
0:06:42 same thing that we did with the world economy which is that we separated currency and capital
0:06:47 the moment that we moved into a fiat currency model in modern capitalism.
0:06:53 We have two major kind of asset types we have capital and we have that and currencies are
0:06:57 really a form of debt at the end of the day and backtracking a little bit the reason we
0:07:01 separated currency from capital is because the transaction velocity of capital is very
0:07:08 low the transaction velocity of a dollar is very high and so as the economy grows exchanging
0:07:12 gold or land or just a barter system is not going to work and so we needed currencies
0:07:19 to accelerate economic growth so far so good the problem is that capital and currencies
0:07:24 respond very differently to economic growth capital appreciates together with economic
0:07:28 growth and currencies actually depreciate as an economy grows.
0:07:34 Inflation is commonly thought of as the printing of new money but really a more traditional
0:07:40 definition of inflation is increases in prices over time as a result of economic growth and
0:07:45 capital as an asset type appreciates as the economy grows because it is more scarce than
0:07:50 currency currency we print more of it to keep up with inflation to keep prices stable and
0:07:53 that’s how they end up devaluing over time because we’re creating more and more and more
0:07:58 what happens over time then is that actually capital becomes more and more concentrated
0:08:02 because its transaction velocity is so low and the problem with that is that we end up
0:08:07 with a lot of people living their lives in in currency and very few people living their
0:08:08 lives in capital.
0:08:13 What part of the promise of crypto networks at least to me is that we are able to combine
0:08:18 currency and capital into a single asset so that then we don’t get the same kind of income
0:08:24 inequality or wealth inequality being created as any individual crypto network grows and
0:08:29 the risk of separating the access token or the work token from the currency token is
0:08:34 that the people who accumulated the access tokens early on that group becomes increasingly
0:08:39 concentrated over time as the economy grows or as the crypto network grows the value of
0:08:44 combining the two. If you have a single token that is both a supply site token and a payment
0:08:50 token in order for the supply site to provide its service it has to take payment in that
0:08:53 token so the token has to be in the user’s hands in order for them to actually consume
0:08:58 the service and so that creates a pressure to sell the work token or sell the capital
0:09:02 token as the network grows because otherwise you’re not going to get any customers. Everyone
0:09:06 in a crypto network can participate from the value created as opposed to just one segment.
0:09:12 I definitely agree there’s a risk to backtracking a little bit with the ethos of crypto separating
0:09:17 the work or the capital and currency. I think one thing even if we end up in a work token
0:09:23 world or taxi medallion world where we’ve separated capital and currency again at least
0:09:29 within those networks the one reassuring thing is that you can’t be a passive accumulator
0:09:35 of capital. You have to be more active like you stake the work token and then you have
0:09:40 to continue to provide work for the network to continue to collect cash flows.
0:09:43 So I would argue that that’s not entirely true and that’s because in these proof of
0:09:49 stake systems you often have the ability to delegate stake and so what happens there is
0:09:53 a marketplace emerges where there are people sitting on capital and they just point that
0:09:58 capital at a worker so it’s very much like taxi medallions work in New York City. They’re
0:10:02 mostly bought up by hedge funds and the like and then they hire drivers to go and drive
0:10:06 the cars and earn passive income on those medallions. And so I think that does speak
0:10:11 to Jule’s point that the separation of these two things does result in a concentration
0:10:17 of capital but I would argue that the flip side is that the users of the service don’t
0:10:22 have to take any risk. So when I get into a taxicab I don’t have to think about whether
0:10:28 Uber or Lyft are going to have a real big impact on the value of my medallion and the
0:10:33 future utility I’ll get out of it. So humans tend to think in one unit of account because
0:10:37 it’s just easier to reason about and you don’t want to be taking risk necessarily when you’re
0:10:42 buying a coffee or a pizza and that ends up going on to be worth millions of dollars.
0:10:49 And so there is this sort of simplicity in the tax now in versus currency model that
0:10:54 I think it’s hard to get away from just because of human nature.
0:10:59 I argue that risk is good because you can’t create value without creating risk. And again
0:11:04 it’s that same kind of thinking that led to the world where we are today. It’s much simpler
0:11:09 to have everyone use dollars because no one has to think about how the markets are moving
0:11:12 and how your value is changing over time. Rather you trust the government. You trust
0:11:18 that they’re going to maintain a certain monetary policy such that you hope that your
0:11:22 hundred thousand dollars today is going to maintain a certain degree of purchasing power.
0:11:28 The problem with that is that by not allowing people to take risks in a way then we’ve cheated
0:11:33 them away from capturing some of the upside of the value. And I think it’s time to rethink
0:11:39 how we think about risk for people more broadly. If we prevent people or the science systems
0:11:44 that make it harder for people to take those risks and participate in the value then we’re
0:11:48 going to end up in the same world where we are today where the values concentrated amongst
0:11:52 those who took the greatest risks. Now obviously it can go really badly if you have a whole
0:11:55 bunch of risk spread a whole whole bunch of people and then everything comes crashing
0:12:00 down then you have a real problem. So it’s not that there isn’t value instability and
0:12:05 value in currencies that make it easier for consumers to go about their daily lives. But
0:12:09 I also don’t think we’re heading to a world where you buy your coffee with Bitcoin. I
0:12:14 also think that the overall argument of well we need to create these systems because otherwise
0:12:18 it’s too risky and then people are not going to use it. I think it’s a little not patronizing
0:12:23 but it underestimates people’s ability to make those decisions for themselves.
0:12:29 I remember asking you about three four years ago whether it’s actually a good idea to make
0:12:35 your users investors because there’s a cognitive overhead that is associated with that. You’re
0:12:42 asking your users to be sophisticated about the underlying infrastructure powering their
0:12:48 daily experience and I think increasingly in in crypto we’re seeing that at least when
0:12:53 you have some skin in the game participants in these networks are willing to go in a little
0:12:58 bit deeper and the types of communities that emerge are on these projects are just fundamentally
0:13:02 different like people have a different relationship with the network. Ethereum participants own
0:13:08 the token and therefore have an emotional connection to the values of the infrastructure
0:13:13 and its goals and what it achieves. Sometimes. Yeah, to a fault sometimes and I think the
0:13:20 question is how scalable is this? Can we scale this model to the entire world or is it limited
0:13:26 to a subset of users who are sophisticated enough and want to take on that risk and so
0:13:31 just to push back a little bit on the idea that we will be able to scale this model sort
0:13:35 of all the way up. I think the idea of having this work token currency model it doesn’t
0:13:41 preclude power users from participating in the network and then through delegation the
0:13:45 owner of that token does not need to necessarily provide the work. So a power user can invest
0:13:51 in the success of the network that they’re using but it’s a choice and so I think there’s
0:13:56 flexibility both ways. On one side the assumption is that we should give risk to everyone and
0:14:00 that’s that optimal in the other side is we should let people choose. You can argue both
0:14:05 sides of the table and so what that tells me is let the people make the choice but then
0:14:09 that becomes a question of what’s the default. So you know we’ve all gone through a sign-up
0:14:14 form where the checkboxes are checked by default or not that opt you into certain things and
0:14:19 we’ve learned that people tend to stick with the default and so I would rather see a world
0:14:23 where the default is we’re all participating in the value. Yeah, any citizen can go and
0:14:27 open a brokerage account and buy stocks and participate in the market but most people
0:14:32 don’t because that’s not the default. So I think one thing that we can do here is actually
0:14:37 create a model and train another generation of users to think as users who are staked
0:14:41 in the network in a way that we really couldn’t before and the generational aspect I think
0:14:46 is very important because the other thing that’s going on here is that the way these
0:14:51 assets accrue value is very different to the way that other previous kinds of assets accrue
0:14:57 value. Previously in a more traditional world when we had companies go public it was a lot
0:15:02 of the same philosophy of you, you’re a customer of Walmart, Walmart goes public, you can buy
0:15:06 stock in Walmart and then every time that you go to Walmart and you’re paying with your
0:15:12 dollars you know there is some value accruing back to you as a shareholder of the company.
0:15:15 But in order to properly analyze an equity you have to go to business school and learn
0:15:19 how to do this kind of cash flows and how to figure out you know how to analyze whether
0:15:23 management is doing things correctly and so on. What’s different here is that we’re dealing
0:15:28 with these decentralized networks where if they’re properly constructed the value doesn’t
0:15:32 really necessarily depend on the actions of a management team but rather on the overall
0:15:33 network.
0:15:39 Yes, we were talking about the consumer perspective and them being a shareholder and that’s one
0:15:43 side of the argument, the other is from the developer and people who are actually building
0:15:44 these protocols.
0:15:49 So I want to take a step back because we’ve been saying user network participant those
0:15:57 kinds of things and I think that it really depends on which network participant we’re
0:15:58 referring to.
0:16:03 So early on in a network where it’s mostly the supply side right because the supply side
0:16:08 has to come on board to actually provision the asset and presumably the supply side is
0:16:15 going to be a much more sophisticated person at least in their understanding of the network
0:16:16 than the demand side.
0:16:20 At least as we scale out over the long term I think that’s an assumption we can make.
0:16:24 And so that supply side can tolerate that earlier risk because presumably they’re more
0:16:27 sophisticated and so that all makes sense.
0:16:32 Where I think we start to encounter more problems especially if crypto goes mainstream we will
0:16:38 have you know hundreds of millions billions of end users on the demand side and I think
0:16:43 it’s a stretch to ask the demand side to be an expert in the network.
0:16:46 The supply side this is different from our current equity environment.
0:16:52 If you set aside stock based comp traditionally an employee at a company doesn’t necessarily
0:16:54 have exposure to the upside of that company.
0:16:59 Stock based comp has changed that a lot but that employee mostly gets paid in fiat currency
0:17:03 which goes back to Joel’s point of you know they don’t get to participate in the capital
0:17:08 appreciation as much as the management and the concentrated owners of capital.
0:17:11 So for me that’s a big improvement.
0:17:18 Early on demand-siders they’ll be early adopters but I think as this space grows we will abstract
0:17:22 a lot of that complexity away from them and the demand side will just pay in whatever
0:17:27 they want to pay it be they in Kenya be they in the US be they in Korea you pay for the
0:17:32 service with whatever currency you want that ends up getting converted and the supply side
0:17:37 will get paid in the native asset of the network through whatever it may be.
0:17:41 But the key is that the supply side gets access to that risk and that capital appreciation
0:17:42 for me for me.
0:17:47 So I think an interesting sort of avenue to go down here is to again come back to this
0:17:51 idea of the different layers in the stack and how it may be different at each of them.
0:17:56 So at the base layer it makes a lot of sense for there to be sort of one currency or you
0:18:01 know because because you do really want to align the incentives of the supply side and
0:18:07 the users because it’s this very general substrate upon which you know like all all kinds of
0:18:11 more complex applications can be built but this because it’s very general you want there
0:18:17 to be this sort of network effect that everyone is everyone converges on the values and the
0:18:22 goal of this general substrate as you get further up the stack and you build more complex
0:18:28 applications something like a stable coin for example requires sort of more specialization
0:18:32 in terms of the type of work that’s being done and this is to Joel’s point earlier that
0:18:39 it’s it’s more human work and more specialization generally means more expertise.
0:18:45 And so once you get into those types of applications I think it becomes a little bit harder for everyday
0:18:49 users to be sophisticated about the work that’s going on behind the scenes.
0:18:54 People probably don’t need to understand how Ethereum computers you know determine consensus
0:18:59 but I think the difference is that in order for a stable coin to work it’s a much more
0:19:03 you know complicated system and it has parameters that need to be tuned and they probably need
0:19:08 to be tuned by experts whereas computation is deterministic it’s sort of a binary outcome
0:19:11 right or wrong and it can be verified by computers.
0:19:16 And so I would argue that the base layer of the system being more general lends itself
0:19:20 to this sort of single token model a little bit better but as you move further up the
0:19:24 stack I think you do want this separation between sort of the management and the users
0:19:27 because it requires expertise.
0:19:29 And that brings us to governance.
0:19:30 Right.
0:19:34 And maybe an analogy is the base layer sort of like a country it’s like you’re a citizen
0:19:40 of America you get to vote and participate in you know elections and to decide kind of
0:19:41 policy.
0:19:46 Taxation policy is the substrate for all economic activity built on top of America and then
0:19:51 there’s more specific corporations inside of America and they have their own governance
0:19:56 practices they have their own equity and then participating in them is much more specialized
0:20:01 but you have this broad substrate that everyone else builds on top of with different governance
0:20:02 parameters.
0:20:07 So I like to think of or we like to think of crypto networks as emerging economies and
0:20:13 what’s interesting about that is that if you compare crypto network to a country you start
0:20:18 to see a number of similarities there’s a currency that’s exchanged between buyers and
0:20:23 sellers there’s if you think of the executive team or the executive branch is the core development
0:20:28 team and you think of the blockchain as the court or the legislative system where you
0:20:32 know all the rules go there and then you have the supply side which are the miners or the
0:20:36 producers and you have the demand side which are the users who are consuming the service
0:20:39 that starts to look a lot like a small economy.
0:20:44 And then what’s cool about that is that you can use this model to think through whether
0:20:47 a crypto network is properly constructed or not.
0:20:52 So for example things that we’ve learned over time that we like to see in physical economies
0:20:58 like low degrees of corruption and sound monetary policies and fiscal policies and a rich supply
0:21:03 side and an active demand side it ends up getting us into the topic of governance because
0:21:08 one of the things that can determine the success of a national economy or not is how that economy
0:21:09 is governed.
0:21:14 There’s a broader conversation about where is the value of governance in crypto networks
0:21:18 and governance is a difficult topic because it is so broad but I’ll bring it even further
0:21:24 back to the history of information technology in the 50s and 60s.
0:21:27 That era was based around the hardware how quickly you could iterate on hardware and
0:21:31 how quickly could you get computers in the market and IBM won that war because they were
0:21:36 able to design custom computers for custom use cases faster than than anybody else.
0:21:40 That business started breaking down in the 70s and 80s following the introduction of
0:21:45 the microprocessor which consolidated a whole bunch of those circuits into a single part
0:21:49 that was widely available and that created two things.
0:21:53 It first it unbundled the hardware industry and we went from effectively one computer
0:21:58 manufacturer to dozens of PC manufacturers and so on.
0:22:01 But then what happened is that value moved one layer up to the software layer and so
0:22:05 we have Microsoft and the PC software boom of the 70s and 80s.
0:22:09 We saw that get built on top of the microprocessor standard or platform.
0:22:15 Fast forward to the end of the 80s and into the 90s and we went from having dozens of
0:22:22 independent software ventures to having Microsoft consolidating that entire ecosystem and building
0:22:27 its business on the basis of proprietary software and proprietary distribution of that software.
0:22:30 What happened in the 90s is we got two things.
0:22:35 We got the internet and we got Linux which was free software and free distribution and
0:22:38 so that directly challenged Microsoft.
0:22:43 And so we got into the web era where the value moved again one layer up to data which is
0:22:44 where we live today.
0:22:50 We have the big tech companies of today Google, Apple, Facebook, Amazon and so on.
0:22:54 Their main asset their main capital asset is all the data that they’ve been able to
0:22:56 accrue over time as people have used their service.
0:23:01 So we’ve been in this stage for about 20 years now and right on schedule we get the
0:23:07 arrival of a new open technology which are blockchains that directly challenge the proprietary
0:23:11 data business model just as the internet and Linux challenged the proprietary software
0:23:12 and distribution business model.
0:23:16 It gets into a question of okay if we see this pattern of value moving one layer up,
0:23:20 one layer up, one layer up and we’ve gone from hardware to software and from software
0:23:22 to data and now data is free.
0:23:23 What’s above data?
0:23:29 The layer that exists at a higher level and to us that’s governance because it becomes
0:23:33 a question of how do we manage, how do we control, how do we manipulate the data and
0:23:38 how do we agree on a single source of truth which is the whole thinking behind designing
0:23:40 these consensus systems.
0:23:44 The crypto networks at the end of the day are systems that we designed to arrive at a shared
0:23:49 understanding of what is the right data to observe in a world where all the data is open.
0:23:51 And ultimately that is a governance system.
0:23:57 With that model do you think the base layer, the computational substrate, does that become
0:23:58 a commodity?
0:24:02 How do you think about the value of the base layer and how governance of the base layer
0:24:05 relates to the governance of the applications on top of it?
0:24:11 I think we need to be careful of thinking of it statically because just as Joel just
0:24:15 went through there’s this evolution of value capture that we saw with information technology
0:24:22 and I think we will see an evolution of value capture within crypto where value will start
0:24:24 to move up the protocol stack.
0:24:29 Right now we’re focused a lot on developer protocols because we believe it’s the developer
0:24:35 era of crypto and those are the most valuable people and developer attention is the most
0:24:38 valuable resource I would argue within crypto right now.
0:24:44 And we may have this period of value accrual and developer facing networks which then may
0:24:48 become commoditized and shift up to more consumer facing protocols.
0:24:50 But it’s more that there’s this evolution.
0:24:53 So how do you justify that more concretely?
0:24:58 For example, two transactions can use the same amount of gas, they could use the same
0:25:04 amount of the computational resource of Ethereum, but their economic value can be drastically
0:25:05 different.
0:25:09 I mean that brings a question of there’s sort of this tragedy of the commons problem that
0:25:14 one user of the protocol derives way more value from this underlying substrate than
0:25:15 another.
0:25:19 We can compare the foundational layer of a blockchain smart contract platform and decentralized
0:25:23 applications being built on top of it to general cloud computing infrastructure such
0:25:30 as Azure or Google Cloud or Amazon AWS and all the super valuable kind of union or applications
0:25:31 built on top of it.
0:25:37 They all utilize this commodity layer and kind of pay for computation at the level of
0:25:42 the resource, but the value they derive from the cloud platform is immensely higher, hence
0:25:46 their market caps combined are much higher than the cloud platforms underneath them.
0:25:50 So I don’t think every layer one protocol will capture a ton of value.
0:25:53 I think most of them will get commoditized and the ones that don’t will be the ones that
0:25:59 become these stores of value for these really important settlement protocols within the
0:26:00 space.
0:26:05 But I think that as we move up past that, we could see these middleware protocols actually
0:26:09 have more scale than the underlying smart contract protocols.
0:26:13 Start now with the developer in the future with the consumer, but still largely in the
0:26:14 protocol layer.
0:26:18 And I mean, we’re talking about a multi-decadal evolution here, but an evolution nonetheless.
0:26:23 I have a different way of thinking about value, which is through the lens of cost.
0:26:27 And this goes back to Econ 101, one of the first things they teach you is marginal benefit
0:26:30 equals marginal cost at equilibrium.
0:26:36 And bringing that into the discussion of where does value accrue, then you can then extend
0:26:40 that and okay, value will accrue to where there is the highest cost.
0:26:42 And it doesn’t really matter where in the layer that is.
0:26:46 It matters more what is the kind of service that’s being provided and what is the cost
0:26:47 of that service.
0:26:49 And then the other dimension is scale.
0:26:54 I think people sometimes confuse commodity with value less.
0:26:58 You can have something that is a complete commodity like milk and still get enormous amounts of
0:27:03 scale that permit value to accrue at that layer regardless.
0:27:08 As you move up the different layers or you start thinking about where does value accrue
0:27:14 in different places, just bringing back the whole governance umbrella, making a decision.
0:27:15 What’s the value of a decision?
0:27:19 Well, it might be in precisely where they’re implementing it in the governance over the
0:27:24 standard and over the protocol over time, where perhaps over time as the protocol becomes
0:27:30 more important and the standard becomes more widespread, then the value or the cost of
0:27:33 making a protocol decision increases over time because it affects a greater number of
0:27:35 people.
0:27:39 And so that’s one way in which you can kind of use the lens of cost to kind of chase down
0:27:44 where value might accrue in different services or across an ecosystem.
0:27:51 I think one helpful analogy with the governance space and value accrue from the perspective
0:27:57 of cost, if you look at, for example, United States, the United States has a fixed supply
0:28:05 of one president, but the cost of being that president has grown over time as the network
0:28:08 of the United States has also grown over time.
0:28:12 And this is actually where I think fixed supply works in a governance token setting.
0:28:16 If you have a fixed supply of that asset, but the cost and value of governing that network
0:28:21 is going up, then so too should the cost per token of that asset.
0:28:23 And so that’s a useful analogy for thinking through it.
0:28:28 And just thinking about America being a substrate for businesses built on top of it, the incentives
0:28:35 there is that corporations that exist within America are taxed to fund that substrate.
0:28:38 Better infrastructure, all kinds of services.
0:28:42 And to my earlier point, that there is no kind of economic relationship in the same
0:28:46 way that there is in America between Ethereum, for example, and the applications built on
0:28:52 top of it, you could imagine that there is some way in protocol to say it’s an upgrade
0:29:00 to ERC 20 standard, that 10% of the tokens every quarter, it’s a tax that goes to fund
0:29:01 the base chain.
0:29:08 And maybe this is a way to solve the problem of how do we fund innovation of the base chain.
0:29:09 I’m curious what you think of that.
0:29:16 I think traditionally taxation is one of the mechanisms through which a currency can boot
0:29:17 strap value.
0:29:22 And this goes back to, for example, the shardless theory that the government has to spend the
0:29:25 currency first and get it into circulation and then collect taxation.
0:29:28 And so that kick starts the economic flywheel.
0:29:32 People are experimenting with different forms of taxation in the space.
0:29:35 And really a transaction fee is a form of tax.
0:29:43 But we haven’t seen direct taxation to fund core developers beyond the inflationary model.
0:29:47 And that’s really kind of taxation through senior age or dilution.
0:29:52 For example, there’s Zcash or Decred or some of these networks that are working with this
0:29:57 idea of, okay, part of the monetary policy, we’re going to mint out over time and Decred
0:30:03 allocates 10% of each Coinbase reward to the developer pool, which will be allocated through
0:30:05 governance and the community’s decision to write it.
0:30:07 That’s effectively a 10% tax.
0:30:10 And same with Zcash, it’s a 20, 30% tax.
0:30:14 And because it’s of the income, right, if you think of every time a new block is produced,
0:30:15 that’s income that’s going to the miners.
0:30:18 But 10% of that is coming back to development.
0:30:22 It’s an implicit as opposed to an explicit tax that then funds the network.
0:30:27 You know, if you start thinking about governance first through the lens of this idea of taxation,
0:30:31 you can think of your kind of analogy of we have a fixed supply of one precedent and then
0:30:35 we can figure out how much it’s cost to run for precedent over time and figure out, you
0:30:38 know, what’s the cost of that governance model.
0:30:41 But there’s something that we can observe more concretely, which is just a tax rate
0:30:43 over time in world economies.
0:30:49 And we have seen tax rates as a proportion of GDP increase over time, as GDP increases
0:30:54 because the cost of governing the economy grows together with the growth of the economy.
0:30:59 And so how we can translate that into crypto networks is thinking through what is the cost
0:31:04 of governing a crypto network and the cost of maintaining an economic system over time.
0:31:08 And how does that change as that network grows or contracts over time?
0:31:13 So again, we keep coming back to this conversation around different layers in the stack and we
0:31:17 did a whole podcast on this about how the emergence of base layer crypto networks are
0:31:24 like cities in that there’s a bunch of people that have a vested interest in the infrastructure
0:31:27 upon which they’re building and they’re pulling it in different directions.
0:31:30 But there’s actually at the birth of a lot of cities, there’s not this like formal process
0:31:35 for coordinating that it just consensus emerges to this very rough process.
0:31:39 A lot of these ideas around rough consensus and running code are from Venkatesh Rao’s
0:31:43 post on Breaking Smart, which we’ve referenced a number of times on the podcast and we keep
0:31:44 coming back to.
0:31:50 We also talked about the emergence of internet standards and specifically the internet engineering
0:31:55 task force was this loose group of academics and engineers that were working on base layer
0:32:00 internet protocols and they had this very formal policy to not have voting, but instead
0:32:05 sort of weekly coordinated consensus mechanism whereby people argued their points with strong
0:32:06 opinions.
0:32:10 It was sort of a robust and sort of scientific approach and the best ideas were converged
0:32:11 upon.
0:32:17 So it’s this idea that a rough consensus emerges from one strong opinions weekly held to running
0:32:20 code in order to form those opinions.
0:32:24 And that’s very much the governance model of Bitcoin and Ethereum today.
0:32:27 Other projects are taking a different approach with formal on chain governance.
0:32:32 I think our view is that the more general the network, in terms of the service that it’s
0:32:37 providing, the more it lends itself to this process of rough consensus.
0:32:43 If you go all the way down the stack down to IP protocol, it’s this very general protocol.
0:32:48 It’s completely un-opinionated about the packets that it’s moving from A to B and the rough
0:32:53 consensus process or work there, it’s very lightweight and easy to integrate.
0:32:59 I would make the argument that general computation platform would lend itself to that same process
0:33:04 because you want it to do this very general thing and a thing that is very deterministic.
0:33:07 It doesn’t require human subjectivity to validate.
0:33:11 Then as you go further up the stack, there are applications built on top of the substrate
0:33:17 and as Dennis made earlier, it’s like the businesses built on top of America.
0:33:19 Each of them are providing different services.
0:33:23 They require different expertise in order to provide those services.
0:33:27 I would make the argument that at that layer in the stack, governance does become important.
0:33:32 The expertise requires specialization and these applications need to be more dynamic and responsive
0:33:38 to their users versus general computation, which over time should hopefully remain fairly
0:33:43 consistent such that these applications can build on top without the rules changing on
0:33:44 them.
0:33:51 There are many factors that go into governance decision processes that are informal.
0:33:55 One factor is the original roadmap of the project.
0:34:01 Does this change fit within the original vision outlined by both the founders and the community?
0:34:04 In Bitcoin, there is a promise of 21 million Bitcoins.
0:34:05 That is very important to it.
0:34:09 Also, there’s figureheads, Vitalik, for example, has very strong opinions.
0:34:13 While people trust him, he is one of many voices in the community.
0:34:15 I think it’s not dictatorial.
0:34:19 He proposes changes and there’s a debate and a conversation.
0:34:25 The base layer today in a platform like Ethereum is both the medium of exchange and the reward
0:34:28 for the supply side of the network.
0:34:31 Importantly, the governance process is this process of rough consensus where there is
0:34:35 no default setting for upgrading the network.
0:34:41 This is an important connection to draw because the users of the network have an active interest
0:34:44 in how the network evolves.
0:34:47 They are incentivized to participate in this process of rough consensus.
0:34:52 Whereas an application built on top of Ethereum, that layer in the stack, the expertise required
0:34:57 may necessitate that the management of this organization, whether it’s a central company
0:35:02 or a loosely affiliated group of people all over the world, that those stakeholders have
0:35:04 that expertise in order to make those decisions.
0:35:06 I think this is a really critical difference.
0:35:08 The base layer is very general.
0:35:13 Users wanted to do this one thing and do it well, compute things in a deterministic way,
0:35:17 but as you get further up the stack, it becomes a lot harder to reason about those mechanics.
0:35:22 I’ve started to use the term power tokens instead of governance tokens to refer to tokens
0:35:28 that represent the power to change the rules or change the makeup of a crypto network or
0:35:32 at least one vote to change the rules of the crypto network.
0:35:36 One way that I like to describe it is that crypt economics are the rules of the game
0:35:41 and governance is the power to change the rules of the game.
0:35:46 My belief is that as the game becomes more valuable, then the power to change the rules
0:35:48 becomes more valuable as well.
0:35:52 That’s the umbrella that I like to use to think through what’s the value of power, what’s
0:35:54 the value of changing the rules of the game.
0:36:00 If power tokens were the only means of making decisions, it’s sort of a very heavy-handed
0:36:06 specific tool that if manipulated incorrectly, it will lead to negative outcomes.
0:36:13 This more informal process through hard forks, you also, in addition, have checks and balances
0:36:17 where, as you mentioned earlier, there are different classes of participants, the miners,
0:36:22 the developers, the users, different bodies of the government that developers proposed
0:36:23 the code.
0:36:28 It has to be agreed upon by the miners to implement it, so it’s a more multifaceted multi-stakeholder
0:36:33 operation as opposed to who owns the tokens gets to, and the other problem is it creates
0:36:34 a default.
0:36:41 If the network automatically upgrades into some specific version of the code, the catastrophic
0:36:45 scenarios are much worse because everyone opts into a default, whereas in this more
0:36:49 weak consensus model, everyone has to agree in a broader way.
0:36:53 I have a twist on rough consensus and running code, which is crypt economic consensus and
0:36:59 running code, and my take on it is rough consensus works well when you’re small, and when the
0:37:03 number of stakeholders is fairly small, when the IETF was working through these protocol
0:37:09 iterations, you didn’t have to find consensus amongst a very large group of people in terms
0:37:11 of who is really affected by these decisions.
0:37:15 Today, it’s a completely different kind of dynamic where the underlying protocols have
0:37:20 remained fairly stagnant, and all of the innovation in terms of use cases has happened above,
0:37:24 where governance is a lot more fluid in the sense that each individual application can
0:37:28 construct its environment or its system in the way they prefer.
0:37:34 But bringing these ideas back to crypto, there’s another model that I like to think about which
0:37:37 is on-chain governance, off-chain diplomacy.
0:37:42 So by off-chain diplomacy, I mean every time that a core developer team meets with each
0:37:47 other to make a decision, or a user or an application of that protocol wants to lobby
0:37:48 for a change.
0:37:54 The human process by which we arrive at proposals and ultimately decisions, you still have meetings
0:38:00 between people debating and arriving at decisions and proposing different ideas.
0:38:01 All sorts of issues can emerge, right?
0:38:06 You can have concentration in the power token that enables a small group of people to really
0:38:10 control the network, but you can have the same kind of dynamic today with rough consensus.
0:38:14 If you don’t have a formal process that allows everyone that they can participate in that
0:38:18 process, you can end up with a clique that effectively governs the network through their
0:38:21 own rough consensus and cuts out everyone else.
0:38:26 Let’s make sure everyone in the network has a fundamental right to participate in that
0:38:27 governance process.
0:38:32 And let’s have mechanisms for people to get together and have sophisticated discussions
0:38:33 about how decisions were made.
0:38:37 But let’s make them convince the community and convince the network that that is the
0:38:38 right thing to do.
0:38:44 I think also, if you don’t have formal governance mechanisms and clarity around it, you devolve
0:38:47 into governance by defection.
0:38:48 And we saw that with Bitcoin.
0:38:51 We’re not necessarily arguing for complicated governance.
0:38:59 We’re arguing for rules and transparency such that the network participants all understand
0:39:02 not only what the rules are, but how the rules are changed.
0:39:07 Yeah, I guess where we might diverge a bit is that we are arguing that the more general
0:39:11 the service of a crypto network, the more it should be ossified.
0:39:17 So again, coming back to this IP is ossified, Bitcoin has ossified, maybe a general computation
0:39:23 substrate is better ossified because it lends itself to the trust that the developers building
0:39:27 on top need in order to feel comfortable building there.
0:39:31 However, those applications that developers build because they’re complex and dynamic
0:39:35 and interfacing with end users need to be able to change.
0:39:41 And in order for them to do so, a formal governance process is probably necessary.
0:39:45 The default is to take whatever the coin holders vote upon and upgrade the system.
0:39:50 And so I think one of the assumptions is that the token holders participating in the vote
0:39:56 and prior to that participating in the diplomacy are experts and have the best interests of
0:39:58 their users in mind.
0:40:01 Their interests are hopefully aligned with their end users.
0:40:03 And they pay for it if they mess up through dilution.
0:40:04 Right.
0:40:05 That’s right.
0:40:09 And so the incentives are tightly coupled so that the experts make decisions for the
0:40:12 benefit of their users and for the benefit of themselves.
0:40:15 And you end up with this very dynamic system where the end users don’t necessarily need
0:40:19 to think about the complexity of the underlying mechanic of the thing.
0:40:24 And importantly, there’s explicit enforceability over one canonical group of contracts that
0:40:26 control the system.
0:40:32 And so the output of the governance process doesn’t require, for example, end users or
0:40:36 other participants in the network to download new software and run it.
0:40:41 The analogy of a blockchain is we’re traveling down a highway.
0:40:47 And then we decide that we want to take a turn, one of two turns into two different futures.
0:40:52 And that will be upgrading– miners upgrade their software to a new fork version.
0:40:57 What forks are in protocols is different than forks are on base layers.
0:41:02 You can’t fork in the case of a stable coin because it has all this collateral.
0:41:06 You can’t fork the collateral because your DAP doesn’t control that.
0:41:09 You are using the existing platforms collateral.
0:41:10 So the upgrade process has to be different.
0:41:14 In the governance debate, there’s nuance in what it means to upgrade a DAP that’s running
0:41:16 on top of ProCo and the protocol itself.
0:41:19 You brought up the example of a computation network.
0:41:22 It’s a very well-defined kind of service.
0:41:24 A computer runs the code, and the code has an output.
0:41:28 If you focus on something more controversial, like how value is distributed, then all of
0:41:30 a sudden you find that the governance is really important.
0:41:37 And I’d also argue that power token voting excludes the needs of miners and the needs
0:41:38 of users.
0:41:39 That depends on how you design it.
0:41:43 If you think of power as a source of value, then you want to make sure that that power
0:41:46 is evenly distributed amongst the participants of the network.
0:41:49 Otherwise, you end up, again, where we are in the modern economy today with very few
0:41:53 people with a lot of power, and most people with not that much power.
0:41:58 So just one thing I wanted to add is that one of the problems with thinking about on-chain
0:42:03 governance as a way for users to actively participate in the evolution of a network
0:42:07 is that in practice, if we look at the real world and how governance systems work, there’s
0:42:09 actually very low participation.
0:42:16 This is because as an individual user, an individual voter, my one vote, my little say doesn’t
0:42:19 have all that much influence on its own.
0:42:22 And so there’s this apathy about participation.
0:42:28 And so a assumption is that when we talk about users affecting or risking together and participating
0:42:33 in governance, is that they have some sort of emotional stake in the outcome of governance
0:42:34 processes.
0:42:41 And that may be the case in very niche applications where users feel a strong affinity to the application
0:42:42 they’re using.
0:42:46 If I don’t have a strong connection to participating in voting, maybe I’d be more likely to sell
0:42:48 my vote to someone who does.
0:42:53 And the result of that is the same risk that Joel described earlier, where the capital
0:42:57 and the currency become disaggregated.
0:42:59 And this could undermine on-chain governance processes.
0:43:03 It’s important to note that on-chain governance is really hard for a number of reasons.
0:43:07 In the context of blockchains, it’s very difficult to know who the participants are.
0:43:12 You can be hundreds of people just by generating multiple keys.
0:43:17 And so when you have this system that is synonymous, it’s difficult to enforce behavior patterns
0:43:22 that lead to good governance, say, in shareholder governance where shareholders are bound by
0:43:23 fiduciary law.
0:43:28 There’s a great post by a researcher at Cornell who we recently had on the podcast named
0:43:32 Phil Diane, and he went deep down the rabbit hole into different attacks that you could
0:43:33 launch.
0:43:34 The rise of dark doubts?
0:43:39 Yeah, the post dives in deep and our podcast does as well on different sort of attacks
0:43:43 that you can launch to bride participants in on-chain governance schemes.
0:43:49 I think a lot of arguments against on-chain governance are kind of primitive in their
0:43:52 thinking around how is that governance applied.
0:43:57 One common element of a lot of the counterarguments is the assumption that governance power is
0:44:02 linear and one token equals one vote, whereas in crypto networks, we have the opportunity
0:44:05 to design much more intricate systems with much more intricate rules.
0:44:09 And so, yes, you can replicate yourself across a thousand different addresses, while you
0:44:15 can make it such that your governance power is amplified if you have all your tokens
0:44:20 in one wallet and so you can make it actually more powerful to basically voluntarily disclose
0:44:24 still within the pseudonymous system how much of the token you have because you get more
0:44:30 power by aggregating your assets together and you can kind of change how the shape of
0:44:35 that curve, depending on the context, you can also do things like look at the age of
0:44:36 an address.
0:44:40 One thing that we learned actually from a non-crypto company that was in the online
0:44:44 community abuse space is that the length an account has been open is the greatest determinant
0:44:45 of whether it’s a troll or not.
0:44:50 And so you can use things like the life span of an account or of a wallet or the tokens
0:44:51 that it received.
0:44:55 And you can even make distinctions around whether the tokens in that wallet were purchased
0:44:58 from an exchange or were received directly through mining.
0:45:02 And you can factor all of those things into how you design, let’s call it your governance
0:45:03 curve.
0:45:07 Maybe those are all examples of on-chain reputation that’s sort of native to the system.
0:45:08 Exactly.
0:45:10 So yeah, thank you guys very much for coming on.
0:45:14 It’s been awesome sort of recapping the space and how it’s evolved over the last few years
0:45:20 and super excited to see how our theses play out over the years going forward.
0:45:21 Well thanks for having us.

0:00:03 – The content here is for informational purposes only,
0:00:05 should not be taken as legal business tax
0:00:07 or investment advice or be used to evaluate
0:00:10 any investment or security and is not directed
0:00:14 at any investors or potential investors in any A16Z fund.
0:00:18 For more details, please see a16z.com/disclosures.
0:00:22 – Hi everyone, welcome to the A16Z podcast, I’m Sonal.
0:00:25 Today’s episode is all about blockchain-based voting systems
0:00:29 which has implications for crypto economic security
0:00:32 and for governance, especially when you think
0:00:34 about the differences, both good and bad,
0:00:36 between real world and online systems
0:00:38 for coordinating groups of people to vote on something,
0:00:40 whether it’s a decision in a boardroom
0:00:42 or an election or anything else.
0:00:44 This episode was recorded as part
0:00:46 of our New York City podcast road show
0:00:49 and so it features Phil Dayan, a PhD at Cornell Tech,
0:00:51 working with Ari Jules there.
0:00:53 His research focuses on broad questions
0:00:55 of security of distributed systems,
0:00:56 specifically blockchains.
0:00:58 He also wrote a post last year
0:01:00 with Tyler Kell, Ian Mears and Ari Jules
0:01:05 on quote, “On-chain vote buying and the rise of dark DAOs.”
0:01:06 Joining Phil in this hallway style jam
0:01:09 to discuss these topics is Ali Yaya,
0:01:10 who was previously a software engineer
0:01:13 and machine learning researcher at Google X and Google Brain.
0:01:16 He also gave a talk at A16Z Summit on crypto
0:01:18 and the evolution of trust, which you can find
0:01:22 on our website, and he’s a partner on A16Z Crypto.
0:01:24 Speaking of, please note that the content here
0:01:26 is for informational purposes only,
0:01:29 should not be taken as legal business tax
0:01:32 or investment advice, or be used to evaluate
0:01:34 any investment or security and is not directed
0:01:38 at any investors or potential investors in any fund.
0:01:39 For more details, please also see
0:01:42 A16ZCrypto.com/disclosures.
0:01:44 The conversation that follows covers ways
0:01:46 in which blockchain systems are different
0:01:48 from real-world voting systems,
0:01:50 ways the system can be gamed
0:01:52 and what that means for security,
0:01:54 as well as possible solutions
0:01:56 and more importantly, questions
0:01:58 all blockchain system designers should think about
0:02:01 instead of making naive assumptions.
0:02:04 But first, Phil and Ali began by very briefly summing up
0:02:06 the issues in real-world elections
0:02:08 and electronic voting systems.
0:02:11 The first voice you’ll hear is Phil’s, followed by Ali’s.
0:02:14 – So one challenge people have seen is straight up hacking.
0:02:17 Of course, if there is electronic voting in use,
0:02:19 just tampering with the integrity of the election itself
0:02:21 or the integrity of the registration.
0:02:23 Another challenge that people have been worried about
0:02:26 in the past is vote buying and selling.
0:02:27 So if I want you to vote a certain way,
0:02:30 maybe I directly bribe you to do so
0:02:32 or maybe even in the current system,
0:02:34 I can indirectly do it.
0:02:37 But it’s very difficult to bribe someone in person
0:02:39 and sort of understand how they’re going
0:02:40 to act in an election.
0:02:42 – Yeah, you have this great example of how
0:02:44 if the price of a vote is a beer
0:02:45 and you take me out for a beer and say,
0:02:48 “Ali, I want you to vote for ex-candidate.”
0:02:50 I could drink your beer and then go to the poll
0:02:53 and submit whichever ballot I want.
0:02:56 And you have no real mechanism to enforce my vote
0:02:58 in one way or another.
0:03:01 And you then point out how this is not so much the case
0:03:03 when you go to the world of electronic voting.
0:03:04 – Yes, the price of the vote as a beer
0:03:06 is actually kind of realistic.
0:03:09 Like vote buying in general is empirically pretty cheap
0:03:10 for two reasons.
0:03:11 Number one, it’s actually the poorest
0:03:13 and least advantaged people that are the most inclined
0:03:15 to sell their votes.
0:03:17 And number two is most people are disinterested
0:03:18 in most elections.
0:03:20 So this actually makes vote buying pretty cheap.
0:03:22 And in electronic voting, this is a big problem
0:03:25 because with many electronic voting protocols,
0:03:26 you can actually tell at the end of the protocol
0:03:27 how someone voted.
0:03:30 So it becomes much easier for me to bribe you
0:03:31 because I can just say essentially I’ll give you a beer
0:03:34 if I check afterwards and you voted with my candidate
0:03:36 rather than sort of trusting you to go in the polling booth
0:03:38 and make the right decision where socially
0:03:40 I can’t follow you into that booth
0:03:41 and look over your shoulder.
0:03:42 – Exactly.
0:03:44 You point out how in the world of human voting,
0:03:47 there are three things that tend to make vote buying
0:03:48 a little bit more difficult.
0:03:49 And it’s the inefficiencies of the human world
0:03:51 that actually work to your advantage here.
0:03:53 So the first is that in the human world,
0:03:55 it’s a crime to buy votes and that itself
0:03:57 kind of can serve as a deterrent,
0:03:59 which doesn’t really exist so much
0:04:01 in the jurisdiction-less crypto world.
0:04:03 The second one was that ballots
0:04:04 tend to be casted in secrecy.
0:04:06 So there’s no way of me to produce a proof
0:04:08 that I voted in one way or another,
0:04:11 which makes the buying of the vote difficult to enforce.
0:04:13 And the third one you mentioned is that if you tell me
0:04:15 that you’re going to pay me in the future
0:04:16 for voting one direction or another,
0:04:18 I have a hard time trusting you
0:04:20 that you will actually in the end pay me.
0:04:22 And so there’s sort of counterparty risk.
0:04:25 And so in the same way that sort of blockchains
0:04:30 mitigate trust and improve coordination for good purposes,
0:04:34 they can also be used to improve coordination
0:04:36 for sort of malicious purposes.
0:04:38 In this case, vote buying is like a double-edged sword.
0:04:41 Blockchains can be used to increase the efficiency
0:04:44 and effectiveness of bribery and vote buying.
0:04:45 Yes.
0:04:46 In the traditional world,
0:04:48 there’s been a long line of academic research.
0:04:51 So very early on people said we want to vote electronically.
0:04:52 It’ll make tallying cheaper.
0:04:54 It can maybe use cryptography
0:04:55 to increase the integrity of our elections.
0:04:57 So we don’t rely on these pieces of paper
0:05:00 sort of with this weird chain of human custody
0:05:01 and things like that.
0:05:03 But early schemes sort of suffered from this receipt property
0:05:06 where I could produce a proof that like here is the outcome
0:05:09 and here is what I actually voted to lead to this outcome.
0:05:11 So there was a wide range of work early on
0:05:12 on how to sort of solve this issue
0:05:15 and create voting schemes that are receipt free,
0:05:17 which means that after the fact,
0:05:19 I cannot produce a receipt or a proof
0:05:20 to tell you which way I voted.
0:05:22 And it’s sort of equally likely from your perspective
0:05:24 that I voted in any direction.
0:05:27 Later work sort of said that this is not strong enough.
0:05:29 Essentially the high level is
0:05:31 if you’re looking over my shoulder electronically,
0:05:33 like you have a virus on my computer
0:05:35 or you’re just physically looking over my shoulder,
0:05:36 at the time that I’m voting,
0:05:38 even receipt freedom is not enough
0:05:40 because you might be able to see in real time
0:05:42 the direction in which I’m voting
0:05:43 and enforce my vote that way.
0:05:45 So that led to an even stronger property
0:05:46 called coercion resistance,
0:05:48 which is that even if you compromise me
0:05:49 for some period of time,
0:05:52 you still are not able to get me to vote a certain way
0:05:54 in a way that you can trust.
0:05:55 – Yeah, that’s very interesting.
0:05:59 And so let’s connect this to sort of the blockchain world.
0:06:01 These questions of electronic voting have existed
0:06:03 for decades and predate the world
0:06:05 of blockchains and crypto networks.
0:06:07 But now there’s like a resurgence of research
0:06:10 in this direction because so many blockchain
0:06:11 and crypto network projects
0:06:14 want to use on-chain voting for all sorts of purposes.
0:06:16 – So I mean, in blockchain networks in general,
0:06:17 you often need to make decisions.
0:06:20 That’s like part of the attractive point of blockchains
0:06:22 that it makes coordinating group decisions
0:06:24 among actors who don’t trust each other
0:06:26 a little bit easier.
0:06:29 And to make these decisions sort of a natural response
0:06:30 is just vote, right?
0:06:31 That’s something you see in the real world.
0:06:34 It’s something you see in corporations with stockholders.
0:06:35 It’s something you see in boardrooms.
0:06:37 It’s something you see in political elections
0:06:39 and all sorts of other social systems.
0:06:41 So it’s just, I think, a natural human tendency
0:06:43 when asking sort of how to organize these things
0:06:47 that voting is the only real clear shelling point answer
0:06:48 that we can come up with.
0:06:49 So I think an important distinction
0:06:52 on why this stuff really matters in the blockchain world
0:06:54 is that the blockchain world and the real world
0:06:56 don’t operate in the same models.
0:06:57 If you’re going to a boardroom with someone,
0:06:59 you’re sitting next to the person, right?
0:07:02 We’re sort of operating in this model of social honesty
0:07:04 where people can see each other face to face.
0:07:05 You have shared interests in the company.
0:07:08 You sort of know their history at least somewhat.
0:07:10 Whereas in blockchains, you’re operating in an economic,
0:07:13 sort of an economically rational game theoretic model.
0:07:15 So you need much stronger guarantees from your systems.
0:07:17 Your systems need to be strong
0:07:20 even in the presence of economically motivated adversaries.
0:07:21 And they need to be secure,
0:07:24 assuming people are rational rather than honest.
0:07:25 So we don’t get to lean on this sort of honesty
0:07:28 that we have in the real world in blockchains.
0:07:29 And I think that’s where a lot of the mechanisms
0:07:32 that people try to sort of port over naively break down.
0:07:34 – Right, and this is especially important
0:07:36 because in most of the crypto networks
0:07:38 that are actually interesting,
0:07:42 the model is one where anyone can participate.
0:07:44 And people refer to this as the permissionless setting
0:07:46 and that anyone can connect to the network.
0:07:49 Anyone can sort of participate in the decisions
0:07:51 that are made through the governance processes
0:07:53 of the crypto network,
0:07:55 which makes the environment the very hostile one
0:07:57 because anyone anywhere can opt to participate
0:07:59 and they have an economic incentive to do so
0:08:01 because if they can game the system
0:08:03 or if they can sort of subvert it in some way,
0:08:05 then they could potentially profit.
0:08:06 – Exactly.
0:08:08 Satoshi released his white paper in ’09
0:08:10 and academics first started looking at Bitcoin
0:08:12 and its success and its rise
0:08:14 and asking like what is actually the interesting lesson
0:08:16 to be learned here from what we’ve been doing
0:08:17 for the last 20 years.
0:08:19 There was a whole space of consensus protocols
0:08:21 and Byzantine fault-tolerant protocols
0:08:22 that came to consensus on something
0:08:25 even in the presence of malicious users.
0:08:27 But what was really new about Bitcoin
0:08:30 is that it let anyone join and leave the network at any time.
0:08:32 And these people didn’t need to ask the people
0:08:35 who are already participating in the network
0:08:36 whether they can join or not.
0:08:38 So in most consensus protocols,
0:08:40 you have a sort of quorum that’s coming to decisions
0:08:41 and if you want to join,
0:08:43 you need to ask the quorum to join
0:08:45 because the quorum needs to agree on who’s in the quorum.
0:08:46 So they need to sort of come to consensus
0:08:48 on the fact that you’re allowed to join.
0:08:50 Whereas in something like Bitcoin,
0:08:51 if you want to start mining Bitcoin,
0:08:53 you just turn on your rig and as soon as you succeed,
0:08:55 people will accept that mathematically.
0:08:57 They don’t need any sort of membership proof
0:08:58 or anything like that.
0:08:59 What I think is relevant to voting
0:09:01 is that fundamental to the permissionless model
0:09:03 if you’re gonna use cryptography,
0:09:04 which all blockchains do,
0:09:06 is that if I can join and leave at any time,
0:09:08 I need to be able to like generate my own key
0:09:09 and join at any time.
0:09:10 – Right.
0:09:13 I mean, the uses of on-chain voting,
0:09:16 we’re voting within blockchain projects,
0:09:19 range all the way from setting the parameters,
0:09:21 like some parameter in the protocol
0:09:22 that may be something minor,
0:09:25 kind of like the price of gas, for example,
0:09:28 all the way over to sort of some intermediate level
0:09:30 where people use governance
0:09:32 and voting to decide how to allocate funds.
0:09:34 And then this goes all of the way over
0:09:37 to actually deciding how to change the protocol itself.
0:09:40 And so there are projects that are sort of self-amending
0:09:45 and that they use governance as a way of proposing updates
0:09:47 to the protocol and then deciding on which updates
0:09:50 should go through and which updates should not.
0:09:51 And so the stakes are high
0:09:55 and that if you have a governance system that can be gamed,
0:09:59 then all of these use cases may end up being vulnerable
0:10:00 to that kind of attack.
0:10:02 One way of thinking of governance that I quite like
0:10:04 that I think was proposed by Vitalik
0:10:07 is the coordination model of governance
0:10:10 and that really all governance decisions are in essence
0:10:13 a way of coordinating collective action.
0:10:16 He talks about how there are multiple layers
0:10:17 to governance, right?
0:10:19 The bottom layer is like what’s closest
0:10:20 to the real and physical world.
0:10:22 – Yeah, so maybe let’s go bottom up
0:10:24 on everywhere you have voting in blockchains.
0:10:25 At the very base level,
0:10:27 all consensus mechanisms are a vote,
0:10:29 so proof of work itself is a form of voting
0:10:30 on which block is valid
0:10:32 and which history is accepted by the network.
0:10:34 So you have voting at that layer.
0:10:35 Then that half layer up, like you said,
0:10:38 is this governance layer of how do blockchains
0:10:39 actually change their underlying code
0:10:42 and respond to attacks or new situations
0:10:45 or new technology or whatever it may be.
0:10:48 Traditionally, this has sort of gone with the fork model
0:10:50 where you just sort of spin up new code
0:10:52 and try to lobby everyone to just run this new system
0:10:53 instead of the old one.
0:10:55 This model has seen a lot of political strife,
0:10:59 a lot of inefficiency, a lot of sort of lobbying
0:11:01 and traditional politics like nastiness
0:11:03 in the blockchain space.
0:11:06 You can look at the Bitcoin block size debate,
0:11:07 whether to change the one to a two,
0:11:11 which spawned like a year long rift between the communities
0:11:13 that ended up in like several summits and agreements
0:11:16 and eventually a permanent split.
0:11:17 So some people look at that and say,
0:11:18 maybe we can make this more efficient
0:11:20 by just using voting and allowing the coin holders
0:11:22 to express their preference
0:11:23 and sort of just going with that.
0:11:25 And then another layer up from that,
0:11:27 you have the application layers like you were saying.
0:11:30 So these are your DAOs, these are your smart contracts
0:11:32 that wanna use voting to make decisions.
0:11:35 They could be, for example, on how to allocate funds.
0:11:37 They could be on how to change parameters
0:11:38 within their own smart contract.
0:11:41 So you really have voting throughout the blockchain stack.
0:11:42 A lot of projects are using it
0:11:46 and it has a very sort of wide impact as a general problem.
0:11:49 – So one observation that comes out of all of this
0:11:52 is that today’s governance systems
0:11:54 and sort of blockchains and crypto networks,
0:11:56 the way that they exist today will likely devolve
0:11:59 into plutocracy simply because the mechanisms
0:12:02 for vote buying are so effective as you described.
0:12:05 And some proponents of on-chain governance
0:12:07 will argue that plutocracy may not actually
0:12:08 be that bad of a thing.
0:12:10 They may be a bad thing for democracies,
0:12:12 but not so much for blockchains.
0:12:15 In the blockchain world for a crypto network,
0:12:17 it’s not so much a bad thing
0:12:20 because it’s in a sense incentive compatible,
0:12:21 at least at a surface level.
0:12:24 If they are voting using their coins
0:12:26 for any one upgrade to the protocol,
0:12:29 they will want to vote in the interest of other people
0:12:31 who also hold the coins in the interest of the network
0:12:34 because they own it and they have a stake in it.
0:12:37 And also their incentive to protect the network
0:12:38 is proportional to how many coins they own.
0:12:41 So like larger voters or stakeholders
0:12:43 who have more coins in the network
0:12:46 have an even greater incentive to protect the network.
0:12:48 What are your thoughts there?
0:12:49 – So I think every blockchain project
0:12:51 should take a step back and ask,
0:12:52 do we want plutocracy?
0:12:53 Do we want vote buying in our system?
0:12:55 And what are the consequences of that?
0:12:56 For many of them,
0:12:58 maybe it’s more acceptable than for others.
0:13:02 For example, if you have like a small closed sort of contract
0:13:03 that has a few shareholders,
0:13:05 something like an investment firm
0:13:06 and you have like one guy
0:13:08 who decides whether people get in or not,
0:13:10 maybe you’re not so concerned about vote buying
0:13:11 in that kind of a scheme.
0:13:15 Or if you have even like some sort of closed setting
0:13:17 where you can say things about the participants,
0:13:19 maybe you’re not so concerned about vote buying.
0:13:21 In a wider system where let’s say
0:13:23 the whole world is participating in it eventually,
0:13:25 I think the fundamental point is that
0:13:27 most people are disinterested in most votes
0:13:29 and the utility they get from the system
0:13:31 is not directly sort of correlated
0:13:34 with whether they vote A or B on this given issue.
0:13:36 Nonetheless, there are certain groups of people
0:13:37 who are extremely interested
0:13:39 in whether people vote A or B on a certain issue
0:13:41 and these are often pretty moneyed groups.
0:13:42 So in this way,
0:13:44 that kind of governance does sort of degenerate
0:13:45 into plutocracy.
0:13:48 And if that’s acceptable for your system, that’s fine.
0:13:49 I think for many systems, it’s not.
0:13:51 You need to care about these attacks
0:13:52 and you need to reason about
0:13:54 why your system is secure against this
0:13:56 and why your system actually doesn’t degenerate
0:13:57 to plutocracy.
0:13:58 People have tried to get around this
0:14:00 in two ways in blockchains.
0:14:02 The first one is they add some sort of identity.
0:14:03 So they have a third party service
0:14:05 that like you send your cell phone number
0:14:07 or something like that and it sends you a text
0:14:09 and sort of anti-sibles you that way
0:14:11 and then you’re able to participate in a vote.
0:14:15 So at least you can sort of attach some entity
0:14:17 to the person and then count votes per entity
0:14:18 rather than per coin.
0:14:21 This actually still degenerates into plutocracy
0:14:22 because of the way the Dark Dow works
0:14:25 because as long as these identities are keys
0:14:27 that people can sort of generate at any time,
0:14:30 they can be bought and sold and using the Dark Dow model
0:14:31 and you can essentially sell people
0:14:33 like the right to your identity
0:14:35 or you can sell people the right to a certain vote
0:14:38 using your identity or even more specific things than that.
0:14:40 So that kind of doesn’t work
0:14:42 unless you have a strong social protection
0:14:44 where like the person has to come in very often
0:14:46 and the network sort of authenticates
0:14:49 that they’re human or something like that.
0:14:50 That becomes very complicated
0:14:52 and steps much more into the messy world
0:14:53 of real world elections
0:14:56 and maybe doesn’t work for a global blockchain community.
0:14:58 Another way people have tried to get around it
0:15:00 which also kind of requires identity
0:15:03 is this new line of work by Vitalik Glenn Whale
0:15:05 and a few other people which is quadratic voting
0:15:07 where you actually allow vote buying.
0:15:09 So you allow people to buy votes
0:15:11 but only at an exponentially increasing price.
0:15:13 And this may kind of look like plutocracy
0:15:15 because you’re allowing people to buy votes
0:15:17 but if you actually do the math on the incentives,
0:15:20 it turns out that through this increasing function,
0:15:21 essentially people will express
0:15:22 their true preferences in the end.
0:15:26 And one rich person who really cares about A versus B
0:15:29 won’t be able to sort of overwhelm a disinterested majority
0:15:30 that weakly prefers A
0:15:32 and maybe each don’t have as many funds
0:15:33 as that one individual.
0:15:35 So this fixes some known pathologies
0:15:37 in real world voting systems
0:15:39 and also blockchain voting systems.
0:15:41 But it does require identity
0:15:43 and it’s extremely vulnerable to manipulation.
0:15:45 If this one rich person can pretend
0:15:47 that they’re two rich people or something like that,
0:15:48 the gig is sort of up.
0:15:51 And that’s what these new coordination mechanisms allow.
0:15:53 – Yes, I think this dependence on identity
0:15:55 that you are pointing out is very important
0:15:57 because as you pointed out,
0:15:59 anyone can pretend to be more than one person.
0:16:02 They can generate 10 different sets of key pairs
0:16:03 or hundreds of sets of key pairs
0:16:05 and pretend to be hundreds of people.
0:16:06 – Yeah, and the only thing you can do
0:16:07 is wait by coins basically.
0:16:08 – Exactly.
0:16:10 In that world, you end up with unfair representation
0:16:13 of you’re trying to assign a single vote to a key pair.
0:16:17 So proponents of on-chain coin holder governance
0:16:21 which means that one coin gives you one vote will argue.
0:16:23 It’s at the very least, civil resistant,
0:16:26 which means that if you have like 10 million coins
0:16:28 staked on one particular vote,
0:16:31 they’re basically used to vote for one particular outcome.
0:16:34 It’s very hard to argue that those 10 million coins
0:16:37 come from trolls that are trying to sway the election
0:16:39 because there’s real weight and real capital
0:16:41 that’s staked in one direction or another.
0:16:43 Whereas if you’re not using coin voting,
0:16:45 then that becomes more possible.
0:16:48 And so if you have a mechanism for identity
0:16:52 wherein you securely associate one human to one vote
0:16:53 or something like that,
0:16:57 then more sophisticated voting schemes become possible.
0:17:00 I think today, because we lack that kind of a mechanism,
0:17:03 people end up gravitating towards this simple
0:17:07 and somewhat, perhaps somewhat naive one coin, one vote model
0:17:09 which is vulnerable to this foot buying attack.
0:17:13 – Yeah, and this opens up a range of other issues.
0:17:15 So one problem that people have
0:17:16 when they analyze blockchain systems
0:17:18 and they sort of design these mechanisms
0:17:19 is that they look at their mechanism
0:17:21 and reason about its security properties,
0:17:23 but they do that in isolation.
0:17:25 And an important point is that none of these systems
0:17:27 really exist in a vacuum, right?
0:17:29 So take a look at any sort of blockchain
0:17:31 that uses coinholder voting to decide
0:17:33 the outcome of its consensus rules.
0:17:35 And there’s at least two such blockchains
0:17:37 that are sort of using this model.
0:17:39 If these two very large projects are approximately
0:17:41 the same size or one is a little bit bigger
0:17:43 than the other one or one is twice as big
0:17:45 as the other one or something like that,
0:17:47 it’s in the economic interests of everyone
0:17:49 who holds coins in the bigger project
0:17:51 to buy up coins on the smaller project
0:17:53 and influence votes in ways
0:17:55 that are sort of counter competitive.
0:17:57 And maybe even if they can’t buy up
0:17:58 enough of a blocked influence votes,
0:18:02 they can sow chaos and confusion and things like that.
0:18:05 So while one of these systems, you may say in isolation,
0:18:07 like, okay, the coinholders interests are represented
0:18:09 by this plutocracy, that doesn’t really work
0:18:11 when you have a whole world around it
0:18:13 that’s full of money that can frictionlessly
0:18:15 enter and exit the system at any time.
0:18:17 There’s no guarantee whatsoever
0:18:19 that the people who are economically in right this second
0:18:21 have an interest in that system,
0:18:22 especially when there are much bigger systems
0:18:24 that are competing with it.
0:18:25 So I think that’s a very important point
0:18:26 that people overlook.
0:18:28 And again, we mentioned that there’s this sort of stack
0:18:30 of voting, even at the consensus layer,
0:18:32 that has implications on the whole stack.
0:18:34 So if you have a fork that’s like 10%
0:18:36 of the size of a project,
0:18:38 and this fork could potentially impact
0:18:40 the price of the larger project,
0:18:42 it’s absolutely in the interest of that larger project
0:18:45 to launch attacks on that base layer proof of workflow
0:18:46 and do things like censorship,
0:18:48 use some small percentage of their hash power
0:18:51 to do 51% attacks or denial of service
0:18:52 or whatever they need to do to make sure
0:18:54 that that network goes down in price.
0:18:56 And that attack might even be profitable,
0:18:58 especially if there are mechanisms to short
0:18:59 that sort of smaller project.
0:19:00 – Yeah, that’s a very good point.
0:19:03 I think most proponents of a coinholder voting
0:19:07 would argue that it is just not in your interest
0:19:10 to sell your vote because you’d be damaging
0:19:11 the value of the asset that you hold.
0:19:14 And you hold a coin, and if you sell the votes
0:19:15 associated with that coin,
0:19:17 and that might reduce the value of the coin
0:19:20 in some way that sort of results in a net loss for you.
0:19:22 But that analysis happens entirely in a vacuum.
0:19:26 It happens sort of assuming that there aren’t any kind
0:19:28 of external mechanisms via which you could profit
0:19:31 from the loss of value of this particular coin.
0:19:32 Like for example, what you’re mentioning,
0:19:33 competition between blockchains.
0:19:36 If I’m a stakeholder, a much larger stakeholder
0:19:37 in a competing network,
0:19:39 then I might have a strong interest
0:19:41 in reducing the value of this particular coin,
0:19:42 and that that’s associated
0:19:44 with this one competing crypto network
0:19:46 because it may result in a larger profit
0:19:48 outside of the system.
0:19:51 And so I think, yeah, the incentive structures
0:19:53 that are built in aggregate tend to be far more complex
0:19:55 and they kind of interact in ways
0:19:58 that tend to be difficult to analyze
0:19:59 and could result in complexity
0:20:01 that could ultimately result in attacks.
0:20:03 And you post, you talk a little bit
0:20:05 about what you refer to as the dark DAO,
0:20:07 which sounds like a fairly dark picture
0:20:10 of what could end up being the case.
0:20:11 In your view, what is the worst case scenario here?
0:20:14 How could this unfold in a bad way?
0:20:15 – Yeah, so there’s a lot of different variants
0:20:17 of the dark DAO which have different assumptions
0:20:19 in the post, some of them require trusted hardware,
0:20:20 some of them don’t.
0:20:22 But the ultimate point of the dark DAO
0:20:23 is that it’s a private smart contract
0:20:26 for attacking a vote, for vote buying,
0:20:28 that essentially hides from the rest of the world
0:20:30 how much money is committed to this contract,
0:20:33 who is participating in the vote buying contract,
0:20:35 and sort of how far along the contract is.
0:20:37 But sort of is a way to frictionlessly
0:20:40 and permissionlessly form a vote buying cartel
0:20:41 for a particular vote.
0:20:42 And this could be sort of a funding pool,
0:20:44 anyone can come contribute money to it.
0:20:45 So if it’s outcome specific,
0:20:47 it could be funded by anyone who’s interested
0:20:48 in such an outcome,
0:20:50 whether it be other blockchain projects,
0:20:53 users on the system, outside groups, whatever it may be.
0:20:55 So once this dark DAO is funded,
0:20:57 what it does is sort of offer up vote buying
0:20:58 to people in the system.
0:21:01 And if people in the system come take this vote buying,
0:21:02 they retain access to their funds,
0:21:05 they keep using their wallet as they normally do,
0:21:07 but they’re sort of shackled by the dark DAO
0:21:08 that for this particular vote,
0:21:10 they can only vote in this certain way.
0:21:10 And this is trustless
0:21:12 because both sides have some guarantees.
0:21:16 So the vote buyers or vote buying network
0:21:18 or whatever it may be has guarantees
0:21:19 that potentially no one will find out
0:21:21 who’s being bought or sold
0:21:23 and how much money is pledged to it.
0:21:24 They’re guaranteed that if they pay for a vote,
0:21:27 this vote will actually be executed in the protocol,
0:21:29 even if the protocol does have
0:21:31 the classic properties of coercion resistance.
0:21:33 Another sort of sidebar of the dark DAO
0:21:36 is that trusted hardware, which is a new technology,
0:21:38 sort of breaks all classical coercion resistance voting
0:21:40 schemes in the blockchain world
0:21:42 and in the regular election world.
0:21:43 So once they launch this attack
0:21:45 and they start buying and selling people’s votes,
0:21:47 they have a number of options available to them.
0:21:48 One cool thing you can do
0:21:50 is you can tell everyone in the cartel
0:21:52 when a certain threshold is reached.
0:21:54 Let’s say when like 70% of the,
0:21:57 or 10% of the votes are locked into this DAO.
0:21:59 And you can do this in a way that’s deniable
0:22:01 such that everyone inside the cartel can check,
0:22:03 yes, 70% is reached,
0:22:05 but no one outside the cartel has any way of knowing
0:22:07 that this is actually reached.
0:22:09 So you can enforce an information asymmetry
0:22:12 that allows for profiting through things like shorting.
0:22:14 You can also enforce stronger information asymmetries,
0:22:16 so not even allow the people who are being bribed
0:22:19 to know at any time how much money is in it
0:22:23 or even potentially whether they voted at all
0:22:25 if the scheme is receipt free.
0:22:27 So it’s a very, very powerful class of attack.
0:22:29 You can spin it up however you want.
0:22:31 It allows people to pool their money and buy votes
0:22:34 in a way that they can keep any part of that secret
0:22:35 to any group of people that they want.
0:22:37 And the outside system has no way of knowing
0:22:39 sort of how far along the attack is.
0:22:41 In some ways, it also represents a credible threat.
0:22:42 If I were to launch a dark DAO,
0:22:44 I might not even need to necessarily have people
0:22:45 participate in it.
0:22:47 Just its existence might be enough
0:22:50 to shake people’s confidence in that underlying vote.
0:22:52 So when we publish that blog post,
0:22:54 we’ve had a lot of reactions from voting projects
0:22:55 and other people in the space.
0:22:57 And I think there is a good question
0:22:58 of why haven’t we seen this already?
0:23:01 But at the end of the day, these systems are tiny, right?
0:23:03 Blockchains today are a drop in the bucket
0:23:05 of like the world financial system
0:23:07 and the incentives just aren’t there yet.
0:23:08 But if we are to use these technologies
0:23:10 and if we are to scale things,
0:23:13 I think these are absolutely realistic scenarios
0:23:15 and potentially nightmare scenarios.
0:23:16 – Yeah, that sounds insane.
0:23:18 And that’s definitely an outcome that is to be prevented.
0:23:21 And I think, I mean, this matters because
0:23:24 if we just take a step back and think about why is governance
0:23:26 so topical and so important in the world of crypto
0:23:27 and blockchains today?
0:23:32 It is because so much of what drives the space forward
0:23:35 in what is sort of the underlying philosophical motivation
0:23:39 is that power over these networks is decentralized.
0:23:41 And so decentralization here refers to
0:23:42 a bunch of different things at the same time.
0:23:44 Like people talk about decentralization
0:23:46 as it refers to sort of consensus,
0:23:50 like who gets to decide like who modifies
0:23:51 the underlying ledger,
0:23:52 but also decentralization applies
0:23:54 to who gets to modify the code.
0:23:56 These networks are decentralized in that
0:23:58 they’re kind of like self-governing organizations
0:24:00 and they don’t have at least philosophically
0:24:03 any central points of control where any one individual
0:24:06 can decide how to sort of modify the code
0:24:08 or make it work in any particular way.
0:24:12 And so all of these initiatives to try to build in governance
0:24:15 into the protocols are an effort to try to
0:24:17 sort of decentralize even that aspect
0:24:20 and to try to make it so that the code itself can evolve
0:24:22 in a way that is still community driven
0:24:24 and not kind of centrally controlled
0:24:27 by the core developer team.
0:24:28 – Yeah, I think the promise of a lot of these systems
0:24:31 is sort of this crypto economic security, right?
0:24:33 You have this mechanism and because the mechanism works
0:24:34 and the incentives are set up right,
0:24:36 everyone comes together harmoniously
0:24:38 and produces something that is bulletproof
0:24:42 and very strong because of the incentives and the mechanism.
0:24:43 An example of this is Bitcoin.
0:24:45 Because of the money paid to miners,
0:24:48 people are burning a small country’s worth of electricity
0:24:50 to try to secure this transaction ledger
0:24:52 that has actually worked fantastically so far.
0:24:53 So when you design these systems,
0:24:56 there needs to be some sort of underlying mechanism
0:24:57 and some sort of reasoning about the security
0:24:59 of that mechanism.
0:25:01 But what these technologies like the dark dial
0:25:03 and private smart contracts allow you to do
0:25:06 is use external money to sort of alter the incentives
0:25:09 inside that game and alter the security properties
0:25:12 that people are actually getting from their project
0:25:15 in a permissionless and trustless way.
0:25:17 So this does sort of speak
0:25:20 to the fundamental coordination of blockchains, right?
0:25:23 Like how do we design these games to coordinate people
0:25:26 to make choices in a way that’s not controlled
0:25:28 by one particular individual, as you said,
0:25:30 or some social trust hierarchy,
0:25:32 but by the economics of the system itself?
0:25:33 And in that model,
0:25:35 if you can’t be secure against economic attacks,
0:25:38 then you’re sort of building something
0:25:40 that doesn’t make much sense in my opinion.
0:25:43 And so I guess that’s a lot of what my work is looking at.
0:25:45 – Right, what do you think are the implications
0:25:47 of vote buying on proof of stake?
0:25:50 – So proof of work is where people use hardware
0:25:51 to sort of solve hard problems.
0:25:53 And if they solve the problem,
0:25:55 then they can post a block to the network.
0:25:56 Rather than using this mechanism,
0:25:59 proof of stake allows people to vote using their coins.
0:26:01 So they lock up their coins for some long period of time
0:26:04 and they can use any number of protocols to do this.
0:26:07 The core idea here is that instead of proof of work
0:26:09 where the economic security you get
0:26:12 is because people are doing this useless computation problem
0:26:13 that is sort of burning money
0:26:16 and there’s some costs associated with doing this,
0:26:18 is that people are paying liquidity costs
0:26:20 to lock up these coins for a long, long period of time
0:26:22 and they’re also taking risks
0:26:23 that they may incur penalties
0:26:25 if they misbehave in the protocol.
0:26:27 And with these liquidity costs,
0:26:28 they’re taking like massive volatility risks
0:26:30 in cryptocurrencies, right?
0:26:32 So if they do something that crashes the system,
0:26:33 well, their coins are locked up
0:26:34 and they’re going to lose money.
0:26:36 If the network decides they misbehaved,
0:26:38 well, they can get rid of all their coins
0:26:39 and they’re gonna lose money.
0:26:40 So it’s this idea of bootstrapping
0:26:42 the economic security of the network from the coins
0:26:45 rather than from some external hardware source.
0:26:46 Obviously that comes with a lot of trade-offs
0:26:48 that are maybe beyond the scope of this discussion,
0:26:50 but at the end of the day, it’s also a voting protocol.
0:26:53 You have these people with coins, they decide how to vote.
0:26:55 So where does vote buying come in here?
0:26:57 Well, obviously this proof-of-stake protocol has an outcome.
0:27:00 It decides what history of the network is valid
0:27:02 and this outcome has all sorts of economic implications.
0:27:05 It decides who gets to send money to who.
0:27:07 It decides who is censored in the system.
0:27:10 It decides what order transactions happen in canonically
0:27:12 according to everyone in the system
0:27:14 and with that comes a lot of profit opportunity.
0:27:17 So I can potentially profit by censoring you
0:27:19 or I can profit by putting my transactions in front of yours
0:27:22 when you wanna execute an order on a decentralized exchange
0:27:25 or I can profit in sort of any number of different ways
0:27:26 by manipulating this vote.
0:27:28 So what you can do with the dark DAO
0:27:30 is to start a staking pool where I say like,
0:27:32 you know, let me do my algorithmic trading
0:27:34 and decide what order of transactions
0:27:35 makes me the most money.
0:27:36 You don’t necessarily care
0:27:38 if someone who’s doing a transaction on a dex
0:27:40 gets front-run and loses like $5, right?
0:27:43 So you say, okay, I’ll happily participate in this.
0:27:45 It’ll still keep the value of my coins high,
0:27:46 especially if I don’t have a lot of coins
0:27:48 and you’re paying me like twice as much
0:27:50 as any other staking pool.
0:27:52 So it sort of opens these coordination mechanisms
0:27:54 for attacks on the underlying transaction history
0:27:56 and the underlying consensus.
0:27:56 – Do you think that there’s a way
0:27:59 of making a proof-of-stake network secure?
0:28:01 – It depends on your definition of secure.
0:28:04 I think it really depends on the type of security you want,
0:28:05 I guess.
0:28:07 – Yeah, and this all gets to the broader question
0:28:10 of like economic security of a blockchain.
0:28:11 And in the case of proof-of-stake,
0:28:13 the resource that’s used to secure the blockchain
0:28:14 is internal to the network.
0:28:15 In the case of proof-of-work,
0:28:18 it’s sort of electricity and like hardware
0:28:21 that’s used external to the network to secure the ledger.
0:28:22 And there are many other kind of approaches.
0:28:26 Like people are experimenting with doing useful work.
0:28:28 Instead of burning electricity uselessly
0:28:30 as you do in proof-of-work,
0:28:32 people try to build a sort of proof-of-space
0:28:34 or proof-of-spacetime protocols
0:28:39 where like for example, you’re able to store files
0:28:41 and storage becomes the resource that people use
0:28:43 to then secure the network.
0:28:44 What do you think of that kind of approach?
0:28:46 – So fundamentally to vote buying,
0:28:48 it doesn’t actually matter what resource you’re using.
0:28:50 Vote buying works for proof-of-work too.
0:28:52 So I could use dark DAO like technology
0:28:53 to start the mining pool.
0:28:55 And the properties of the mining pool would be
0:28:56 you come, you mine here.
0:28:57 I’ll pay you more than we’re making
0:28:59 because I have some external incentive
0:29:02 to censor someone or reorder transactions or whatever.
0:29:04 And then you get the dark DAO privacy properties
0:29:06 of no one knows how much hash power is participating
0:29:09 in this pool or who’s getting paid or things like that.
0:29:11 So these certainly also apply to systems
0:29:15 that use things like files and other useful work properties.
0:29:17 I think there’s a whole class of other questions
0:29:19 on the economic security of those systems.
0:29:21 So you have to be really careful
0:29:23 about where the economic security comes from.
0:29:24 I think you have to be really careful
0:29:26 with what useful means.
0:29:28 Whether the fact that it’s useful also introduces
0:29:31 any external incentives to mess with it, right?
0:29:35 So you could imagine like if the useful thing
0:29:37 the network was doing was like powering a search engine
0:29:38 or something, right?
0:29:39 Those results are valuable
0:29:42 and they bring external actors in who want to manipulate that.
0:29:43 And there’s sort of this feedback loop
0:29:45 between the mechanism securing the protocol
0:29:48 and the utility of what the protocol is actually providing.
0:29:50 But there’s definitely some people in the community
0:29:53 that look at that and say this is all way too complicated.
0:29:54 This is never going to work.
0:29:55 You have to have it be useless
0:29:57 because there’s no external incentives
0:29:59 and messy things that way.
0:30:02 I personally think that’s an open question.
0:30:03 – Yeah, there’s this argument that people make
0:30:06 that if the resource that is used to secure the network
0:30:08 is very commoditized
0:30:09 and just generally exists in the world
0:30:12 in the world in sort of plentiful quantities
0:30:13 that for example, in the case of storage,
0:30:16 if storage is the resource that’s used to secure the network
0:30:17 then anyone with a bunch of storage
0:30:19 could presumably attack the network.
0:30:22 Whereas in the case of a network like say Bitcoin
0:30:25 where you have ASICs that are specific to the network
0:30:27 in order to attack the network
0:30:28 you have to get your hands on those ASICs
0:30:30 and those ASICs aren’t useful for anything
0:30:31 but mining Bitcoin.
0:30:35 So people would argue the security of that kind of
0:30:38 the economic security of that kind of model is better.
0:30:40 – Yeah, and Joe Bono has a fascinating line of work
0:30:41 on these problems.
0:30:42 So if you Google Goldfinger attacks
0:30:44 he has a paper and a presentation.
0:30:47 There’s also the question of like buying versus renting.
0:30:48 So if something is very commoditized
0:30:50 you may be able to rent it
0:30:52 which substantially subsidizes the tax.
0:30:54 You may be able to buy it, perform the attack
0:30:56 and then resell it into the commodity market
0:30:59 which again substantially subsidizes the attack.
0:31:02 So these are all open and very complex questions
0:31:04 but people will build the systems and we’ll see.
0:31:06 This is sort of a classic pattern you see
0:31:08 in traditional finance.
0:31:10 And then you’ll have sort of black swan
0:31:13 and tail risk like events that surprise people.
0:31:15 – So we’ve talked a lot about governance in general
0:31:19 but you obviously are working on a ton of interesting stuff
0:31:21 just generally with respect to economic security
0:31:23 for crypto networks and blockchain
0:31:24 just the computer security.
0:31:27 What are some of the other interesting ideas
0:31:29 or sort of lines of work that you’re exploring?
0:31:32 – So one that I’m extremely personally interested in
0:31:35 is fairness guarantees for users around these systems.
0:31:36 A lot of what attracted me to them in the first place
0:31:39 was this promise of sort of eliminating the middleman
0:31:41 and making things in control of the user.
0:31:43 Like be your own bank, you don’t need these institutions
0:31:45 to tell you how to set your money supply
0:31:47 or how to route your transactions
0:31:50 or what exchange to use, et cetera, et cetera.
0:31:52 I look a lot at those guarantees
0:31:54 and sort of the ways in which modern blockchain solutions
0:31:56 are failing to meet those guarantees.
0:31:59 So one example of that is in the decentralized exchange space.
0:32:01 That’s something that’s seen a lot of promise
0:32:03 from people who wanna build these exchanges
0:32:06 that aren’t vulnerable to hacks and other user fund theft.
0:32:08 Unfortunately, the way these mechanisms
0:32:10 that people are building interact with the blockchain
0:32:13 is very complex and opens the door for external actors
0:32:15 to make a lot of money from front running them
0:32:17 and make a lot of money from doing algorithmic trading
0:32:19 on the network and everything that you see
0:32:21 in the traditional financial world.
0:32:24 So some of my work is around how large is that economy
0:32:27 and what are the failures of those guarantees?
0:32:30 – What are some interesting results so far on that front?
0:32:33 – So it’s actually probably a bigger market than you think,
0:32:36 even though DEXs have not seen substantial volume.
0:32:38 So this is a big problem for users.
0:32:41 It also highlights a lot of weird quirks of these systems,
0:32:44 such as like allowing for typos that end up costing users
0:32:47 a lot of money when programmatic actors soup in
0:32:50 and sort of take advantage of these inefficient mechanisms.
0:32:53 And it also raises fundamental questions about, I guess,
0:32:55 whether we’ll be able to do something that’s different
0:32:57 from the current financial system
0:32:59 because there are still these information asymmetries
0:33:01 that come up and this is a worldwide network.
0:33:02 And at the end of the day,
0:33:04 someone is still ordering transactions.
0:33:07 So is this rent sort of implicit to all blockchains?
0:33:09 How large is it?
0:33:12 And does it threaten the security of the overall blockchain?
0:33:13 Which I think it may.
0:33:15 – So I think one very interesting line of work
0:33:18 that you did was around gas token
0:33:20 and tokenizing gas on the Ethereum network.
0:33:23 – So this sort of came out of this arbitrage project.
0:33:25 We wrote a blog post very early on,
0:33:27 last I think October, November,
0:33:29 essentially saying decentralized exchanges are flawed.
0:33:31 You can just run this 20 line Python script
0:33:33 and you can profit off of users in a way
0:33:34 that was maybe not foreseen
0:33:37 and is not sort of explicitly stated to them
0:33:39 because of how inefficient these mechanisms are.
0:33:40 And before we wrote this blog post,
0:33:42 we were actually doing this to test it, right?
0:33:44 And we said, we made X dollars, whatever.
0:33:45 After we wrote the blog post,
0:33:47 sort of this cottage industry spawned
0:33:50 of like a few dozen people who are competing
0:33:53 in sort of this market and trying to outbid each other
0:33:57 to get their transactions first in that mind order
0:33:59 and take advantage of these opportunities.
0:34:00 So we’ve been studying that market for quite a while
0:34:02 and competing against these guys.
0:34:04 And unfortunately at some point,
0:34:05 they started out competing us.
0:34:07 So we started competing on what’s called gas,
0:34:08 which is the price you’re willing to pay
0:34:10 per unit of transaction.
0:34:12 The way it works is you make a typo Ali,
0:34:14 it puts a million dollars on the table
0:34:16 for anyone who can get their order in ahead of that typo
0:34:18 and sort of take advantage of your typo.
0:34:20 And then I would like to do a $5 transaction
0:34:23 to take advantage of Ali’s mistake, right?
0:34:26 And then maybe someone else is willing
0:34:27 to do a $10 transaction
0:34:29 ’cause it’s a million dollar opportunity, right?
0:34:31 So we sort of get into this bidding war of like,
0:34:33 minor, please pick me first, minor, please pick me first.
0:34:35 That’s inherent to how these transactions
0:34:36 are ordered by miners.
0:34:38 And what we noticed is that when you have like 10 of these,
0:34:39 we were rarely profiting
0:34:40 because we didn’t have the best latency,
0:34:42 we didn’t have the best infrastructure
0:34:44 and they were getting their bids out faster.
0:34:46 They were getting them two miners faster
0:34:48 and they were willing to bid up higher than we were
0:34:50 to essentially take these opportunities.
0:34:51 So that’s where gas token came in.
0:34:54 It’s a way to sort of store this gas for the longer term
0:34:57 rather than just paying for it when you do your transaction.
0:34:59 So gas is the transaction fee.
0:35:00 And usually you say, okay,
0:35:03 I’m willing to pay a $100 fee for this transaction.
0:35:05 Instead, what you could do is sort of bank
0:35:06 a transaction’s worth of gas
0:35:08 and then just deploy that bank gas
0:35:09 and not pay as much fee
0:35:11 for the transaction you are doing.
0:35:12 And that works by taking advantage
0:35:16 of this fundamental issue in Ethereum’s resource model
0:35:18 which has to do with how you pay
0:35:22 to sort of incentivize people to clean up after themselves.
0:35:24 So in Ethereum, you actually give people a refund in gas
0:35:27 if they delete something they stored in the network previously
0:35:29 to incentivize them to not leave garbage around
0:35:30 that everyone has to store forever.
0:35:32 So what we do is when gas is cheap,
0:35:35 we fill the Ethereum state with junk
0:35:36 and then when it’s expensive, we delete this junk
0:35:39 which gives us a refund at that higher price
0:35:42 that we can use to subsidize these arbitrage transactions
0:35:44 which often costs thousands and thousands of dollars in fees.
0:35:47 Like people are bidding multiple thousands,
0:35:49 even tens of thousands in fees on these transactions.
0:35:51 – Right, and so to clarify for those not already familiar,
0:35:54 so gas is basically the resource that you use
0:35:56 to pay for computational resources
0:35:57 on the Ethereum blockchain.
0:36:00 So if you want it to buy computations, say instructions
0:36:03 that miners will execute for you, you pay for those in gas.
0:36:05 If you want it to buy a storage,
0:36:08 you similarly also pay for storage in gas.
0:36:10 And the current model of Ethereum is that you buy
0:36:14 some storage on the blockchain for a fixed price upfront
0:36:16 and then that storage sort of remains
0:36:18 on the blockchain forever.
0:36:20 And the Ethereum blockchain has this mechanism
0:36:22 that if you were to delete that storage,
0:36:23 if you were to free it,
0:36:26 then you will receive a refund for the amount that you paid.
0:36:30 Some refund for what you paid originally
0:36:31 for that amount of storage.
0:36:35 And so you’re basically saying that when gas is very cheap,
0:36:38 you can sort of fill storage on the blockchain
0:36:42 and then reclaim a refund later once gas is expensive.
0:36:45 And sort of the gas will be worth more at that point
0:36:47 than it was when you stored it.
0:36:48 And you could sort of leverage that
0:36:50 to kind of increase the amount of gas
0:36:51 that’s available to you.
0:36:53 – Yeah, and our fundamental observation was that
0:36:54 this is basically a derivative on gas.
0:36:57 It’s like a call option on some gas.
0:36:58 It led to the broader question
0:37:00 of how are these resources actually priced?
0:37:03 Like how do people choose how much is paid for storage?
0:37:06 How do people choose how much is paid for computation?
0:37:07 And in what ways are these suboptimal?
0:37:10 – So you mentioned the current model of pay one store forever.
0:37:12 That’s something we certainly address in our work,
0:37:14 proposing more of a rentful scheme
0:37:17 where you have to pay for ongoing costs at market rate.
0:37:19 There’s also the issue of who’s getting the payment.
0:37:21 So the fact that the miners get payment for storage
0:37:24 when the miners actually don’t need to store the whole state
0:37:26 and it’s the full nodes that bear the cost.
0:37:29 So this sort of asymmetry between who’s bearing the cost
0:37:31 like where the externality is
0:37:32 and like who’s actually profiting
0:37:34 is super important to study.
0:37:36 It leads to a sort of tragedy of the commons
0:37:38 in the worst case where the miners are happy to take payment
0:37:40 for as much storage as you want
0:37:42 because they don’t have to store it and they don’t care.
0:37:43 As long as they don’t break the whole network,
0:37:46 they’ll happily push out as many full nodes as they can.
0:37:47 So these are broader questions.
0:37:50 We have a broader initiative called Project Chicago,
0:37:53 which you can see at projectchicago.io.
0:37:55 That basically is studying these questions
0:37:56 of crypto commodities.
0:37:59 What are the underlying commodities behind blockchains?
0:38:02 For example, computation, relay network and storage.
0:38:03 How are these commodities priced?
0:38:05 How can you exploit these commodities?
0:38:07 How can you exploit like the relay network
0:38:09 to get information about people’s transactions earlier
0:38:13 or the computation layer to sort of, I don’t know,
0:38:16 do this kind of gas refund or something like that.
0:38:19 So there’s a lot of interesting work in that direction.
0:38:21 – Yeah, by the way, why is it called Project Chicago?
0:38:23 – So it’s called Project Chicago
0:38:25 ’cause our inspiration is sort of the Chicago mercantile
0:38:26 exchange.
0:38:28 That’s how businesses hedge against volatility
0:38:31 and sort of price commodities in real world markets.
0:38:35 So we think of this as sort of exploring something similar
0:38:36 on blockchains and asking like,
0:38:38 is that the right model or can we do better
0:38:40 now that we have all these decentralized tools
0:38:41 at our disposal?
0:38:42 – Best painting.
0:38:44 Well, thank you so much for coming on the podcast.
0:38:46 – Yeah, thanks for having me.