AI transcript
0:00:02 (upbeat music)
0:00:07 Pushkin.
0:00:12 – I’m Dr. Laurie Santos, and to welcome the new year,
0:00:13 my podcast, The Happiness Lab,
0:00:16 is releasing a series of happiness how-to guides
0:00:18 to help you in 2025.
0:00:20 I’ll distill the wisdom of world-class experts
0:00:22 into easy to digest actionable tips,
0:00:24 struggling with tough emotions.
0:00:25 We have a how-to guide.
0:00:27 Worried that you’re not enough?
0:00:28 We got you.
0:00:30 Self-obsessed and want to get over yourself,
0:00:32 there’s a guide for that too.
0:00:35 The Happiness Lab’s how-to season starts January 1st.
0:00:37 Listen on the iHeart radio app, Apple Podcast,
0:00:39 or wherever you get your podcast.
0:00:48 – AI works amazingly well.
0:00:52 It works terrifyingly well even for virtual things,
0:00:56 for words, for pictures, for videos.
0:01:00 This is true in large part because of the internet.
0:01:03 The internet provides this wildly abundant,
0:01:07 readily available source of words, pictures, and videos
0:01:10 to train AI models.
0:01:13 But there is no analogous wildly abundant,
0:01:17 readily available data set for the physical world.
0:01:22 There is no gargantuan internet-like repository of data
0:01:25 that describes how things move and bend
0:01:29 and break in real physical space.
0:01:33 And as a result, we do not yet have robust AI
0:01:35 for the physical world.
0:01:37 But people are working on it.
0:01:42 And if they succeed, they’ll change the way the world works.
0:01:45 Not just the world as it appears on our screens,
0:01:47 but the actual physical world.
0:01:51 The world where if you drop something on your foot, it hurts.
0:01:54 (upbeat music)
0:01:57 – I’m Jacob Goldstein, and this is What’s Your Problem,
0:01:59 the show where I talk to people who are trying
0:02:02 to make technological progress.
0:02:05 My guest today is Edward Mayer.
0:02:08 He’s the co-founder and CEO of Machina Labs.
0:02:10 Edward’s problem is this.
0:02:15 How can you use AI to turn robots from dumb,
0:02:20 inflexible machines into skilled, versatile craftsmen?
0:02:22 Before he started Machina Labs,
0:02:24 Edward worked in the rocket ship business,
0:02:29 first at SpaceX, and then at a company called Relativity Space.
0:02:31 And in the rocket business,
0:02:35 Edward saw firsthand the problems of traditional manufacturing.
0:02:37 It’s the kind of problem he’s now trying to solve
0:02:39 with AI and robots.
0:02:43 It’s a problem he calls the rigid factory problem.
0:02:49 So I’ve heard you use this phrase that’s interesting to me,
0:02:50 and it’s the real thing.
0:02:54 And it’s interesting to me, and it’s the rigid factory problem.
0:02:56 What’s the rigid factory problem?
0:03:00 – The main problem with the factories today
0:03:02 is that rigidity, meaning that if you have to build
0:03:06 a physical product, you pretty much have to build a factory
0:03:09 that’s designed for it and built for it.
0:03:12 There’s a lot of components that goes into the factory,
0:03:15 from machinery all the way to the tooling
0:03:16 that is required to build products
0:03:19 that are specifically designed for the geometry,
0:03:22 for the material that you’re trying to use.
0:03:24 The moment you want to change that,
0:03:25 you have to change your factory,
0:03:26 which is a huge investment.
0:03:30 I always give an example from when I was at SpaceX.
0:03:33 You think of SpaceX as a very innovative company,
0:03:36 and it is on the edge of a hardware space
0:03:38 in terms of innovation.
0:03:43 In the past 24 years, 23, 24 years that they have existed,
0:03:45 they have two rocket families.
0:03:48 It’s a Starship and there’s Falcon, right?
0:03:50 Because the moment you decide on diameter of,
0:03:53 for example, Falcon 9 or the Falcon family in general,
0:03:55 the diameter of that core, it’s very hard to change it.
0:03:56 A lot of tooling and machinery
0:03:59 specifically built for that diameter.
0:04:02 And that’s why for Starship, they had to start from scratch.
0:04:04 – Start from scratch, meaning like not just design,
0:04:06 but like the factory itself.
0:04:09 Like they actually had to build a whole new factory
0:04:12 ’cause they wanted to make a different size rocket.
0:04:14 – Yes, different size, different material.
0:04:15 All the tooling has to change, right?
0:04:19 Almost, yeah, you have to basically assume,
0:04:21 building from scratch, ground up factory,
0:04:22 what does it need to be there
0:04:24 for us to build this new product?
0:04:28 – I heard you describe, was this from your own experience,
0:04:31 that sort of era at SpaceX
0:04:36 when the fact that you couldn’t make the rocket wider
0:04:39 led to all these kind of difficult things
0:04:40 people were trying to do to be like,
0:04:42 how can we do all these things
0:04:44 under this fundamental constraint?
0:04:46 Like can you talk a little bit about that?
0:04:48 – Yeah, this is a lot of conversation happening
0:04:52 in 2012, 2013, 2014 time
0:04:57 when the diameter of the Falcon 9 could not get any larger.
0:05:00 And if you look at actually different Falcon versions,
0:05:02 the height of that vehicle kept going higher,
0:05:04 but the diameter could not change.
0:05:07 So it was about, what space can you find
0:05:11 to put new features and new designs
0:05:13 that exist within the vehicle?
0:05:15 So there was a lot of stuff basically being crammed
0:05:17 into the space that you have already got.
0:05:19 – So that’s true for building rockets.
0:05:21 I mean, what are some other,
0:05:24 just different kinds of manufactured products
0:05:26 where that kind of rigidity is a problem?
0:05:29 – Yeah, I think it is just common,
0:05:31 almost in all manufacturing.
0:05:32 That’s why this phenomenon,
0:05:35 I think it’s kind of funny, people take it for granted
0:05:37 that a thing called economies of scale.
0:05:42 Like people take it for granted as if it’s rule of nature.
0:05:43 It’s actually not.
0:05:46 – And just to be clear, it’s basically the more you build
0:05:48 of a thing, the cheaper each one of those things gets.
0:05:49 If you build one, it’s really expensive.
0:05:52 If you build a million, each one’s a lot cheaper.
0:05:53 – Yes, exactly.
0:05:54 But then people don’t think about it.
0:05:56 It’s like, oh, okay, intuitively makes sense.
0:05:56 But why?
0:05:58 It’s actually not that intuitive.
0:06:01 It’s actually a limitation of technology, right?
0:06:05 Why economies of scale is a thing
0:06:08 is because you have to make a huge amount of investment
0:06:09 to make the first thing.
0:06:11 And the moment you make the second thing
0:06:14 and the third thing, then you can break even your investment
0:06:17 onto more products that you’re gonna come out of it.
0:06:19 But that’s only true if the second product
0:06:22 can be built for the first investment.
0:06:24 You have to turn this concept and say, oh, this is a given.
0:06:26 This is an axiom of the world,
0:06:28 that the economies of scale is a thing.
0:06:31 But in reality, it’s a technological challenge, right?
0:06:34 It means that you build a car, you ask what application?
0:06:36 You, once you build a factory for a car
0:06:40 and all the toolings and dyes that goes into stamping
0:06:42 of the panels of that car,
0:06:46 that’s $150 million investment just for stamping.
0:06:47 This is a number, for example, from Tesla.
0:06:50 Tesla spends $150 million in a stamping plant
0:06:54 they have in a factory in Texas, right?
0:06:57 And that can only make Model Y or Model 3, right?
0:06:59 The moment you have to change that,
0:07:04 that means go through every 80 to 130 sheet metal panels
0:07:08 that exist on that car and design a new tool for it.
0:07:10 And each of those tools is gonna be a few hundred thousand
0:07:12 dollars to sometimes a million dollars
0:07:14 or a million and a half million dollars.
0:07:17 And you’re talking about like 80 to 130 tools per vehicle.
0:07:18 – And like all you’re doing,
0:07:20 you’re not reinventing the car there.
0:07:23 You’re just making a car that’s a slightly different shape.
0:07:26 – Yes, maybe you may get sedan
0:07:28 and you’re not trying to do a slightly longer version,
0:07:30 a slightly bigger version.
0:07:32 And that’s why economies of scale are the thing.
0:07:34 You’re saying, okay, I made a factory.
0:07:38 Now it only pays back if I make a million of this car, right?
0:07:41 Because I had to just drop $150 million
0:07:42 on just a stamping plant.
0:07:44 So yeah, it’s all over manufacturing.
0:07:47 We abstract this whole concept and gave it a name
0:07:49 says economies of scale.
0:07:53 – Yeah, so you left SpaceX and you went to
0:07:55 Relativity Space, right?
0:07:57 A company that was also in the space business
0:07:59 that was using 3D printing, right?
0:08:01 That was the idea of the company,
0:08:04 which seems like an approach to this problem
0:08:05 that you’re talking about, right?
0:08:10 An advantage of 3D printing is that it is much more flexible
0:08:12 and less rigid than traditional manufacturing, right?
0:08:14 So tell me about that.
0:08:18 – Yeah, so yeah, we saw this challenge at SpaceX
0:08:20 and I joined Relativity very early on.
0:08:22 Well, I was the fourth person on that team.
0:08:26 And the goal over there was, okay,
0:08:29 let’s just think about this fundamentally.
0:08:32 Can we build a rocket that all built
0:08:33 with flexible technology?
0:08:37 And at the time, 3D printing was a forefront
0:08:40 of everybody’s minds because people were already starting
0:08:42 to build at NASA, at SpaceX.
0:08:44 People were already starting to build engines
0:08:46 out of 3D printing.
0:08:47 And the concept was like, well, that’s great.
0:08:48 It’s very flexible.
0:08:51 3D printing has this promise of geometry agnostic,
0:08:52 material agnostic.
0:08:53 You can just feed it a design and you can build
0:08:55 a product for it.
0:08:57 And it worked very well with rocket engines.
0:09:00 I think probably in the future, all rocket engines
0:09:02 would be 3D printed.
0:09:06 And the concept was, can we take this
0:09:08 and scale it to the whole vehicle, right?
0:09:10 Can we build the whole vehicle with a process
0:09:14 like 3D printing so that it is flexible?
0:09:16 Today, if you want to build a rocket
0:09:18 with 12 foot diameter, we can do it.
0:09:20 And then if our calculation changes
0:09:21 and we wanted to go to another orbit
0:09:23 or do a different type of emission,
0:09:25 then we can change that 12 foot diameter
0:09:27 to 20 diameter, 20 feet diameter.
0:09:28 – Don’t have to build a new factory.
0:09:30 Don’t have to build new machinery,
0:09:32 just use the same 3D printer, yeah.
0:09:34 – Exactly.
0:09:37 So that was the concept behind relativity.
0:09:39 That’s the thesis behind relativity.
0:09:39 And that was the goal there.
0:09:42 The goal was 3D print the whole rocket
0:09:44 so that you can be flexible.
0:09:48 – But it hasn’t worked at least in the kind of maximalist
0:09:49 version, right?
0:09:50 Like they just haven’t been able to do it.
0:09:53 They’ve sort of backed off of that big dream,
0:09:54 as I understand it.
0:09:55 – Yeah, yeah.
0:10:00 So I think the challenge was that 3D printing
0:10:02 is just one process.
0:10:07 And it’s necessarily not good for every type of part.
0:10:09 Manufacturing is very versatile.
0:10:11 You do different types of geometries,
0:10:12 different types of material.
0:10:14 And 3D printing has a very small reach.
0:10:17 There’s certain type of parts, like rocket engines,
0:10:18 very good fit.
0:10:22 You’re building a tank, maybe not so, right?
0:10:24 So yeah, it’s good for certain type of parts,
0:10:25 but there’s a whole lot of other parts.
0:10:28 Like I said, you’re building a fuel tank,
0:10:30 which is a basically large sheet metal
0:10:33 or thin walled structure.
0:10:35 Then maybe 3D printing is not as good of a fit
0:10:36 because it takes a long time.
0:10:38 And also because it’s thin,
0:10:40 you have a lot of physical challenges
0:10:43 in terms of controlling the geometry and the tolerances.
0:10:48 So we realize soon that maybe other processes
0:10:51 are also need to be automated
0:10:53 the same way 3D printing is.
0:10:56 You need to have more flexible processes
0:10:57 that are not just one process,
0:10:59 more flexible platforms that can do different types
0:11:02 of processes, not just 3D printing,
0:11:07 to be able to cover a whole variety of products
0:11:09 in a flexible manner, the same way the 3D printing,
0:11:12 that’s for certain type of products.
0:11:14 And that was actually the thinking behind Machina Labs
0:11:17 is that, okay, can we step back and say,
0:11:18 what do we need to build?
0:11:20 What is this flexible platform
0:11:23 that can do 3D printing if needed,
0:11:25 or can do sheet forming if it’s needed,
0:11:27 it can do machining if it’s needed,
0:11:29 but chooses the right operation,
0:11:31 right flexible operation for the right part,
0:11:34 but still very agile and doesn’t require a lot of tooling
0:11:35 and it’s not inflexible.
0:11:39 – So it’s sort of zooming out more.
0:11:42 It’s saying 3D printing is not gonna do everything.
0:11:44 The way manufacturing works now,
0:11:47 it’s just too rigid, too hard to change things,
0:11:50 too reliant on scale to make the economics work out.
0:11:55 So like that’s a very big, very abstract thought.
0:11:59 To start a company, you gotta make something,
0:12:02 or you gotta make something that makes something.
0:12:04 Like what do you actually do?
0:12:07 – Yeah, so it was interesting, right?
0:12:11 You know, actually the solution was in our past, right?
0:12:13 If you look at–
0:12:14 – It’s like the lesson in a movie.
0:12:16 It’s like the Wizard of Oz or something.
0:12:19 – Exactly, if you look at manufacturing,
0:12:21 I mean up to Industrial Revolution,
0:12:23 it was arts and crafts, right?
0:12:26 It was basically humans trying to figure out
0:12:27 how to conquer nature, right?
0:12:30 Like how am I gonna use my hands and my brains
0:12:35 and very few primitive tools to deform a product
0:12:39 or shape a product from raw material, right?
0:12:43 And to this date, if you are in a very high mix manufacturing,
0:12:45 still a lot of that creativity exists.
0:12:49 There is a person at SpaceX, his name is Big John.
0:12:51 I don’t think he’s there anymore.
0:12:53 But there was this guy who’s like, you know,
0:12:57 very skilled maker at Craftsman.
0:12:59 He could figure out how to use simpler tools
0:13:01 to build different things in a creative way.
0:13:03 Maybe it’s not a repeatable way like, you know,
0:13:07 a stamping works or injection molding works,
0:13:08 but you can be flexible.
0:13:09 You can do different types of things.
0:13:11 You can be creative about it and do different types of things.
0:13:13 So the inspiration came from
0:13:16 how actually humans used to do manufacturing.
0:13:19 But realize in order to be flexible,
0:13:21 you actually need two components.
0:13:25 You need intelligence and you need set of simple tools
0:13:27 with a lot of kinematic freedom.
0:13:29 Now you can pick up those simple tools.
0:13:30 And as long as you have the intelligence
0:13:32 on how to use tools and what sequence
0:13:34 and what kind of a process parameters,
0:13:35 how to use those tools,
0:13:38 you can actually do a whole variety of projects.
0:13:40 – And so when you say kinematic freedom,
0:13:42 you basically mean like robot arms
0:13:44 that can move in lots of different ways.
0:13:46 Is that practically what kinematic freedom
0:13:47 means in this context?
0:13:50 – Yes, basically can apply these tools
0:13:52 with a lot of freedom to the material, right?
0:13:55 The same way humans do it, right?
0:13:57 As a human, you know, if you think about it,
0:14:00 you can pick up a welder and weld something,
0:14:02 and then you drop the welder and you pick up a drill
0:14:04 and you put a hole in it and you drop the drill
0:14:09 and you pick up a hammer and maybe hammer it into shape.
0:14:11 So you actually have a few set of tools,
0:14:14 but you have a lot of good kinematic freedom
0:14:16 and most importantly, very creative mind
0:14:19 to tell you how to apply these tools to the material
0:14:22 so they can actually get very complex set of products
0:14:24 and a lot of diversity.
0:14:27 – So plainly, instead of big John, you want a robot, right?
0:14:29 That’s where the, that’s the kinematic freedom.
0:14:32 The tools are kind of like old tools,
0:14:34 but optimized for the robot.
0:14:36 And then when you say intelligence,
0:14:38 that’s the one where it’s like feels more frontier-ish.
0:14:42 Like, does that mean like clever engineers figuring out
0:14:43 how to automate the robots?
0:14:44 Does it mean AI?
0:14:45 Does it mean both?
0:14:48 – Yeah, so I think, yeah, you’re basically getting
0:14:51 to the crux of how do you scale it, right?
0:14:53 You have to have those three components
0:14:56 and how does the intelligent piece,
0:14:57 which is the most important piece,
0:15:00 comes into play in an automated fashion.
0:15:03 So early days, we started from basic intelligence of humans,
0:15:07 but then we had a plan to capture data and train AI
0:15:10 so that you can replace the thinking and the creativity
0:15:12 that human had to put in.
0:15:14 – What’s the first thing you decide to try and build?
0:15:17 What’s the first sort of problem you want to solve?
0:15:17 – Yeah, I think so.
0:15:19 I left for relativity in 2018.
0:15:22 And the idea when I left for relativity was there, right?
0:15:24 I was like, okay, we need to build basically
0:15:27 what I had in my mind, call it robot craftsman.
0:15:29 Robo craftsman, we call it at that time.
0:15:31 How can you build a robot system to your point?
0:15:33 Can pick up different tools, has the same kinematic,
0:15:35 but also have to have the intelligence.
0:15:38 The challenge is, you said,
0:15:39 in order to train these robots with AI,
0:15:42 you need to have a lot of data.
0:15:44 And this is not the data you can find on internet.
0:15:48 – Right, this is the AI robotics problem, it seems, right?
0:15:50 Like, unlike with large language models,
0:15:52 like that’s why we have large language models
0:15:54 and not AI robots, right?
0:15:56 Because we have the data
0:15:58 just sort of randomly sitting around on the internet
0:16:01 and we don’t have that physical world data for robots, right?
0:16:02 – Exactly.
0:16:06 – So basically the problem narrowed down into, okay,
0:16:08 how can I generate enough data?
0:16:09 How can I create a business
0:16:12 that has a sustaining way of generating data
0:16:13 so I can actually build these models?
0:16:17 I can build this intelligence for these robots.
0:16:19 And the thinking was, okay,
0:16:23 I need to create a solution that can scale in the industry
0:16:28 with limited amount of data and some heuristic.
0:16:30 But then because it’s a scaling,
0:16:32 we can generate a lot of data
0:16:33 and it starts building the AI model.
0:16:34 – Right.
0:16:37 You need a first thing that you can actually do
0:16:39 before you really have AI
0:16:42 to generate the data that will get you to AI.
0:16:44 – Exactly, exactly.
0:16:46 So we’re thinking about, okay,
0:16:49 it needs to be a large enough market, right?
0:16:51 Where we can get mass adoption
0:16:54 and we need to solve a problem that’s big enough.
0:16:56 It’s 10 times at least better than the current solution.
0:16:58 So it can actually get adoption, right?
0:17:01 – Meaning you can’t just do something as well.
0:17:03 You have to do it 10 times better.
0:17:05 – Yeah, because I think what we realized
0:17:07 is that through the last two companies,
0:17:09 if something is not 10 times better,
0:17:12 it cannot overcome the inertia
0:17:14 that exists in an industry for adoption.
0:17:16 Because if you’re doing something for the same way,
0:17:17 and you’re manufacturing,
0:17:19 people have been doing things the old way
0:17:21 for a hundred of years, right?
0:17:22 – And it’s a risk, right?
0:17:24 If they’re gonna try working with you,
0:17:26 they’re immediately taking a risk.
0:17:27 And if it’s only gonna be a little better,
0:17:29 why should I take that risk?
0:17:31 – Exactly.
0:17:32 So the idea was, okay,
0:17:34 we need to find a large enough market
0:17:35 for our first application,
0:17:36 and we need to have a solution
0:17:38 that at least 10 times better.
0:17:39 So that landed us.
0:17:40 We actually looked at a lot of things,
0:17:42 from 3D printing to forging to a lot of things.
0:17:44 And then landed on sheet metal.
0:17:47 So sheet metal is the largest metal processing sector
0:17:48 out of all.
0:17:50 It’s a $280 billion industry today.
0:17:54 And forming complex sheet metal shapes
0:17:55 is very tool intensive.
0:17:58 So what we started to do was,
0:18:01 okay, can we make a robot craftsman’s first operation
0:18:03 to be forming sheet metal?
0:18:05 Basically, forming sheet metal the same way
0:18:07 a sheet shape or hammer is a sheet into shape.
0:18:09 – And when I think about sheet metal,
0:18:11 I mean, what do, I don’t know anything about sheet metal.
0:18:13 I think of like, I think of cars,
0:18:15 I think of planes, right?
0:18:17 I think of like, you know, Detroit, like stamping.
0:18:19 Is that, am I thinking about the right thing?
0:18:22 Am I missing huge, a huge sheet metal universe?
0:18:23 Like, what’s the sheet metal universe?
0:18:26 – Yeah, so sheet metal almost is everywhere.
0:18:28 I think is the most common metal part
0:18:32 that you see on day to day, right?
0:18:33 Because most of the time we use metal
0:18:36 to be a container for other things.
0:18:38 So it’s usually a thin metal structure
0:18:41 that’s formed in complex shape to hold something else.
0:18:44 Now, you know, it can be from case of a computer,
0:18:48 you know, to a car, right?
0:18:49 You know, you’re sitting in a freeway,
0:18:50 you’re in a sea of sheet metal,
0:18:54 or to a airplane, you’re in a sheet metal can,
0:18:57 to a rocket body, for a lot of rockets,
0:19:00 some of them were composites with a lot of them are sheet metal.
0:19:05 And to agricultural heavy equipment machinery,
0:19:09 think of combines, tractors, to even building equipment.
0:19:14 So you look at your HVAC ducts are all sheet metal, right?
0:19:16 Because it just makes sense.
0:19:19 We mostly use metal parts to container other things.
0:19:21 And we give it complex shapes,
0:19:23 and that’s where sheet forming comes into play.
0:19:25 So you pretty much see it everywhere.
0:19:28 But the challenge is that in almost in all cases,
0:19:29 you have to create tooling,
0:19:30 it goes back to that first problem.
0:19:33 You have to create tooling for each of those geometries.
0:19:37 And that’s why, you know, a Ford needs to make sure
0:19:40 they can sell a million of an F-150
0:19:41 before they can invest in a plant
0:19:44 that makes a new version of F-150, right?
0:19:48 – ‘Cause you basically have to build a bespoke factory
0:19:52 just to shape sheet metal in a new way.
0:19:55 – Exactly, for a new geometry, for a new design, exactly.
0:19:57 – Where is that a particular problem?
0:20:00 Like, where is it, where does that acutely bind
0:20:03 the fact that sheet metal is so hard to do
0:20:05 if you’re not working at scale, so expensive?
0:20:07 – Yeah, so I think now you’re coming to the,
0:20:11 even the third stage of how do you scale this technology?
0:20:13 You need to first find, you know,
0:20:14 we said we need to be 10x better.
0:20:15 We need to go in an area. – Who’s your first customer?
0:20:18 – Yeah, you go in an area that has a lot of pain
0:20:19 with today’s technology. – Yeah, who’s like,
0:20:21 “Oh my God, thank God you walked through the door.
0:20:23 “We’ve been waiting for you.”
0:20:27 – Yeah, so end up being very much defense in aerospace, right?
0:20:31 So think of, you know, think of our military, for example,
0:20:36 right, today they have 50, 60 different weapon system
0:20:38 or defense systems.
0:20:40 You can basically think of like aircrafts
0:20:42 that they’re maintaining.
0:20:44 And some of these systems have been built
0:20:47 from 60, 70, 80 years ago.
0:20:49 Like think of B-50 to C-130.
0:20:52 – Like World War II planes still flying?
0:20:54 – It’s still flying, yes, exactly.
0:20:57 And they have like, you know, 30 of one,
0:20:59 50 of another, 100 of another one,
0:21:01 and these things get break down, right?
0:21:04 And unlike a Ford factory,
0:21:06 there is no factory for 70 different products
0:21:07 that they’re carrying. – Right.
0:21:11 – And presumably the factory they built in 1941
0:21:12 to build this plane doesn’t exist anymore.
0:21:13 Doesn’t exist.
0:21:16 And even the vendor might completely have disappeared, right?
0:21:18 That made that misuse component.
0:21:20 So they’re constantly battling with this challenge
0:21:23 of an aircraft goes down, how can it fix it?
0:21:24 How can it find the part?
0:21:25 And there are thousands of parts
0:21:26 in each of these aircrafts, right?
0:21:28 So any of them can go down.
0:21:29 And that’s a huge challenge.
0:21:31 I mean, if you look at, you know,
0:21:34 government of a government accountability office,
0:21:37 put this report out, I think it was a couple of years ago
0:21:41 or a year ago about how ready each weapon system is
0:21:43 to defend the United States.
0:21:46 Out of the 48, 49 weapon systems they look into,
0:21:50 only one, only one in the past 11 years,
0:21:53 every year was ready, right?
0:21:55 I think only top four had like,
0:21:58 at least half of the years ready, right?
0:22:00 So that means in most years,
0:22:02 these weapons are not ready to fight.
0:22:03 – Like they’re waiting for parts.
0:22:05 – They’re waiting for parts, something is broken,
0:22:07 something is damaged.
0:22:08 And just to go deeper,
0:22:11 some of these components take four years to be replaced.
0:22:12 So if a plane gets damaged,
0:22:14 it needs to sit on the ground for four years
0:22:16 before it can be replaced.
0:22:17 And the cost of replacement
0:22:19 is building another factory, basically.
0:22:21 So some of these parts, and think of it,
0:22:24 a landing gear door that goes on a plane,
0:22:26 it will cost them $800,000, for example,
0:22:27 because they have to go make a role.
0:22:29 – Because it’s bespoke, essentially.
0:22:30 – It’s very spoke, yeah.
0:22:31 – It’s like buying a bespoke suit or something.
0:22:33 It’s just like, it’s gonna cost a lot, yeah.
0:22:35 – Yeah, so the idea started there,
0:22:37 I think that was one of our first customers.
0:22:40 Can we make defense manufacturing more agile,
0:22:42 directly affects our national readiness
0:22:46 for military conflict, and it’s a huge problem.
0:22:47 But then, even in a broader sense,
0:22:50 any defense product or aerospace product,
0:22:53 usually has very low volume, but high mix of products.
0:22:55 Even you’re building a missile,
0:22:57 you make a few thousand a year,
0:23:00 and you might make five, six, seven different versions of it.
0:23:01 It’s very unlike cars,
0:23:05 where you make a million of the same car over and over again.
0:23:08 So that ended up being our first application,
0:23:11 which we got a lot of traction with.
0:23:13 But even outside of that,
0:23:16 you look at companies like Caterpillar,
0:23:17 like John Deere’s of the world,
0:23:19 these folks also are in the same boat.
0:23:22 They make 200 combines, right?
0:23:23 But they need to support them in the field.
0:23:26 And these folks have exactly the same problem, right?
0:23:30 Do I need to run a large factory
0:23:32 to support all these models at all time?
0:23:34 And that will be very expensive
0:23:36 to support like 100 vehicles out there.
0:23:40 (upbeat music)
0:23:41 Still to come on the show,
0:23:44 we’ll talk about the future of AI and robotics
0:23:46 at Machina Labs and beyond.
0:23:52 – Hey, it’s Jacob.
0:23:54 I’m here with Rachel Botsman.
0:23:57 Rachel lectures on trust at Oxford University,
0:24:01 and she is the author of a new Pushkin audiobook
0:24:04 called “How to Trust and Be Trusted.”
0:24:05 Hi, Rachel.
0:24:06 – Hi, Jacob.
0:24:09 – Rachel Botsman, tell me three things
0:24:11 I need to know about trust.
0:24:16 – Number one, do not mistake confidence for competence.
0:24:17 Big trust mistakes.
0:24:19 So when people are making trust decisions,
0:24:23 they often look for confidence versus competence.
0:24:27 Number two, transparency doesn’t equal more trust.
0:24:29 Big myth and misconception.
0:24:32 And a real problem actually in the tech world,
0:24:34 the reason why is because trust
0:24:38 is a confident relationship with the unknown.
0:24:39 So what are you doing?
0:24:41 If you make things more transparent,
0:24:44 you’re reducing the need for trust.
0:24:49 And number three, become a stellar expectation setter.
0:24:54 Inconsistency with expectations really damages trust.
0:24:55 – I love it.
0:24:57 Say the name of the book again
0:24:59 and why everybody should listen to it.
0:25:02 – So it’s called “How to Trust and Be Trusted.”
0:25:04 Intentionally, it’s a two-way title
0:25:08 because we have to give trust and we have to earn trust.
0:25:11 And the reason why I wrote it is because we often hear
0:25:13 about how trust is in a state of crisis
0:25:15 or how it’s in a state of decline.
0:25:17 But there’s lots of things that you can do
0:25:20 to improve trust in your own lives,
0:25:22 to improve trust in your teams,
0:25:24 trusting yourself to take more risks,
0:25:27 or even making smarter trust decisions.
0:25:29 – Rachel Botsman, the new audio book
0:25:32 is called “How to Trust and Be Trusted.”
0:25:33 Great to talk with you.
0:25:34 – It’s so good to talk with you
0:25:36 and I really hope listeners listen to it
0:25:38 because it can change people’s lives.
0:25:44 – And so you got the right market.
0:25:47 Now you gotta make a thing,
0:25:49 you gotta figure out how to actually do the thing,
0:25:52 how to make your idea come true.
0:25:53 Like, how does that work?
0:25:56 – So the idea originally was,
0:25:58 can we get rid of a die, right?
0:26:02 And do it the same way a sheet shaper forms a sheet of metal.
0:26:04 And what does a sheet shaper do?
0:26:06 A sheet shaper starts from a flat sheet of metal
0:26:09 and it slowly hammers it into shape.
0:26:12 So what we wanted to do was have a robot do that, right?
0:26:14 Have a robotic system basically do that
0:26:16 incremental defamation into shape.
0:26:18 We call it robot pouring.
0:26:19 – So you’re sort of bending it, right?
0:26:20 I mean, you’re hammering it,
0:26:22 sort of like if you take a, whatever,
0:26:24 cut open an aluminum can and kind of bend it into shape.
0:26:27 Like that’s a version of what’s happening here, right?
0:26:30 – Obviously in a much more complicated way, yeah.
0:26:31 – Exactly, you’re right.
0:26:35 I mean, the same way a potter forms a clay bowl,
0:26:36 that’s basically what our robots do.
0:26:38 They start from a flat sheet of metal
0:26:40 and it’s slowly deforming in the shape,
0:26:42 the same way a potter would form a clay bowl
0:26:45 or a sheet shaper hammers a sheet into shape.
0:26:46 – So I’ve seen it, right?
0:26:48 So there’ll be a sheet of metal like hanging,
0:26:51 hanging up in whatever above the ground.
0:26:54 And then you have a robot arm on either side, right?
0:26:55 Like one on one side, one on the other.
0:26:58 And then what happens?
0:26:59 – Basically, the robots come together
0:27:01 from both sides of the sheet.
0:27:04 And they pinch the sheet in a certain way
0:27:07 so that that location that they’re pinching
0:27:10 it slightly stretches in the forms, right?
0:27:13 And you, if you start applying this pinching
0:27:16 all over the sheet and incrementally,
0:27:20 you slowly start to form it into a shape, right?
0:27:22 So instead of traditionally we’d use a die
0:27:24 and with sheer pressure of the press,
0:27:26 pushing the sheet against the die to give it a shape.
0:27:28 Now the robots are like a craftsman,
0:27:29 like a tradesperson,
0:27:32 coming in to slowly deform the sheet
0:27:34 into shape by just applying pressure.
0:27:35 So one robot is pushing it,
0:27:37 the other robot is supporting it.
0:27:39 And by applying a pinch,
0:27:41 you slightly stretch the material
0:27:43 and you form it into a shape.
0:27:45 – So, I mean, the way you describe it,
0:27:50 it makes sense and it sounds easy.
0:27:51 I’m sure it wasn’t easy.
0:27:56 Like were there things that just didn’t work for a while?
0:27:58 – So it’s, you should have been here
0:28:01 when the first time we actually tried to form a part.
0:28:04 The part looked like it was like a ghost
0:28:06 of the geometry that they wanted to make.
0:28:09 And actually in the end it tore, right?
0:28:12 So think about it, you have this very flimsy sheet
0:28:14 and applying pressure to it.
0:28:17 And if you just apply pressure slightly wrong, right?
0:28:20 It can potentially tear it.
0:28:22 It can form it into a different shape.
0:28:23 And also the whole sheet is moving.
0:28:26 The whole time you’re trying to form it,
0:28:28 the whole sheet is moving because it’s very flimsy.
0:28:30 It’s not a rigid structure, right?
0:28:34 So the main challenge was how do you get this accurate?
0:28:35 And how do you get this process accurate?
0:28:37 How do you get accuracy?
0:28:40 And the idea was, what does the robot need to do?
0:28:43 Given all of these chaotic nature of the process,
0:28:45 where the sheet moves and if you apply too much pressure
0:28:48 it will deform in a bad way or it might tear.
0:28:52 If you apply too not enough pressure, it might just not form.
0:28:55 So how do you come up with the right set of robot movements
0:28:58 and process parameters to form the part?
0:29:00 And that was the problem we wanted to solve with AI, right?
0:29:04 Well, we didn’t have the data right in the beginning, right?
0:29:08 The idea was that if I form enough parts with this process
0:29:11 and I can capture all the data throughout the process,
0:29:12 where did the robot go?
0:29:13 How much pressure did it apply?
0:29:16 And what was the resulting geometry?
0:29:18 Then I can start building a model that says,
0:29:20 that correlates the inputs to the outputs.
0:29:21 And then I can explore this model and say,
0:29:25 “Okay, in order to get to the right output, I need these inputs.”
0:29:26 But we didn’t have them in the beginning.
0:29:28 So the idea was two things.
0:29:32 One was, maybe we can simulate the data, right?
0:29:35 And very early on we started doing some simulation,
0:29:37 physics-based simulation and we soon realized
0:29:39 in order to get an accurate result,
0:29:42 the simulations are going to be very computationally intensive.
0:29:46 A simulation of a part that took only 15 minutes to form
0:29:51 took us one week on 27 core machine, right?
0:29:52 So we’re like, okay, simulation,
0:29:55 not only is not accurate, it takes forever.
0:29:58 So we realized, okay, so that’s not the right route.
0:29:59 The right route was like, okay,
0:30:03 we can also form a lot of parts and gather the data.
0:30:05 But in order to do that, we go back to that same problem.
0:30:06 We need to have a scale.
0:30:08 We need to have a lot of these machines
0:30:09 forming these parts and get that data.
0:30:12 – I mean, one of the big AI insights of the last,
0:30:15 whatever decade is like, you need a ton of data,
0:30:17 which is easy if it’s words,
0:30:19 but hard if it’s metal, right?
0:30:20 – Yes.
0:30:24 So what we ended up doing was created a hybrid model.
0:30:28 We said, okay, what if we keep the humans in the loop?
0:30:30 So the human can give an instruction
0:30:32 initially based on heuristics.
0:30:35 And then we look at the data and human can adjust
0:30:37 and then iterate on that.
0:30:40 But while we are capturing all these data
0:30:42 and over time, as we’re capturing the data,
0:30:44 we start building the models
0:30:47 that will help the human do less trials, right?
0:30:50 It’s basically guided reinforcement learning, right?
0:30:52 You know, humans are actually guiding it where to go,
0:30:53 but it’s exploring those areas.
0:30:54 But after a while,
0:30:57 once we started forming thousands of parts,
0:30:59 then you can start feeding this data into a model.
0:31:02 Then the model will be like, okay, human,
0:31:04 you don’t need to do 25 different trials.
0:31:06 Now you can do with five trials,
0:31:07 you’re gonna get to the right place,
0:31:09 which is actually the number we are at right now.
0:31:11 – And that’s happening in the physical world,
0:31:12 largely those iterations,
0:31:14 like you’re trying on a piece of metal
0:31:15 and it’s bad and it tears
0:31:16 and you do another piece of metal
0:31:19 and it’s a little less bad and eventually you get it.
0:31:20 – Exactly, exactly.
0:31:22 And that initially would take 25 parts.
0:31:26 Like, you know, before we find a recipe for that design,
0:31:28 but 25 parts still was better than traditional alternative.
0:31:31 – When you say 25 parts, you mean 25 tries,
0:31:34 25 pieces of metal before you make the part the right way.
0:31:35 – Exactly.
0:31:37 And that was like, you know,
0:31:39 they would sit down basically 25 days in a row.
0:31:41 So in a month, they could actually define a recipe
0:31:43 where traditionally making a mold
0:31:45 would take at least three, four months, right?
0:31:46 So we were still better.
0:31:49 But then now, over time, when we generated the data
0:31:53 and now the model can tell the engineer,
0:31:55 okay, maybe you want to choose these parameters.
0:31:58 Is now becoming an advisor, we’re down to five trials.
0:32:01 So in five trials, we can actually get to the right part.
0:32:02 And then hopefully in the future,
0:32:04 we get to a point where, you know,
0:32:06 the machine will tell the robot’s what to do
0:32:09 and the human can be completely out of loop.
0:32:10 But the idea was like,
0:32:12 how do you kind of create that hybrid model that’s efficient
0:32:14 so that we can generate the data
0:32:18 until the model is good enough to do the job itself?
0:32:21 – And you find that the data is sort of generalizable.
0:32:24 I mean, clearly, like making one kind of part
0:32:26 makes the model, the AI,
0:32:28 smarter about making another kind of part.
0:32:30 – Yes, yeah, it is.
0:32:31 It’s kind of interesting.
0:32:33 I think people don’t think about it.
0:32:35 I used to do sheet shaping by hand, right?
0:32:37 That was one of the hobbies I had.
0:32:39 I was working with this shop in Pomona
0:32:41 and we were actually hammer sheets into shape.
0:32:42 And we used to say, you know,
0:32:45 if you spent five years doing it, you’re really good.
0:32:46 You got really good at it.
0:32:48 I was just to think, you know, okay,
0:32:50 after five years of doing this,
0:32:52 yes, you have this intuitive understanding of you.
0:32:52 Look at the sheet and be like,
0:32:54 okay, this place needs to be hammered more.
0:32:55 This place needs to be hammered.
0:32:58 Like it was, it was, it was, it was intuitive.
0:32:59 It was like, you couldn’t explain why you’re thinking
0:33:00 this did to happen.
0:33:02 There was no physical explanation.
0:33:03 None of these people who were sheet shaping
0:33:06 got PhDs in material science.
0:33:08 They just learned over time seeing the pattern
0:33:09 of how the sheet forms.
0:33:10 – That’s craftsmanship.
0:33:12 – That’s craftsmanship, right?
0:33:14 But really reminded me of, okay,
0:33:16 these people can know how to do it,
0:33:18 but without really being able to explain it,
0:33:19 they have to do it for five years.
0:33:22 – So that kind of tacit knowledge, yeah.
0:33:26 – And reminded me of the same challenge we had,
0:33:28 an early machine learning challenge where they were like,
0:33:31 okay, a human can look at two pictures and say,
0:33:33 okay, this is a cat and this is a dog.
0:33:35 Something happens in their brain that knows,
0:33:36 which is a cat, but they cannot really define
0:33:39 why they’re calling this a cat and this one a dog.
0:33:40 So that was where it started to click for me.
0:33:42 If I can capture enough data,
0:33:46 five years worth of data, right, of a human,
0:33:49 then I should be able to get to a very good sheet shaper,
0:33:52 right, and you know, it’s funny, back at the end,
0:33:54 but we’ll say, okay, humans are, you know,
0:33:57 receiving X amount of megabytes a second.
0:33:59 Okay, five years worth of data is that much.
0:34:02 So roughly, I think once we get to certain amount of data,
0:34:06 I think we have enough data to be able to basically replace,
0:34:08 like not replace the mentality
0:34:11 or the model that the sheet shaper has in their mind.
0:34:16 – So how many years of kind of human level craftsman
0:34:19 sheet shaping data does the model have at this point?
0:34:22 – Yeah, no, so I think last time I checked one year ago,
0:34:24 I checked around, we were like three fourths of the way there
0:34:27 in terms of the data that we have for just sheet shaping,
0:34:29 right, so once we get to, I think full, I think,
0:34:31 at least at that point we have no excuse,
0:34:32 we have enough data, the model should be good,
0:34:36 we just need to figure out why it’s not maybe four fourths.
0:34:40 – It is interesting to analogize it
0:34:42 to like human craftsmanship, right,
0:34:44 and I mean, even if you want to zoom out even more
0:34:48 the like 50 year history of AI where first everybody was like,
0:34:50 oh, you just got to teach the machine all the rules
0:34:53 for to use your example, like what’s a cat and what’s a dog,
0:34:55 but then you realize it’s actually wildly hard
0:34:56 to make a list of rules
0:35:00 that can reliably distinguish a cat from a dog,
0:35:04 and the weird thing that has happened in AI is like,
0:35:07 oh, you don’t actually have to make a list,
0:35:09 you just need like image net,
0:35:12 you just need like a giant database of images
0:35:15 and a giant neural network,
0:35:18 and you just throw it at it and say, figure it out,
0:35:21 and it figures it out, and you’re sort of doing that,
0:35:23 but for shaping metal.
0:35:25 – For shaping metal, and then the only challenge
0:35:28 was like cats and dogs pictures were on internet,
0:35:30 and sheet metal forming data wasn’t,
0:35:33 and so that was an additional problem we had to solve
0:35:35 as you pointed out, which is a big problem in physical AI.
0:35:40 – So I want to talk a little bit more about AI and robotics.
0:35:44 Jensen Wong’s been talking about it.
0:35:47 As I’m sure you know, NVIDIA’s VC arm
0:35:49 is an investor in your company.
0:35:51 Other people are working on what you’re working on.
0:35:55 I mean, I’m curious, what is the sort of AI and robotics
0:35:57 path look like to you for the next few years,
0:35:59 and what do you understand about it now
0:36:02 that you didn’t understand whatever five years ago?
0:36:04 Like what have you really come to realize
0:36:06 by working on it all the time?
0:36:09 – I think the biggest problem for physical AI
0:36:13 is data generation for now to train models.
0:36:15 So we need to either, there’s two things need to happen.
0:36:17 Either new types of models needs to be created,
0:36:21 new architectures, new algorithms, basically,
0:36:23 which I’m sure it’s going to happen.
0:36:25 – That can learn more with less data, basically.
0:36:30 – And same way humans kind of learn more with less data, right?
0:36:35 But at the same time, I think we only exposed our models
0:36:40 categorically to 10% of type of data that humans receive.
0:36:43 You think about human interactions.
0:36:45 You and I are now talking.
0:36:47 If it was AI, AI is probably only listening
0:36:50 to the words we’re saying, right?
0:36:51 But that’s only 10% of communication.
0:36:53 I can see your lips moving.
0:36:55 I can see your eyebrows moving.
0:36:57 I can see like maybe you’re folding your arms and okay,
0:36:59 I know that like, okay, maybe there’s all these,
0:37:01 90% of the signals are not captured.
0:37:04 That’s used for learning.
0:37:08 You look at, if you ask chat GPT or Dolly
0:37:12 or any of the even, even, you know, Grock say,
0:37:17 okay, draw me a clock that shows 5.30.
0:37:18 It cannot show you, draw you a clock.
0:37:20 It will draw you a clock, but it doesn’t show 5.30.
0:37:23 Actually, most of the time it shows 10.10.
0:37:25 – 10.10, ’cause that’s where watch hands,
0:37:27 like analog watch hands look good, right?
0:37:28 It’s a nice little V.
0:37:30 – Yeah, because those are all the images
0:37:32 that’s seen on internet because they’re watching.
0:37:33 – It’s almost always 10.10.
0:37:35 It’s the classic watch photo.
0:37:36 – It’s like 5.30 is also 10.10.
0:37:38 ‘Cause I don’t know– – It’s always 10.10, right?
0:37:41 – It’s a little bit to a generative AI.
0:37:42 It’s always 10.10 somewhere.
0:37:45 – So, I think, but that humans, you know,
0:37:47 receive this data of movement when you grow up,
0:37:49 you look at the clock on the wall as a kid,
0:37:51 you’re like, okay, now I intuitively get it.
0:37:52 And I think I know what’s going on.
0:37:53 So I can actually make it work.
0:37:56 So, even though we trained it on a lot of data,
0:37:58 I don’t think we trained it on the right categorically,
0:38:00 right data yet, right?
0:38:03 To get all the intuitive understanding
0:38:04 that we have today.
0:38:06 So I think we have a data problem
0:38:08 and that exists in physical AI.
0:38:10 So I think the applications will win.
0:38:11 There’s a lot of people are working in this.
0:38:13 I think the applications will win
0:38:16 who can either synthetically generate that data
0:38:19 or they can actually scale in the physical world
0:38:21 in a way where they can actually generate the data
0:38:25 for themselves, but the scaling needs to happen
0:38:26 with less data.
0:38:28 And I think that was, that’s why I’m like,
0:38:30 for example, like very bullish on manufacturing.
0:38:33 So I think the data is gonna be the biggest challenge.
0:38:35 And I think, you know, in order for us
0:38:37 to massively change this space,
0:38:39 we need to be able to get to the data.
0:38:42 I don’t think algorithms is a bottleneck there yet.
0:38:43 It’s just the data for us.
0:38:48 – And is it just a matter of people doing what you’re doing
0:38:52 and like finding little wedge places to start
0:38:54 and having people sort of hold the hand of the model
0:38:55 and training up the models?
0:39:00 I mean, that seems slow on a certain level.
0:39:02 Like not, you know, obviously it’s working for you,
0:39:05 but like, is there some kind of breakthrough move
0:39:06 that people can make?
0:39:09 Can you put sensors somewhere in the world to, you know,
0:39:12 train AI without having to, you know,
0:39:14 have a human stand next to it
0:39:17 as it messes up one piece of sheet metal after another?
0:39:21 – Yeah, I think there is another path
0:39:23 which is simulation path.
0:39:26 Make physics-based simulations faster
0:39:29 and kind of learn, let the robots just go play
0:39:31 in a digital playground as opposed to deploying
0:39:32 in real world.
0:39:34 And that becomes a computation problem.
0:39:35 And then, you know, as long as you have enough computation,
0:39:37 you can train the robots.
0:39:39 But I think, you know, I think, you know,
0:39:41 the good examples that we have had success so far
0:39:43 is like autonomous cars, right?
0:39:45 Did the same thing we were doing, but in the car,
0:39:48 like, okay, Tesla, you know,
0:39:50 deploy the fleet of robots that are capturing data,
0:39:52 still be driven by humans,
0:39:55 but the data can be used later on to kind of automate it.
0:39:56 – I mean, that’s an interesting case
0:40:00 because it has been much harder clearly
0:40:03 than many people thought, maybe most people thought, right?
0:40:05 Like, I know that’s a particular instance
0:40:08 where you’re really worried about edge cases.
0:40:12 I don’t know, is autonomous cars like a good model or not?
0:40:13 It seems complicated.
0:40:15 – I think the model of capturing data is there,
0:40:19 but then the task at hand is very hard, right?
0:40:21 So, I think that’s the challenge, right?
0:40:23 So, where it says, like with us,
0:40:25 it’s still a much more structured environment.
0:40:27 And I think that was a thinking where I think,
0:40:30 I think the hardest problem right now in physical AI
0:40:32 is finding the business model
0:40:34 of how do you scale data capture
0:40:36 without requiring billions of dollars in investment.
0:40:38 – So, what are you making today?
0:40:39 I imagine you know.
0:40:43 – So, last time I checked in the facility one,
0:40:46 four of the cells are working on a defense application.
0:40:48 – Is it secret?
0:40:50 Can you tell me what it is?
0:40:51 – It’s a missile.
0:40:55 And two of them were working on an aerospace application.
0:40:58 This is components of an aircraft or a drone.
0:41:01 And one of them, this is an interesting one,
0:41:03 was working on an architectural component,
0:41:07 which is a roof tile for a specific building
0:41:09 that’s used by the Department of,
0:41:11 by Bureau of Water Recognition.
0:41:12 – Oh, I was gonna say, what is it,
0:41:17 something like Frank Gehry nightmare weirdo metal part?
0:41:19 – Well, those we have had those in the past too,
0:41:22 but this one is actually very practical.
0:41:24 It’s this building, it’s actually very interesting.
0:41:26 Exactly, these buildings, these large industrial buildings
0:41:30 that built, they built in the ’60s or ’50s.
0:41:31 And they use these type of roof tiles
0:41:34 that the manufacturer doesn’t exist anymore.
0:41:36 And anybody else who they went to,
0:41:39 quoted them hundreds of thousand dollars to make those tiles.
0:41:41 And we were like, oh, no, we can make it for you.
0:41:43 And, but also that shows kind of the diversity.
0:41:44 I mean, like I say, in the morning,
0:41:47 we have like aerospace parts in the afternoon,
0:41:52 roof tiles for a industrial complex for, you know, for a dam.
0:41:54 – Now you’re in the sheet metal business.
0:41:57 I know your large dream is much larger than that, right?
0:42:02 But like, what, like, tell me where you are now.
0:42:05 Tell me where you are now.
0:42:05 Like what are you doing?
0:42:06 What are you selling?
0:42:09 And then kind of what’s the next big step?
0:42:12 – So some of our systems are now operating out in the wild
0:42:15 and working for the customers.
0:42:18 And, but I think the next phase of growth for us
0:42:21 is getting into each of these applications
0:42:22 and own more of the process.
0:42:25 So we can teach the robot craftsman the future processes,
0:42:28 not just sheet forming, but also maybe how to assemble it,
0:42:31 how to weld it, how to surface finish it, right?
0:42:35 So what we are doing now in the next phase is actually,
0:42:39 instead of selling parts or components or systems,
0:42:40 we’re actually saying, okay,
0:42:44 can we get this robot craftsman
0:42:46 to actually build a US sub-assembly or a full product,
0:42:49 not just a component of it, but a full product.
0:42:51 So that’s something we’re describing with folks.
0:42:53 Can we have the robot craftsman
0:42:54 build the full drone for you?
0:42:59 Can we have the robot craftsman build you a full missile,
0:43:02 as opposed to just build missile, you know, missile scans?
0:43:05 – Is there, that seems like a leap.
0:43:07 Is there not an intermediate step?
0:43:08 – Yes.
0:43:09 – Yeah.
0:43:10 – So, I mean, how we are doing it
0:43:13 is we’re gradually stepping into it, right?
0:43:15 The same way sheet metal was our first application.
0:43:18 So we’re putting a facility that maybe makes drones,
0:43:22 but the main component that we automate today
0:43:24 is sheet forming, which is the bottleneck.
0:43:27 And then we do the welding in a traditional way
0:43:30 on the same robots, but we actually instruct them to do it.
0:43:31 – All right.
0:43:33 – So that way, the robot is kind of back
0:43:36 where it was on sheet metal five years ago,
0:43:38 but it’s learning how to weld now.
0:43:39 – Exactly.
0:43:42 I used to work in a shop that we will do custom cars,
0:43:44 build custom cars with hand.
0:43:47 And so it was also near and near to my heart.
0:43:50 So what we realized is that with our technology
0:43:54 for the first time, we can actually enable
0:43:56 a product that didn’t exist in automotive,
0:44:00 meaning that instead of buying a car that’s mass produced
0:44:02 and every single one of them looked the same,
0:44:06 you can now let the customer design a custom car for them.
0:44:07 You know, right now, if you go buy a car,
0:44:11 you have options of what the seat color would be,
0:44:13 or maybe the color of the car would be
0:44:15 and what some trim options,
0:44:17 but you can’t really choose the design of your car.
0:44:19 You can’t say, oh, I want a different hood,
0:44:21 and I want a different fender
0:44:23 because going to back same problem,
0:44:25 you don’t have to make tooling and mold
0:44:28 for the fender of a certain designs
0:44:29 and you cannot easily change it.
0:44:30 So with our technology, you can.
0:44:34 So what we started doing was like, okay,
0:44:37 applying this freedom that this technology provides
0:44:40 to now automotive is ability of the customers
0:44:41 to be able to go to a website,
0:44:45 design a fully customized car for themselves.
0:44:47 It can be either from already designed panels
0:44:52 from car designer or adding a specific customizations
0:44:53 they want to do, for example,
0:44:55 logo of their company to their door of the car
0:44:56 or the hood of the car
0:44:59 and actually get a completely unique car, right?
0:45:01 Manufactured for them.
0:45:03 And we’re actually working with this
0:45:04 with some of our automotive partners,
0:45:07 automotive OEMs as well, right?
0:45:09 We actually showed some of this work
0:45:11 in the biggest aftermarket show in the United States
0:45:14 called SEMA with our partner Toyota.
0:45:17 So I think this is gonna be, in my opinion,
0:45:20 one of the new product categories in automotive.
0:45:22 We have had autonomous cars,
0:45:24 we have had, you know, electric cars.
0:45:26 And I think now for the first time
0:45:27 with technologies like ours,
0:45:29 you can have custom to order cars,
0:45:31 like all cars that are like, you know,
0:45:33 the same way you choose what T-shirt you wear
0:45:35 and your T-shirt is different than mine,
0:45:37 we also don’t have to drive the same, you know,
0:45:39 Model S or, you know, Model 3.
0:45:41 We can actually have our own customized
0:45:43 Model 3s and Model Ss.
0:45:45 – So what’s the, I mean, is that the,
0:45:48 if you think sort of long-term for Machina,
0:45:49 is like that what you think about?
0:45:53 Like, give me the five-year vision.
0:45:54 – Yeah.
0:45:55 – Or 10-year or whatever.
0:45:58 – Yeah, so I think the long-term motivation
0:46:00 behind our company is,
0:46:02 can it create this democratization
0:46:05 of ideas for people who wanna build anything, right?
0:46:06 Can I express myself?
0:46:09 If I’m a builder, can I go build something
0:46:11 without having to build a factory for it?
0:46:13 So that’s really the long-term goal.
0:46:17 So I imagine in the next five to 10 years,
0:46:20 you can, as a designer, somebody who has an idea,
0:46:25 you can go to a website, get guided through your ideas
0:46:28 on how to make and design a physical product,
0:46:31 hit a button and say, okay, I want 20 of these
0:46:35 and I want them in Chatsworth, California
0:46:38 and the right facility programs the right number of robots
0:46:39 to actually do those operations
0:46:41 without any hardware or investment
0:46:44 that needs to be made for those specific parts
0:46:46 and ship it to you two days later in the right location.
0:46:49 That is the future we’re building towards.
0:46:53 Cars is just one of the products that could be built.
0:46:56 But I imagine that this technology or technology like these,
0:46:58 technologies like these can be used to do
0:46:59 the myriad of designs.
0:47:03 I think the moment you open up this possibility
0:47:05 of any designs could be a reality.
0:47:07 I think so many things will be created
0:47:09 that we’re not even thinking of right now.
0:47:12 You know, the fact that we have cars today
0:47:13 and they’re all looked the same
0:47:15 is limitation of technology.
0:47:18 But the moment you can open up this creativity
0:47:20 of turning ideas into physical reality
0:47:24 without an initial investment or a huge barrier to entry,
0:47:26 then I think we’re gonna have all kinds of drones,
0:47:28 all kinds of satellites, all kinds of rockets,
0:47:31 all kinds of cars, then you’re gonna be just like,
0:47:34 you know, Cambrian explosion of different designs
0:47:35 that’s gonna come into our world.
0:47:37 And I think that’s what future is about.
0:47:38 The future is about, you know,
0:47:39 I call like future is custom.
0:47:41 Like future is about being able to make
0:47:43 these all these ideas in reality.
0:47:45 We had this explosion happening in digital world.
0:47:46 – Yeah.
0:47:48 – You know, now we have even models generating images
0:47:52 and videos and there’s this, you know,
0:47:55 explosion of different ideas and content being created
0:47:58 using the technology, but the link is broken
0:47:59 to the physical world.
0:48:01 And the physical work is still pretty uniform
0:48:03 ’cause it’s very hard to make things in physical work.
0:48:05 Can we bridge that gap?
0:48:08 Can we connect the digital world of creation
0:48:09 to physical world of creation
0:48:12 and create the same variety in the physical world
0:48:14 as we have in the digital world?
0:48:16 I think that’s the goal in our company.
0:48:19 (upbeat music)
0:48:22 We’ll be back in a minute with the lightning round.
0:48:24 (upbeat music)
0:48:28 – Hey, it’s Jacob.
0:48:30 I’m here with Rachel Botsman.
0:48:34 Rachel lectures on trust at Oxford University.
0:48:37 And she is the author of a new Pushkin audio book
0:48:40 called How to Trust and Be Trusted.
0:48:41 Hi, Rachel.
0:48:42 – Hi, Jacob.
0:48:46 – Rachel Botsman, tell me three things
0:48:47 I need to know about trust.
0:48:52 – Number one, do not mistake confidence for competence.
0:48:53 Big trust mistakes.
0:48:55 So when people are making trust decisions,
0:48:59 they often look for confidence versus competence.
0:49:04 Number two, transparency doesn’t equal more trust.
0:49:05 Big myth and misconception.
0:49:08 And a real problem actually in the tech world.
0:49:11 The reason why is because trust
0:49:14 is a confident relationship with the unknown.
0:49:17 So what are you doing if you make things more transparent?
0:49:20 You’re reducing the need for trust.
0:49:25 And number three, become a stellar expectation setter.
0:49:30 Inconsistency with expectations really damages trust.
0:49:31 – I love it.
0:49:33 Say the name of the book again
0:49:35 and why everybody should listen to it.
0:49:38 So it’s called How to Trust and Be Trusted.
0:49:40 Intentionally, it’s a two-way title
0:49:44 because we have to give trust and we have to earn trust.
0:49:47 And the reason why I wrote it is because we often hear
0:49:49 about how trust is in a state of crisis
0:49:51 or how it’s in a state of decline.
0:49:54 But there’s lots of things that you can do
0:49:56 to improve trust in your own lives,
0:49:58 to improve trust in your teams,
0:50:01 trusting yourself to take more risks
0:50:04 or even making smarter trust decisions.
0:50:05 – Rachel Botsman, the new audio book
0:50:08 is called How to Trust and Be Trusted.
0:50:09 Great to talk with you.
0:50:10 – It’s so good to talk with you
0:50:12 and I really hope listeners listen to it
0:50:15 because it can change people’s lives.
0:50:19 – Let’s finish with the lightning round.
0:50:23 Do you drive a customized car?
0:50:26 – I don’t actually yet.
0:50:27 Well, if I am–
0:50:29 – What have I seen on your Instagram?
0:50:31 What’s that truck you keep posting on your Instagram?
0:50:33 – So I have a truck that’s customized.
0:50:36 I don’t drive it around as much.
0:50:38 But maybe this year, I’ll start taking it out this year.
0:50:40 We have been kind of stealth about it.
0:50:41 We’ve been talking about it,
0:50:43 but we haven’t talked about it in a big way
0:50:45 because we have a big release coming soon.
0:50:47 – I mean, you’re literally posting it on Instagram.
0:50:49 It’s not that stealth.
0:50:52 Tell me about that truck you keep posting on Instagram.
0:50:53 What’s going on with that?
0:50:54 – So it’s a truck that’s fully,
0:50:56 the full body is fully customized.
0:50:59 – It says Anvil in the back when you post it.
0:51:00 Is it called Anvil?
0:51:00 – Dumb question.
0:51:02 – We call it Anvil.
0:51:04 I think the idea was that actually the shape,
0:51:05 design of it was inspired by Anvil.
0:51:07 If you look at the front fender,
0:51:10 it actually looks like the front bumper looks like an Anvil.
0:51:12 But also the idea is that like,
0:51:14 we are actually forming sheets on an Anvil.
0:51:16 So it was very fitting.
0:51:17 Tell me about that truck.
0:51:19 Like just tell me what’s it look like?
0:51:20 – Yeah. So for example, like,
0:51:24 we put a lot of form and sharp edges in the hood, right?
0:51:27 Most vehicles have a very hard time.
0:51:30 If you look at most of the hood of the vehicles,
0:51:32 they’re very smooth because it’s very hard
0:51:35 to actually put sharp angles in the hood.
0:51:36 So if you look at this truck,
0:51:37 this truck has a lot of angles,
0:51:41 a lot of sharp detail right in the hood, right?
0:51:44 And that’s very expressive of the type of person,
0:51:46 for example, that I am, right?
0:51:47 I like things that are edgy.
0:51:50 And that truck is certainly edgy, right?
0:51:53 It’s bare metal, right?
0:51:58 There is no blemishes being hidden under the vehicle.
0:52:02 A lot of people, when Cybertruck came out,
0:52:04 we got very excited about, you know,
0:52:06 oh, it’s bare metal, it looks like a metal,
0:52:07 but then there was no form in it
0:52:08 because it’s actually very hard
0:52:11 to make a form metal look nice.
0:52:13 And so that’s one of the things we wanted to show.
0:52:14 We wanted to show that, okay,
0:52:17 you can actually have a form metal with a lot of detail in it
0:52:20 and still keep it bare metal because it will look nice, right?
0:52:24 So yeah, a lot of design features of it for me
0:52:26 kind of represents the type of personality character
0:52:29 that I have, but I think that’s how every car should be.
0:52:31 You know, people should be able to have that freedom
0:52:33 to choose what their cars look like.
0:52:37 – How many skull tattoos do you have?
0:52:39 – I’ve got three.
0:52:41 – Why?
0:52:46 – So yeah, it’s an interesting thing.
0:52:50 So a skull for me represents kind of,
0:52:54 and it’s an abstract for death of ego.
0:52:56 So I have a tattoo on my thumb,
0:52:59 which is a skull that’s holding a microphone
0:53:00 into his ears.
0:53:02 And this was a time where, you know,
0:53:04 I felt like, you know, I had a good platform
0:53:07 and I could talk a lot and people would listen,
0:53:09 but then I realized I should, yes, that’s great,
0:53:12 but I should maybe keep the mic close to my ears
0:53:16 and also listen as opposed to talk all the time, right?
0:53:18 So I think a skull–
0:53:21 – Microphones don’t work that way for the record,
0:53:23 but I like it as a metaphor.
0:53:27 – Exactly, but I think the idea was around, you know,
0:53:28 kind of reminders.
0:53:30 You can see a lot of my tattoos are on my hands,
0:53:34 so it’s really a reminder for myself to know that, you know,
0:53:37 be present and make sure that, you know,
0:53:39 you’re not involved with your ego too much
0:53:42 and you can see other people’s perspective.
0:53:45 – Is there any tension between ego death and custom cars?
0:53:50 – Tension between ego death and custom cars.
0:53:52 – I don’t know, I’m just playing, but like, you know,
0:53:55 custom car kind of seems like, hey, look at me, I’m special.
0:53:58 And ego death seems like, oh, don’t look at me,
0:53:59 I’m not so special.
0:54:01 – Yeah, no, I think the difference is,
0:54:04 I think, yeah, if you have attachment to your custom car,
0:54:06 then maybe there’s tension,
0:54:09 but I more think of it in terms of expression, right?
0:54:10 You know, you can be an artist.
0:54:14 You can design your home the way it expresses you.
0:54:17 You can design the theme of your podcast
0:54:18 the way it expresses you.
0:54:21 You can design your car also the way it expresses you.
0:54:23 I think it’s less so about, oh, look at me, I’m special.
0:54:26 It’s more like, here’s my expression to the world
0:54:28 for the people to see.
0:54:31 But I think that expressiveness is pretty amazing.
0:54:33 I think that’s uniquely one of the unique things
0:54:36 about humans that like, you know, we,
0:54:37 I think all we do when we come to this world
0:54:39 is expressing ourselves, right?
0:54:41 Expressing ourselves through our work,
0:54:43 expressing through ourselves, through our relationships.
0:54:47 And if we can enable people to express themselves better,
0:54:49 better, I think that’s great.
0:54:51 But if you get attached to your expressions
0:54:53 and your ideas and your thoughts,
0:54:55 and you think, oh, I’m better than everybody else,
0:54:56 then I think that that becomes,
0:54:59 that becomes a little bit of an ego-driven trip.
0:55:02 (upbeat music)
0:55:10 – Edward Mayer is the co-founder and CEO of Mockin’ The Labs.
0:55:13 Today’s show was produced by Gabriel Hunter Chang.
0:55:15 It was edited by Lydia Jean Cotte
0:55:18 and engineered by Sarah Bruginger.
0:55:21 You can email us at problem@pushkin.fm.
0:55:24 I’m Jacob Goldstein and we’ll be back next week
0:55:26 with another episode of What’s Your Problem?
0:55:34 – Hey, it’s Jacob.
0:55:35 I’m here with Rachel Botsman.
0:55:39 Rachel lectures on trust at Oxford University
0:55:43 and she is the author of a new Pushkin audiobook
0:55:46 called How to Trust and Be Trusted.
0:55:47 Hi, Rachel.
0:55:48 – Hi, Jacob.
0:55:51 – Rachel Botsman, tell me three things
0:55:53 I need to know about trust.
0:55:58 – Number one, do not mistake confidence for competence.
0:55:59 Big trust mistakes.
0:56:01 So when people are making trust decisions,
0:56:05 they often look for confidence versus competence.
0:56:09 Number two, transparency doesn’t equal more trust.
0:56:11 Big myth and misconception.
0:56:13 And a real problem, actually in the tech world,
0:56:16 the reason why is because trust
0:56:20 is a confident relationship with the unknown.
0:56:23 So what are you doing if you make things more transparent?
0:56:26 You’re reducing the need for trust.
0:56:31 And number three, become a stellar expectation setter.
0:56:36 Inconsistency with expectations really damages trust.
0:56:37 – I love it.
0:56:38 Say the name of the book again
0:56:41 and why everybody should listen to it.
0:56:43 – So it’s called How to Trust and Be Trusted.
0:56:46 Intentionally, it’s a two-way title
0:56:49 because we have to give trust and we have to earn trust.
0:56:53 And the reason why I wrote it is because we often hear
0:56:55 about how trust is in a state of crisis
0:56:57 or how it’s in a state of decline.
0:56:59 But there’s lots of things that you can do
0:57:02 to improve trust in your own lives,
0:57:04 to improve trust in your teams,
0:57:06 trusting yourself to take more risks
0:57:09 or even making smarter trust decisions.
0:57:11 – Rachel Botsman, the new audio book is called
0:57:14 How to Trust and Be Trusted.
0:57:14 Great to talk with you.
0:57:16 – It’s so good to talk with you.
0:57:18 And I really hope listeners listen to it
0:57:21 because it can change people’s lives.
0:00:07 Pushkin.
0:00:12 – I’m Dr. Laurie Santos, and to welcome the new year,
0:00:13 my podcast, The Happiness Lab,
0:00:16 is releasing a series of happiness how-to guides
0:00:18 to help you in 2025.
0:00:20 I’ll distill the wisdom of world-class experts
0:00:22 into easy to digest actionable tips,
0:00:24 struggling with tough emotions.
0:00:25 We have a how-to guide.
0:00:27 Worried that you’re not enough?
0:00:28 We got you.
0:00:30 Self-obsessed and want to get over yourself,
0:00:32 there’s a guide for that too.
0:00:35 The Happiness Lab’s how-to season starts January 1st.
0:00:37 Listen on the iHeart radio app, Apple Podcast,
0:00:39 or wherever you get your podcast.
0:00:48 – AI works amazingly well.
0:00:52 It works terrifyingly well even for virtual things,
0:00:56 for words, for pictures, for videos.
0:01:00 This is true in large part because of the internet.
0:01:03 The internet provides this wildly abundant,
0:01:07 readily available source of words, pictures, and videos
0:01:10 to train AI models.
0:01:13 But there is no analogous wildly abundant,
0:01:17 readily available data set for the physical world.
0:01:22 There is no gargantuan internet-like repository of data
0:01:25 that describes how things move and bend
0:01:29 and break in real physical space.
0:01:33 And as a result, we do not yet have robust AI
0:01:35 for the physical world.
0:01:37 But people are working on it.
0:01:42 And if they succeed, they’ll change the way the world works.
0:01:45 Not just the world as it appears on our screens,
0:01:47 but the actual physical world.
0:01:51 The world where if you drop something on your foot, it hurts.
0:01:54 (upbeat music)
0:01:57 – I’m Jacob Goldstein, and this is What’s Your Problem,
0:01:59 the show where I talk to people who are trying
0:02:02 to make technological progress.
0:02:05 My guest today is Edward Mayer.
0:02:08 He’s the co-founder and CEO of Machina Labs.
0:02:10 Edward’s problem is this.
0:02:15 How can you use AI to turn robots from dumb,
0:02:20 inflexible machines into skilled, versatile craftsmen?
0:02:22 Before he started Machina Labs,
0:02:24 Edward worked in the rocket ship business,
0:02:29 first at SpaceX, and then at a company called Relativity Space.
0:02:31 And in the rocket business,
0:02:35 Edward saw firsthand the problems of traditional manufacturing.
0:02:37 It’s the kind of problem he’s now trying to solve
0:02:39 with AI and robots.
0:02:43 It’s a problem he calls the rigid factory problem.
0:02:49 So I’ve heard you use this phrase that’s interesting to me,
0:02:50 and it’s the real thing.
0:02:54 And it’s interesting to me, and it’s the rigid factory problem.
0:02:56 What’s the rigid factory problem?
0:03:00 – The main problem with the factories today
0:03:02 is that rigidity, meaning that if you have to build
0:03:06 a physical product, you pretty much have to build a factory
0:03:09 that’s designed for it and built for it.
0:03:12 There’s a lot of components that goes into the factory,
0:03:15 from machinery all the way to the tooling
0:03:16 that is required to build products
0:03:19 that are specifically designed for the geometry,
0:03:22 for the material that you’re trying to use.
0:03:24 The moment you want to change that,
0:03:25 you have to change your factory,
0:03:26 which is a huge investment.
0:03:30 I always give an example from when I was at SpaceX.
0:03:33 You think of SpaceX as a very innovative company,
0:03:36 and it is on the edge of a hardware space
0:03:38 in terms of innovation.
0:03:43 In the past 24 years, 23, 24 years that they have existed,
0:03:45 they have two rocket families.
0:03:48 It’s a Starship and there’s Falcon, right?
0:03:50 Because the moment you decide on diameter of,
0:03:53 for example, Falcon 9 or the Falcon family in general,
0:03:55 the diameter of that core, it’s very hard to change it.
0:03:56 A lot of tooling and machinery
0:03:59 specifically built for that diameter.
0:04:02 And that’s why for Starship, they had to start from scratch.
0:04:04 – Start from scratch, meaning like not just design,
0:04:06 but like the factory itself.
0:04:09 Like they actually had to build a whole new factory
0:04:12 ’cause they wanted to make a different size rocket.
0:04:14 – Yes, different size, different material.
0:04:15 All the tooling has to change, right?
0:04:19 Almost, yeah, you have to basically assume,
0:04:21 building from scratch, ground up factory,
0:04:22 what does it need to be there
0:04:24 for us to build this new product?
0:04:28 – I heard you describe, was this from your own experience,
0:04:31 that sort of era at SpaceX
0:04:36 when the fact that you couldn’t make the rocket wider
0:04:39 led to all these kind of difficult things
0:04:40 people were trying to do to be like,
0:04:42 how can we do all these things
0:04:44 under this fundamental constraint?
0:04:46 Like can you talk a little bit about that?
0:04:48 – Yeah, this is a lot of conversation happening
0:04:52 in 2012, 2013, 2014 time
0:04:57 when the diameter of the Falcon 9 could not get any larger.
0:05:00 And if you look at actually different Falcon versions,
0:05:02 the height of that vehicle kept going higher,
0:05:04 but the diameter could not change.
0:05:07 So it was about, what space can you find
0:05:11 to put new features and new designs
0:05:13 that exist within the vehicle?
0:05:15 So there was a lot of stuff basically being crammed
0:05:17 into the space that you have already got.
0:05:19 – So that’s true for building rockets.
0:05:21 I mean, what are some other,
0:05:24 just different kinds of manufactured products
0:05:26 where that kind of rigidity is a problem?
0:05:29 – Yeah, I think it is just common,
0:05:31 almost in all manufacturing.
0:05:32 That’s why this phenomenon,
0:05:35 I think it’s kind of funny, people take it for granted
0:05:37 that a thing called economies of scale.
0:05:42 Like people take it for granted as if it’s rule of nature.
0:05:43 It’s actually not.
0:05:46 – And just to be clear, it’s basically the more you build
0:05:48 of a thing, the cheaper each one of those things gets.
0:05:49 If you build one, it’s really expensive.
0:05:52 If you build a million, each one’s a lot cheaper.
0:05:53 – Yes, exactly.
0:05:54 But then people don’t think about it.
0:05:56 It’s like, oh, okay, intuitively makes sense.
0:05:56 But why?
0:05:58 It’s actually not that intuitive.
0:06:01 It’s actually a limitation of technology, right?
0:06:05 Why economies of scale is a thing
0:06:08 is because you have to make a huge amount of investment
0:06:09 to make the first thing.
0:06:11 And the moment you make the second thing
0:06:14 and the third thing, then you can break even your investment
0:06:17 onto more products that you’re gonna come out of it.
0:06:19 But that’s only true if the second product
0:06:22 can be built for the first investment.
0:06:24 You have to turn this concept and say, oh, this is a given.
0:06:26 This is an axiom of the world,
0:06:28 that the economies of scale is a thing.
0:06:31 But in reality, it’s a technological challenge, right?
0:06:34 It means that you build a car, you ask what application?
0:06:36 You, once you build a factory for a car
0:06:40 and all the toolings and dyes that goes into stamping
0:06:42 of the panels of that car,
0:06:46 that’s $150 million investment just for stamping.
0:06:47 This is a number, for example, from Tesla.
0:06:50 Tesla spends $150 million in a stamping plant
0:06:54 they have in a factory in Texas, right?
0:06:57 And that can only make Model Y or Model 3, right?
0:06:59 The moment you have to change that,
0:07:04 that means go through every 80 to 130 sheet metal panels
0:07:08 that exist on that car and design a new tool for it.
0:07:10 And each of those tools is gonna be a few hundred thousand
0:07:12 dollars to sometimes a million dollars
0:07:14 or a million and a half million dollars.
0:07:17 And you’re talking about like 80 to 130 tools per vehicle.
0:07:18 – And like all you’re doing,
0:07:20 you’re not reinventing the car there.
0:07:23 You’re just making a car that’s a slightly different shape.
0:07:26 – Yes, maybe you may get sedan
0:07:28 and you’re not trying to do a slightly longer version,
0:07:30 a slightly bigger version.
0:07:32 And that’s why economies of scale are the thing.
0:07:34 You’re saying, okay, I made a factory.
0:07:38 Now it only pays back if I make a million of this car, right?
0:07:41 Because I had to just drop $150 million
0:07:42 on just a stamping plant.
0:07:44 So yeah, it’s all over manufacturing.
0:07:47 We abstract this whole concept and gave it a name
0:07:49 says economies of scale.
0:07:53 – Yeah, so you left SpaceX and you went to
0:07:55 Relativity Space, right?
0:07:57 A company that was also in the space business
0:07:59 that was using 3D printing, right?
0:08:01 That was the idea of the company,
0:08:04 which seems like an approach to this problem
0:08:05 that you’re talking about, right?
0:08:10 An advantage of 3D printing is that it is much more flexible
0:08:12 and less rigid than traditional manufacturing, right?
0:08:14 So tell me about that.
0:08:18 – Yeah, so yeah, we saw this challenge at SpaceX
0:08:20 and I joined Relativity very early on.
0:08:22 Well, I was the fourth person on that team.
0:08:26 And the goal over there was, okay,
0:08:29 let’s just think about this fundamentally.
0:08:32 Can we build a rocket that all built
0:08:33 with flexible technology?
0:08:37 And at the time, 3D printing was a forefront
0:08:40 of everybody’s minds because people were already starting
0:08:42 to build at NASA, at SpaceX.
0:08:44 People were already starting to build engines
0:08:46 out of 3D printing.
0:08:47 And the concept was like, well, that’s great.
0:08:48 It’s very flexible.
0:08:51 3D printing has this promise of geometry agnostic,
0:08:52 material agnostic.
0:08:53 You can just feed it a design and you can build
0:08:55 a product for it.
0:08:57 And it worked very well with rocket engines.
0:09:00 I think probably in the future, all rocket engines
0:09:02 would be 3D printed.
0:09:06 And the concept was, can we take this
0:09:08 and scale it to the whole vehicle, right?
0:09:10 Can we build the whole vehicle with a process
0:09:14 like 3D printing so that it is flexible?
0:09:16 Today, if you want to build a rocket
0:09:18 with 12 foot diameter, we can do it.
0:09:20 And then if our calculation changes
0:09:21 and we wanted to go to another orbit
0:09:23 or do a different type of emission,
0:09:25 then we can change that 12 foot diameter
0:09:27 to 20 diameter, 20 feet diameter.
0:09:28 – Don’t have to build a new factory.
0:09:30 Don’t have to build new machinery,
0:09:32 just use the same 3D printer, yeah.
0:09:34 – Exactly.
0:09:37 So that was the concept behind relativity.
0:09:39 That’s the thesis behind relativity.
0:09:39 And that was the goal there.
0:09:42 The goal was 3D print the whole rocket
0:09:44 so that you can be flexible.
0:09:48 – But it hasn’t worked at least in the kind of maximalist
0:09:49 version, right?
0:09:50 Like they just haven’t been able to do it.
0:09:53 They’ve sort of backed off of that big dream,
0:09:54 as I understand it.
0:09:55 – Yeah, yeah.
0:10:00 So I think the challenge was that 3D printing
0:10:02 is just one process.
0:10:07 And it’s necessarily not good for every type of part.
0:10:09 Manufacturing is very versatile.
0:10:11 You do different types of geometries,
0:10:12 different types of material.
0:10:14 And 3D printing has a very small reach.
0:10:17 There’s certain type of parts, like rocket engines,
0:10:18 very good fit.
0:10:22 You’re building a tank, maybe not so, right?
0:10:24 So yeah, it’s good for certain type of parts,
0:10:25 but there’s a whole lot of other parts.
0:10:28 Like I said, you’re building a fuel tank,
0:10:30 which is a basically large sheet metal
0:10:33 or thin walled structure.
0:10:35 Then maybe 3D printing is not as good of a fit
0:10:36 because it takes a long time.
0:10:38 And also because it’s thin,
0:10:40 you have a lot of physical challenges
0:10:43 in terms of controlling the geometry and the tolerances.
0:10:48 So we realize soon that maybe other processes
0:10:51 are also need to be automated
0:10:53 the same way 3D printing is.
0:10:56 You need to have more flexible processes
0:10:57 that are not just one process,
0:10:59 more flexible platforms that can do different types
0:11:02 of processes, not just 3D printing,
0:11:07 to be able to cover a whole variety of products
0:11:09 in a flexible manner, the same way the 3D printing,
0:11:12 that’s for certain type of products.
0:11:14 And that was actually the thinking behind Machina Labs
0:11:17 is that, okay, can we step back and say,
0:11:18 what do we need to build?
0:11:20 What is this flexible platform
0:11:23 that can do 3D printing if needed,
0:11:25 or can do sheet forming if it’s needed,
0:11:27 it can do machining if it’s needed,
0:11:29 but chooses the right operation,
0:11:31 right flexible operation for the right part,
0:11:34 but still very agile and doesn’t require a lot of tooling
0:11:35 and it’s not inflexible.
0:11:39 – So it’s sort of zooming out more.
0:11:42 It’s saying 3D printing is not gonna do everything.
0:11:44 The way manufacturing works now,
0:11:47 it’s just too rigid, too hard to change things,
0:11:50 too reliant on scale to make the economics work out.
0:11:55 So like that’s a very big, very abstract thought.
0:11:59 To start a company, you gotta make something,
0:12:02 or you gotta make something that makes something.
0:12:04 Like what do you actually do?
0:12:07 – Yeah, so it was interesting, right?
0:12:11 You know, actually the solution was in our past, right?
0:12:13 If you look at–
0:12:14 – It’s like the lesson in a movie.
0:12:16 It’s like the Wizard of Oz or something.
0:12:19 – Exactly, if you look at manufacturing,
0:12:21 I mean up to Industrial Revolution,
0:12:23 it was arts and crafts, right?
0:12:26 It was basically humans trying to figure out
0:12:27 how to conquer nature, right?
0:12:30 Like how am I gonna use my hands and my brains
0:12:35 and very few primitive tools to deform a product
0:12:39 or shape a product from raw material, right?
0:12:43 And to this date, if you are in a very high mix manufacturing,
0:12:45 still a lot of that creativity exists.
0:12:49 There is a person at SpaceX, his name is Big John.
0:12:51 I don’t think he’s there anymore.
0:12:53 But there was this guy who’s like, you know,
0:12:57 very skilled maker at Craftsman.
0:12:59 He could figure out how to use simpler tools
0:13:01 to build different things in a creative way.
0:13:03 Maybe it’s not a repeatable way like, you know,
0:13:07 a stamping works or injection molding works,
0:13:08 but you can be flexible.
0:13:09 You can do different types of things.
0:13:11 You can be creative about it and do different types of things.
0:13:13 So the inspiration came from
0:13:16 how actually humans used to do manufacturing.
0:13:19 But realize in order to be flexible,
0:13:21 you actually need two components.
0:13:25 You need intelligence and you need set of simple tools
0:13:27 with a lot of kinematic freedom.
0:13:29 Now you can pick up those simple tools.
0:13:30 And as long as you have the intelligence
0:13:32 on how to use tools and what sequence
0:13:34 and what kind of a process parameters,
0:13:35 how to use those tools,
0:13:38 you can actually do a whole variety of projects.
0:13:40 – And so when you say kinematic freedom,
0:13:42 you basically mean like robot arms
0:13:44 that can move in lots of different ways.
0:13:46 Is that practically what kinematic freedom
0:13:47 means in this context?
0:13:50 – Yes, basically can apply these tools
0:13:52 with a lot of freedom to the material, right?
0:13:55 The same way humans do it, right?
0:13:57 As a human, you know, if you think about it,
0:14:00 you can pick up a welder and weld something,
0:14:02 and then you drop the welder and you pick up a drill
0:14:04 and you put a hole in it and you drop the drill
0:14:09 and you pick up a hammer and maybe hammer it into shape.
0:14:11 So you actually have a few set of tools,
0:14:14 but you have a lot of good kinematic freedom
0:14:16 and most importantly, very creative mind
0:14:19 to tell you how to apply these tools to the material
0:14:22 so they can actually get very complex set of products
0:14:24 and a lot of diversity.
0:14:27 – So plainly, instead of big John, you want a robot, right?
0:14:29 That’s where the, that’s the kinematic freedom.
0:14:32 The tools are kind of like old tools,
0:14:34 but optimized for the robot.
0:14:36 And then when you say intelligence,
0:14:38 that’s the one where it’s like feels more frontier-ish.
0:14:42 Like, does that mean like clever engineers figuring out
0:14:43 how to automate the robots?
0:14:44 Does it mean AI?
0:14:45 Does it mean both?
0:14:48 – Yeah, so I think, yeah, you’re basically getting
0:14:51 to the crux of how do you scale it, right?
0:14:53 You have to have those three components
0:14:56 and how does the intelligent piece,
0:14:57 which is the most important piece,
0:15:00 comes into play in an automated fashion.
0:15:03 So early days, we started from basic intelligence of humans,
0:15:07 but then we had a plan to capture data and train AI
0:15:10 so that you can replace the thinking and the creativity
0:15:12 that human had to put in.
0:15:14 – What’s the first thing you decide to try and build?
0:15:17 What’s the first sort of problem you want to solve?
0:15:17 – Yeah, I think so.
0:15:19 I left for relativity in 2018.
0:15:22 And the idea when I left for relativity was there, right?
0:15:24 I was like, okay, we need to build basically
0:15:27 what I had in my mind, call it robot craftsman.
0:15:29 Robo craftsman, we call it at that time.
0:15:31 How can you build a robot system to your point?
0:15:33 Can pick up different tools, has the same kinematic,
0:15:35 but also have to have the intelligence.
0:15:38 The challenge is, you said,
0:15:39 in order to train these robots with AI,
0:15:42 you need to have a lot of data.
0:15:44 And this is not the data you can find on internet.
0:15:48 – Right, this is the AI robotics problem, it seems, right?
0:15:50 Like, unlike with large language models,
0:15:52 like that’s why we have large language models
0:15:54 and not AI robots, right?
0:15:56 Because we have the data
0:15:58 just sort of randomly sitting around on the internet
0:16:01 and we don’t have that physical world data for robots, right?
0:16:02 – Exactly.
0:16:06 – So basically the problem narrowed down into, okay,
0:16:08 how can I generate enough data?
0:16:09 How can I create a business
0:16:12 that has a sustaining way of generating data
0:16:13 so I can actually build these models?
0:16:17 I can build this intelligence for these robots.
0:16:19 And the thinking was, okay,
0:16:23 I need to create a solution that can scale in the industry
0:16:28 with limited amount of data and some heuristic.
0:16:30 But then because it’s a scaling,
0:16:32 we can generate a lot of data
0:16:33 and it starts building the AI model.
0:16:34 – Right.
0:16:37 You need a first thing that you can actually do
0:16:39 before you really have AI
0:16:42 to generate the data that will get you to AI.
0:16:44 – Exactly, exactly.
0:16:46 So we’re thinking about, okay,
0:16:49 it needs to be a large enough market, right?
0:16:51 Where we can get mass adoption
0:16:54 and we need to solve a problem that’s big enough.
0:16:56 It’s 10 times at least better than the current solution.
0:16:58 So it can actually get adoption, right?
0:17:01 – Meaning you can’t just do something as well.
0:17:03 You have to do it 10 times better.
0:17:05 – Yeah, because I think what we realized
0:17:07 is that through the last two companies,
0:17:09 if something is not 10 times better,
0:17:12 it cannot overcome the inertia
0:17:14 that exists in an industry for adoption.
0:17:16 Because if you’re doing something for the same way,
0:17:17 and you’re manufacturing,
0:17:19 people have been doing things the old way
0:17:21 for a hundred of years, right?
0:17:22 – And it’s a risk, right?
0:17:24 If they’re gonna try working with you,
0:17:26 they’re immediately taking a risk.
0:17:27 And if it’s only gonna be a little better,
0:17:29 why should I take that risk?
0:17:31 – Exactly.
0:17:32 So the idea was, okay,
0:17:34 we need to find a large enough market
0:17:35 for our first application,
0:17:36 and we need to have a solution
0:17:38 that at least 10 times better.
0:17:39 So that landed us.
0:17:40 We actually looked at a lot of things,
0:17:42 from 3D printing to forging to a lot of things.
0:17:44 And then landed on sheet metal.
0:17:47 So sheet metal is the largest metal processing sector
0:17:48 out of all.
0:17:50 It’s a $280 billion industry today.
0:17:54 And forming complex sheet metal shapes
0:17:55 is very tool intensive.
0:17:58 So what we started to do was,
0:18:01 okay, can we make a robot craftsman’s first operation
0:18:03 to be forming sheet metal?
0:18:05 Basically, forming sheet metal the same way
0:18:07 a sheet shape or hammer is a sheet into shape.
0:18:09 – And when I think about sheet metal,
0:18:11 I mean, what do, I don’t know anything about sheet metal.
0:18:13 I think of like, I think of cars,
0:18:15 I think of planes, right?
0:18:17 I think of like, you know, Detroit, like stamping.
0:18:19 Is that, am I thinking about the right thing?
0:18:22 Am I missing huge, a huge sheet metal universe?
0:18:23 Like, what’s the sheet metal universe?
0:18:26 – Yeah, so sheet metal almost is everywhere.
0:18:28 I think is the most common metal part
0:18:32 that you see on day to day, right?
0:18:33 Because most of the time we use metal
0:18:36 to be a container for other things.
0:18:38 So it’s usually a thin metal structure
0:18:41 that’s formed in complex shape to hold something else.
0:18:44 Now, you know, it can be from case of a computer,
0:18:48 you know, to a car, right?
0:18:49 You know, you’re sitting in a freeway,
0:18:50 you’re in a sea of sheet metal,
0:18:54 or to a airplane, you’re in a sheet metal can,
0:18:57 to a rocket body, for a lot of rockets,
0:19:00 some of them were composites with a lot of them are sheet metal.
0:19:05 And to agricultural heavy equipment machinery,
0:19:09 think of combines, tractors, to even building equipment.
0:19:14 So you look at your HVAC ducts are all sheet metal, right?
0:19:16 Because it just makes sense.
0:19:19 We mostly use metal parts to container other things.
0:19:21 And we give it complex shapes,
0:19:23 and that’s where sheet forming comes into play.
0:19:25 So you pretty much see it everywhere.
0:19:28 But the challenge is that in almost in all cases,
0:19:29 you have to create tooling,
0:19:30 it goes back to that first problem.
0:19:33 You have to create tooling for each of those geometries.
0:19:37 And that’s why, you know, a Ford needs to make sure
0:19:40 they can sell a million of an F-150
0:19:41 before they can invest in a plant
0:19:44 that makes a new version of F-150, right?
0:19:48 – ‘Cause you basically have to build a bespoke factory
0:19:52 just to shape sheet metal in a new way.
0:19:55 – Exactly, for a new geometry, for a new design, exactly.
0:19:57 – Where is that a particular problem?
0:20:00 Like, where is it, where does that acutely bind
0:20:03 the fact that sheet metal is so hard to do
0:20:05 if you’re not working at scale, so expensive?
0:20:07 – Yeah, so I think now you’re coming to the,
0:20:11 even the third stage of how do you scale this technology?
0:20:13 You need to first find, you know,
0:20:14 we said we need to be 10x better.
0:20:15 We need to go in an area. – Who’s your first customer?
0:20:18 – Yeah, you go in an area that has a lot of pain
0:20:19 with today’s technology. – Yeah, who’s like,
0:20:21 “Oh my God, thank God you walked through the door.
0:20:23 “We’ve been waiting for you.”
0:20:27 – Yeah, so end up being very much defense in aerospace, right?
0:20:31 So think of, you know, think of our military, for example,
0:20:36 right, today they have 50, 60 different weapon system
0:20:38 or defense systems.
0:20:40 You can basically think of like aircrafts
0:20:42 that they’re maintaining.
0:20:44 And some of these systems have been built
0:20:47 from 60, 70, 80 years ago.
0:20:49 Like think of B-50 to C-130.
0:20:52 – Like World War II planes still flying?
0:20:54 – It’s still flying, yes, exactly.
0:20:57 And they have like, you know, 30 of one,
0:20:59 50 of another, 100 of another one,
0:21:01 and these things get break down, right?
0:21:04 And unlike a Ford factory,
0:21:06 there is no factory for 70 different products
0:21:07 that they’re carrying. – Right.
0:21:11 – And presumably the factory they built in 1941
0:21:12 to build this plane doesn’t exist anymore.
0:21:13 Doesn’t exist.
0:21:16 And even the vendor might completely have disappeared, right?
0:21:18 That made that misuse component.
0:21:20 So they’re constantly battling with this challenge
0:21:23 of an aircraft goes down, how can it fix it?
0:21:24 How can it find the part?
0:21:25 And there are thousands of parts
0:21:26 in each of these aircrafts, right?
0:21:28 So any of them can go down.
0:21:29 And that’s a huge challenge.
0:21:31 I mean, if you look at, you know,
0:21:34 government of a government accountability office,
0:21:37 put this report out, I think it was a couple of years ago
0:21:41 or a year ago about how ready each weapon system is
0:21:43 to defend the United States.
0:21:46 Out of the 48, 49 weapon systems they look into,
0:21:50 only one, only one in the past 11 years,
0:21:53 every year was ready, right?
0:21:55 I think only top four had like,
0:21:58 at least half of the years ready, right?
0:22:00 So that means in most years,
0:22:02 these weapons are not ready to fight.
0:22:03 – Like they’re waiting for parts.
0:22:05 – They’re waiting for parts, something is broken,
0:22:07 something is damaged.
0:22:08 And just to go deeper,
0:22:11 some of these components take four years to be replaced.
0:22:12 So if a plane gets damaged,
0:22:14 it needs to sit on the ground for four years
0:22:16 before it can be replaced.
0:22:17 And the cost of replacement
0:22:19 is building another factory, basically.
0:22:21 So some of these parts, and think of it,
0:22:24 a landing gear door that goes on a plane,
0:22:26 it will cost them $800,000, for example,
0:22:27 because they have to go make a role.
0:22:29 – Because it’s bespoke, essentially.
0:22:30 – It’s very spoke, yeah.
0:22:31 – It’s like buying a bespoke suit or something.
0:22:33 It’s just like, it’s gonna cost a lot, yeah.
0:22:35 – Yeah, so the idea started there,
0:22:37 I think that was one of our first customers.
0:22:40 Can we make defense manufacturing more agile,
0:22:42 directly affects our national readiness
0:22:46 for military conflict, and it’s a huge problem.
0:22:47 But then, even in a broader sense,
0:22:50 any defense product or aerospace product,
0:22:53 usually has very low volume, but high mix of products.
0:22:55 Even you’re building a missile,
0:22:57 you make a few thousand a year,
0:23:00 and you might make five, six, seven different versions of it.
0:23:01 It’s very unlike cars,
0:23:05 where you make a million of the same car over and over again.
0:23:08 So that ended up being our first application,
0:23:11 which we got a lot of traction with.
0:23:13 But even outside of that,
0:23:16 you look at companies like Caterpillar,
0:23:17 like John Deere’s of the world,
0:23:19 these folks also are in the same boat.
0:23:22 They make 200 combines, right?
0:23:23 But they need to support them in the field.
0:23:26 And these folks have exactly the same problem, right?
0:23:30 Do I need to run a large factory
0:23:32 to support all these models at all time?
0:23:34 And that will be very expensive
0:23:36 to support like 100 vehicles out there.
0:23:40 (upbeat music)
0:23:41 Still to come on the show,
0:23:44 we’ll talk about the future of AI and robotics
0:23:46 at Machina Labs and beyond.
0:23:52 – Hey, it’s Jacob.
0:23:54 I’m here with Rachel Botsman.
0:23:57 Rachel lectures on trust at Oxford University,
0:24:01 and she is the author of a new Pushkin audiobook
0:24:04 called “How to Trust and Be Trusted.”
0:24:05 Hi, Rachel.
0:24:06 – Hi, Jacob.
0:24:09 – Rachel Botsman, tell me three things
0:24:11 I need to know about trust.
0:24:16 – Number one, do not mistake confidence for competence.
0:24:17 Big trust mistakes.
0:24:19 So when people are making trust decisions,
0:24:23 they often look for confidence versus competence.
0:24:27 Number two, transparency doesn’t equal more trust.
0:24:29 Big myth and misconception.
0:24:32 And a real problem actually in the tech world,
0:24:34 the reason why is because trust
0:24:38 is a confident relationship with the unknown.
0:24:39 So what are you doing?
0:24:41 If you make things more transparent,
0:24:44 you’re reducing the need for trust.
0:24:49 And number three, become a stellar expectation setter.
0:24:54 Inconsistency with expectations really damages trust.
0:24:55 – I love it.
0:24:57 Say the name of the book again
0:24:59 and why everybody should listen to it.
0:25:02 – So it’s called “How to Trust and Be Trusted.”
0:25:04 Intentionally, it’s a two-way title
0:25:08 because we have to give trust and we have to earn trust.
0:25:11 And the reason why I wrote it is because we often hear
0:25:13 about how trust is in a state of crisis
0:25:15 or how it’s in a state of decline.
0:25:17 But there’s lots of things that you can do
0:25:20 to improve trust in your own lives,
0:25:22 to improve trust in your teams,
0:25:24 trusting yourself to take more risks,
0:25:27 or even making smarter trust decisions.
0:25:29 – Rachel Botsman, the new audio book
0:25:32 is called “How to Trust and Be Trusted.”
0:25:33 Great to talk with you.
0:25:34 – It’s so good to talk with you
0:25:36 and I really hope listeners listen to it
0:25:38 because it can change people’s lives.
0:25:44 – And so you got the right market.
0:25:47 Now you gotta make a thing,
0:25:49 you gotta figure out how to actually do the thing,
0:25:52 how to make your idea come true.
0:25:53 Like, how does that work?
0:25:56 – So the idea originally was,
0:25:58 can we get rid of a die, right?
0:26:02 And do it the same way a sheet shaper forms a sheet of metal.
0:26:04 And what does a sheet shaper do?
0:26:06 A sheet shaper starts from a flat sheet of metal
0:26:09 and it slowly hammers it into shape.
0:26:12 So what we wanted to do was have a robot do that, right?
0:26:14 Have a robotic system basically do that
0:26:16 incremental defamation into shape.
0:26:18 We call it robot pouring.
0:26:19 – So you’re sort of bending it, right?
0:26:20 I mean, you’re hammering it,
0:26:22 sort of like if you take a, whatever,
0:26:24 cut open an aluminum can and kind of bend it into shape.
0:26:27 Like that’s a version of what’s happening here, right?
0:26:30 – Obviously in a much more complicated way, yeah.
0:26:31 – Exactly, you’re right.
0:26:35 I mean, the same way a potter forms a clay bowl,
0:26:36 that’s basically what our robots do.
0:26:38 They start from a flat sheet of metal
0:26:40 and it’s slowly deforming in the shape,
0:26:42 the same way a potter would form a clay bowl
0:26:45 or a sheet shaper hammers a sheet into shape.
0:26:46 – So I’ve seen it, right?
0:26:48 So there’ll be a sheet of metal like hanging,
0:26:51 hanging up in whatever above the ground.
0:26:54 And then you have a robot arm on either side, right?
0:26:55 Like one on one side, one on the other.
0:26:58 And then what happens?
0:26:59 – Basically, the robots come together
0:27:01 from both sides of the sheet.
0:27:04 And they pinch the sheet in a certain way
0:27:07 so that that location that they’re pinching
0:27:10 it slightly stretches in the forms, right?
0:27:13 And you, if you start applying this pinching
0:27:16 all over the sheet and incrementally,
0:27:20 you slowly start to form it into a shape, right?
0:27:22 So instead of traditionally we’d use a die
0:27:24 and with sheer pressure of the press,
0:27:26 pushing the sheet against the die to give it a shape.
0:27:28 Now the robots are like a craftsman,
0:27:29 like a tradesperson,
0:27:32 coming in to slowly deform the sheet
0:27:34 into shape by just applying pressure.
0:27:35 So one robot is pushing it,
0:27:37 the other robot is supporting it.
0:27:39 And by applying a pinch,
0:27:41 you slightly stretch the material
0:27:43 and you form it into a shape.
0:27:45 – So, I mean, the way you describe it,
0:27:50 it makes sense and it sounds easy.
0:27:51 I’m sure it wasn’t easy.
0:27:56 Like were there things that just didn’t work for a while?
0:27:58 – So it’s, you should have been here
0:28:01 when the first time we actually tried to form a part.
0:28:04 The part looked like it was like a ghost
0:28:06 of the geometry that they wanted to make.
0:28:09 And actually in the end it tore, right?
0:28:12 So think about it, you have this very flimsy sheet
0:28:14 and applying pressure to it.
0:28:17 And if you just apply pressure slightly wrong, right?
0:28:20 It can potentially tear it.
0:28:22 It can form it into a different shape.
0:28:23 And also the whole sheet is moving.
0:28:26 The whole time you’re trying to form it,
0:28:28 the whole sheet is moving because it’s very flimsy.
0:28:30 It’s not a rigid structure, right?
0:28:34 So the main challenge was how do you get this accurate?
0:28:35 And how do you get this process accurate?
0:28:37 How do you get accuracy?
0:28:40 And the idea was, what does the robot need to do?
0:28:43 Given all of these chaotic nature of the process,
0:28:45 where the sheet moves and if you apply too much pressure
0:28:48 it will deform in a bad way or it might tear.
0:28:52 If you apply too not enough pressure, it might just not form.
0:28:55 So how do you come up with the right set of robot movements
0:28:58 and process parameters to form the part?
0:29:00 And that was the problem we wanted to solve with AI, right?
0:29:04 Well, we didn’t have the data right in the beginning, right?
0:29:08 The idea was that if I form enough parts with this process
0:29:11 and I can capture all the data throughout the process,
0:29:12 where did the robot go?
0:29:13 How much pressure did it apply?
0:29:16 And what was the resulting geometry?
0:29:18 Then I can start building a model that says,
0:29:20 that correlates the inputs to the outputs.
0:29:21 And then I can explore this model and say,
0:29:25 “Okay, in order to get to the right output, I need these inputs.”
0:29:26 But we didn’t have them in the beginning.
0:29:28 So the idea was two things.
0:29:32 One was, maybe we can simulate the data, right?
0:29:35 And very early on we started doing some simulation,
0:29:37 physics-based simulation and we soon realized
0:29:39 in order to get an accurate result,
0:29:42 the simulations are going to be very computationally intensive.
0:29:46 A simulation of a part that took only 15 minutes to form
0:29:51 took us one week on 27 core machine, right?
0:29:52 So we’re like, okay, simulation,
0:29:55 not only is not accurate, it takes forever.
0:29:58 So we realized, okay, so that’s not the right route.
0:29:59 The right route was like, okay,
0:30:03 we can also form a lot of parts and gather the data.
0:30:05 But in order to do that, we go back to that same problem.
0:30:06 We need to have a scale.
0:30:08 We need to have a lot of these machines
0:30:09 forming these parts and get that data.
0:30:12 – I mean, one of the big AI insights of the last,
0:30:15 whatever decade is like, you need a ton of data,
0:30:17 which is easy if it’s words,
0:30:19 but hard if it’s metal, right?
0:30:20 – Yes.
0:30:24 So what we ended up doing was created a hybrid model.
0:30:28 We said, okay, what if we keep the humans in the loop?
0:30:30 So the human can give an instruction
0:30:32 initially based on heuristics.
0:30:35 And then we look at the data and human can adjust
0:30:37 and then iterate on that.
0:30:40 But while we are capturing all these data
0:30:42 and over time, as we’re capturing the data,
0:30:44 we start building the models
0:30:47 that will help the human do less trials, right?
0:30:50 It’s basically guided reinforcement learning, right?
0:30:52 You know, humans are actually guiding it where to go,
0:30:53 but it’s exploring those areas.
0:30:54 But after a while,
0:30:57 once we started forming thousands of parts,
0:30:59 then you can start feeding this data into a model.
0:31:02 Then the model will be like, okay, human,
0:31:04 you don’t need to do 25 different trials.
0:31:06 Now you can do with five trials,
0:31:07 you’re gonna get to the right place,
0:31:09 which is actually the number we are at right now.
0:31:11 – And that’s happening in the physical world,
0:31:12 largely those iterations,
0:31:14 like you’re trying on a piece of metal
0:31:15 and it’s bad and it tears
0:31:16 and you do another piece of metal
0:31:19 and it’s a little less bad and eventually you get it.
0:31:20 – Exactly, exactly.
0:31:22 And that initially would take 25 parts.
0:31:26 Like, you know, before we find a recipe for that design,
0:31:28 but 25 parts still was better than traditional alternative.
0:31:31 – When you say 25 parts, you mean 25 tries,
0:31:34 25 pieces of metal before you make the part the right way.
0:31:35 – Exactly.
0:31:37 And that was like, you know,
0:31:39 they would sit down basically 25 days in a row.
0:31:41 So in a month, they could actually define a recipe
0:31:43 where traditionally making a mold
0:31:45 would take at least three, four months, right?
0:31:46 So we were still better.
0:31:49 But then now, over time, when we generated the data
0:31:53 and now the model can tell the engineer,
0:31:55 okay, maybe you want to choose these parameters.
0:31:58 Is now becoming an advisor, we’re down to five trials.
0:32:01 So in five trials, we can actually get to the right part.
0:32:02 And then hopefully in the future,
0:32:04 we get to a point where, you know,
0:32:06 the machine will tell the robot’s what to do
0:32:09 and the human can be completely out of loop.
0:32:10 But the idea was like,
0:32:12 how do you kind of create that hybrid model that’s efficient
0:32:14 so that we can generate the data
0:32:18 until the model is good enough to do the job itself?
0:32:21 – And you find that the data is sort of generalizable.
0:32:24 I mean, clearly, like making one kind of part
0:32:26 makes the model, the AI,
0:32:28 smarter about making another kind of part.
0:32:30 – Yes, yeah, it is.
0:32:31 It’s kind of interesting.
0:32:33 I think people don’t think about it.
0:32:35 I used to do sheet shaping by hand, right?
0:32:37 That was one of the hobbies I had.
0:32:39 I was working with this shop in Pomona
0:32:41 and we were actually hammer sheets into shape.
0:32:42 And we used to say, you know,
0:32:45 if you spent five years doing it, you’re really good.
0:32:46 You got really good at it.
0:32:48 I was just to think, you know, okay,
0:32:50 after five years of doing this,
0:32:52 yes, you have this intuitive understanding of you.
0:32:52 Look at the sheet and be like,
0:32:54 okay, this place needs to be hammered more.
0:32:55 This place needs to be hammered.
0:32:58 Like it was, it was, it was, it was intuitive.
0:32:59 It was like, you couldn’t explain why you’re thinking
0:33:00 this did to happen.
0:33:02 There was no physical explanation.
0:33:03 None of these people who were sheet shaping
0:33:06 got PhDs in material science.
0:33:08 They just learned over time seeing the pattern
0:33:09 of how the sheet forms.
0:33:10 – That’s craftsmanship.
0:33:12 – That’s craftsmanship, right?
0:33:14 But really reminded me of, okay,
0:33:16 these people can know how to do it,
0:33:18 but without really being able to explain it,
0:33:19 they have to do it for five years.
0:33:22 – So that kind of tacit knowledge, yeah.
0:33:26 – And reminded me of the same challenge we had,
0:33:28 an early machine learning challenge where they were like,
0:33:31 okay, a human can look at two pictures and say,
0:33:33 okay, this is a cat and this is a dog.
0:33:35 Something happens in their brain that knows,
0:33:36 which is a cat, but they cannot really define
0:33:39 why they’re calling this a cat and this one a dog.
0:33:40 So that was where it started to click for me.
0:33:42 If I can capture enough data,
0:33:46 five years worth of data, right, of a human,
0:33:49 then I should be able to get to a very good sheet shaper,
0:33:52 right, and you know, it’s funny, back at the end,
0:33:54 but we’ll say, okay, humans are, you know,
0:33:57 receiving X amount of megabytes a second.
0:33:59 Okay, five years worth of data is that much.
0:34:02 So roughly, I think once we get to certain amount of data,
0:34:06 I think we have enough data to be able to basically replace,
0:34:08 like not replace the mentality
0:34:11 or the model that the sheet shaper has in their mind.
0:34:16 – So how many years of kind of human level craftsman
0:34:19 sheet shaping data does the model have at this point?
0:34:22 – Yeah, no, so I think last time I checked one year ago,
0:34:24 I checked around, we were like three fourths of the way there
0:34:27 in terms of the data that we have for just sheet shaping,
0:34:29 right, so once we get to, I think full, I think,
0:34:31 at least at that point we have no excuse,
0:34:32 we have enough data, the model should be good,
0:34:36 we just need to figure out why it’s not maybe four fourths.
0:34:40 – It is interesting to analogize it
0:34:42 to like human craftsmanship, right,
0:34:44 and I mean, even if you want to zoom out even more
0:34:48 the like 50 year history of AI where first everybody was like,
0:34:50 oh, you just got to teach the machine all the rules
0:34:53 for to use your example, like what’s a cat and what’s a dog,
0:34:55 but then you realize it’s actually wildly hard
0:34:56 to make a list of rules
0:35:00 that can reliably distinguish a cat from a dog,
0:35:04 and the weird thing that has happened in AI is like,
0:35:07 oh, you don’t actually have to make a list,
0:35:09 you just need like image net,
0:35:12 you just need like a giant database of images
0:35:15 and a giant neural network,
0:35:18 and you just throw it at it and say, figure it out,
0:35:21 and it figures it out, and you’re sort of doing that,
0:35:23 but for shaping metal.
0:35:25 – For shaping metal, and then the only challenge
0:35:28 was like cats and dogs pictures were on internet,
0:35:30 and sheet metal forming data wasn’t,
0:35:33 and so that was an additional problem we had to solve
0:35:35 as you pointed out, which is a big problem in physical AI.
0:35:40 – So I want to talk a little bit more about AI and robotics.
0:35:44 Jensen Wong’s been talking about it.
0:35:47 As I’m sure you know, NVIDIA’s VC arm
0:35:49 is an investor in your company.
0:35:51 Other people are working on what you’re working on.
0:35:55 I mean, I’m curious, what is the sort of AI and robotics
0:35:57 path look like to you for the next few years,
0:35:59 and what do you understand about it now
0:36:02 that you didn’t understand whatever five years ago?
0:36:04 Like what have you really come to realize
0:36:06 by working on it all the time?
0:36:09 – I think the biggest problem for physical AI
0:36:13 is data generation for now to train models.
0:36:15 So we need to either, there’s two things need to happen.
0:36:17 Either new types of models needs to be created,
0:36:21 new architectures, new algorithms, basically,
0:36:23 which I’m sure it’s going to happen.
0:36:25 – That can learn more with less data, basically.
0:36:30 – And same way humans kind of learn more with less data, right?
0:36:35 But at the same time, I think we only exposed our models
0:36:40 categorically to 10% of type of data that humans receive.
0:36:43 You think about human interactions.
0:36:45 You and I are now talking.
0:36:47 If it was AI, AI is probably only listening
0:36:50 to the words we’re saying, right?
0:36:51 But that’s only 10% of communication.
0:36:53 I can see your lips moving.
0:36:55 I can see your eyebrows moving.
0:36:57 I can see like maybe you’re folding your arms and okay,
0:36:59 I know that like, okay, maybe there’s all these,
0:37:01 90% of the signals are not captured.
0:37:04 That’s used for learning.
0:37:08 You look at, if you ask chat GPT or Dolly
0:37:12 or any of the even, even, you know, Grock say,
0:37:17 okay, draw me a clock that shows 5.30.
0:37:18 It cannot show you, draw you a clock.
0:37:20 It will draw you a clock, but it doesn’t show 5.30.
0:37:23 Actually, most of the time it shows 10.10.
0:37:25 – 10.10, ’cause that’s where watch hands,
0:37:27 like analog watch hands look good, right?
0:37:28 It’s a nice little V.
0:37:30 – Yeah, because those are all the images
0:37:32 that’s seen on internet because they’re watching.
0:37:33 – It’s almost always 10.10.
0:37:35 It’s the classic watch photo.
0:37:36 – It’s like 5.30 is also 10.10.
0:37:38 ‘Cause I don’t know– – It’s always 10.10, right?
0:37:41 – It’s a little bit to a generative AI.
0:37:42 It’s always 10.10 somewhere.
0:37:45 – So, I think, but that humans, you know,
0:37:47 receive this data of movement when you grow up,
0:37:49 you look at the clock on the wall as a kid,
0:37:51 you’re like, okay, now I intuitively get it.
0:37:52 And I think I know what’s going on.
0:37:53 So I can actually make it work.
0:37:56 So, even though we trained it on a lot of data,
0:37:58 I don’t think we trained it on the right categorically,
0:38:00 right data yet, right?
0:38:03 To get all the intuitive understanding
0:38:04 that we have today.
0:38:06 So I think we have a data problem
0:38:08 and that exists in physical AI.
0:38:10 So I think the applications will win.
0:38:11 There’s a lot of people are working in this.
0:38:13 I think the applications will win
0:38:16 who can either synthetically generate that data
0:38:19 or they can actually scale in the physical world
0:38:21 in a way where they can actually generate the data
0:38:25 for themselves, but the scaling needs to happen
0:38:26 with less data.
0:38:28 And I think that was, that’s why I’m like,
0:38:30 for example, like very bullish on manufacturing.
0:38:33 So I think the data is gonna be the biggest challenge.
0:38:35 And I think, you know, in order for us
0:38:37 to massively change this space,
0:38:39 we need to be able to get to the data.
0:38:42 I don’t think algorithms is a bottleneck there yet.
0:38:43 It’s just the data for us.
0:38:48 – And is it just a matter of people doing what you’re doing
0:38:52 and like finding little wedge places to start
0:38:54 and having people sort of hold the hand of the model
0:38:55 and training up the models?
0:39:00 I mean, that seems slow on a certain level.
0:39:02 Like not, you know, obviously it’s working for you,
0:39:05 but like, is there some kind of breakthrough move
0:39:06 that people can make?
0:39:09 Can you put sensors somewhere in the world to, you know,
0:39:12 train AI without having to, you know,
0:39:14 have a human stand next to it
0:39:17 as it messes up one piece of sheet metal after another?
0:39:21 – Yeah, I think there is another path
0:39:23 which is simulation path.
0:39:26 Make physics-based simulations faster
0:39:29 and kind of learn, let the robots just go play
0:39:31 in a digital playground as opposed to deploying
0:39:32 in real world.
0:39:34 And that becomes a computation problem.
0:39:35 And then, you know, as long as you have enough computation,
0:39:37 you can train the robots.
0:39:39 But I think, you know, I think, you know,
0:39:41 the good examples that we have had success so far
0:39:43 is like autonomous cars, right?
0:39:45 Did the same thing we were doing, but in the car,
0:39:48 like, okay, Tesla, you know,
0:39:50 deploy the fleet of robots that are capturing data,
0:39:52 still be driven by humans,
0:39:55 but the data can be used later on to kind of automate it.
0:39:56 – I mean, that’s an interesting case
0:40:00 because it has been much harder clearly
0:40:03 than many people thought, maybe most people thought, right?
0:40:05 Like, I know that’s a particular instance
0:40:08 where you’re really worried about edge cases.
0:40:12 I don’t know, is autonomous cars like a good model or not?
0:40:13 It seems complicated.
0:40:15 – I think the model of capturing data is there,
0:40:19 but then the task at hand is very hard, right?
0:40:21 So, I think that’s the challenge, right?
0:40:23 So, where it says, like with us,
0:40:25 it’s still a much more structured environment.
0:40:27 And I think that was a thinking where I think,
0:40:30 I think the hardest problem right now in physical AI
0:40:32 is finding the business model
0:40:34 of how do you scale data capture
0:40:36 without requiring billions of dollars in investment.
0:40:38 – So, what are you making today?
0:40:39 I imagine you know.
0:40:43 – So, last time I checked in the facility one,
0:40:46 four of the cells are working on a defense application.
0:40:48 – Is it secret?
0:40:50 Can you tell me what it is?
0:40:51 – It’s a missile.
0:40:55 And two of them were working on an aerospace application.
0:40:58 This is components of an aircraft or a drone.
0:41:01 And one of them, this is an interesting one,
0:41:03 was working on an architectural component,
0:41:07 which is a roof tile for a specific building
0:41:09 that’s used by the Department of,
0:41:11 by Bureau of Water Recognition.
0:41:12 – Oh, I was gonna say, what is it,
0:41:17 something like Frank Gehry nightmare weirdo metal part?
0:41:19 – Well, those we have had those in the past too,
0:41:22 but this one is actually very practical.
0:41:24 It’s this building, it’s actually very interesting.
0:41:26 Exactly, these buildings, these large industrial buildings
0:41:30 that built, they built in the ’60s or ’50s.
0:41:31 And they use these type of roof tiles
0:41:34 that the manufacturer doesn’t exist anymore.
0:41:36 And anybody else who they went to,
0:41:39 quoted them hundreds of thousand dollars to make those tiles.
0:41:41 And we were like, oh, no, we can make it for you.
0:41:43 And, but also that shows kind of the diversity.
0:41:44 I mean, like I say, in the morning,
0:41:47 we have like aerospace parts in the afternoon,
0:41:52 roof tiles for a industrial complex for, you know, for a dam.
0:41:54 – Now you’re in the sheet metal business.
0:41:57 I know your large dream is much larger than that, right?
0:42:02 But like, what, like, tell me where you are now.
0:42:05 Tell me where you are now.
0:42:05 Like what are you doing?
0:42:06 What are you selling?
0:42:09 And then kind of what’s the next big step?
0:42:12 – So some of our systems are now operating out in the wild
0:42:15 and working for the customers.
0:42:18 And, but I think the next phase of growth for us
0:42:21 is getting into each of these applications
0:42:22 and own more of the process.
0:42:25 So we can teach the robot craftsman the future processes,
0:42:28 not just sheet forming, but also maybe how to assemble it,
0:42:31 how to weld it, how to surface finish it, right?
0:42:35 So what we are doing now in the next phase is actually,
0:42:39 instead of selling parts or components or systems,
0:42:40 we’re actually saying, okay,
0:42:44 can we get this robot craftsman
0:42:46 to actually build a US sub-assembly or a full product,
0:42:49 not just a component of it, but a full product.
0:42:51 So that’s something we’re describing with folks.
0:42:53 Can we have the robot craftsman
0:42:54 build the full drone for you?
0:42:59 Can we have the robot craftsman build you a full missile,
0:43:02 as opposed to just build missile, you know, missile scans?
0:43:05 – Is there, that seems like a leap.
0:43:07 Is there not an intermediate step?
0:43:08 – Yes.
0:43:09 – Yeah.
0:43:10 – So, I mean, how we are doing it
0:43:13 is we’re gradually stepping into it, right?
0:43:15 The same way sheet metal was our first application.
0:43:18 So we’re putting a facility that maybe makes drones,
0:43:22 but the main component that we automate today
0:43:24 is sheet forming, which is the bottleneck.
0:43:27 And then we do the welding in a traditional way
0:43:30 on the same robots, but we actually instruct them to do it.
0:43:31 – All right.
0:43:33 – So that way, the robot is kind of back
0:43:36 where it was on sheet metal five years ago,
0:43:38 but it’s learning how to weld now.
0:43:39 – Exactly.
0:43:42 I used to work in a shop that we will do custom cars,
0:43:44 build custom cars with hand.
0:43:47 And so it was also near and near to my heart.
0:43:50 So what we realized is that with our technology
0:43:54 for the first time, we can actually enable
0:43:56 a product that didn’t exist in automotive,
0:44:00 meaning that instead of buying a car that’s mass produced
0:44:02 and every single one of them looked the same,
0:44:06 you can now let the customer design a custom car for them.
0:44:07 You know, right now, if you go buy a car,
0:44:11 you have options of what the seat color would be,
0:44:13 or maybe the color of the car would be
0:44:15 and what some trim options,
0:44:17 but you can’t really choose the design of your car.
0:44:19 You can’t say, oh, I want a different hood,
0:44:21 and I want a different fender
0:44:23 because going to back same problem,
0:44:25 you don’t have to make tooling and mold
0:44:28 for the fender of a certain designs
0:44:29 and you cannot easily change it.
0:44:30 So with our technology, you can.
0:44:34 So what we started doing was like, okay,
0:44:37 applying this freedom that this technology provides
0:44:40 to now automotive is ability of the customers
0:44:41 to be able to go to a website,
0:44:45 design a fully customized car for themselves.
0:44:47 It can be either from already designed panels
0:44:52 from car designer or adding a specific customizations
0:44:53 they want to do, for example,
0:44:55 logo of their company to their door of the car
0:44:56 or the hood of the car
0:44:59 and actually get a completely unique car, right?
0:45:01 Manufactured for them.
0:45:03 And we’re actually working with this
0:45:04 with some of our automotive partners,
0:45:07 automotive OEMs as well, right?
0:45:09 We actually showed some of this work
0:45:11 in the biggest aftermarket show in the United States
0:45:14 called SEMA with our partner Toyota.
0:45:17 So I think this is gonna be, in my opinion,
0:45:20 one of the new product categories in automotive.
0:45:22 We have had autonomous cars,
0:45:24 we have had, you know, electric cars.
0:45:26 And I think now for the first time
0:45:27 with technologies like ours,
0:45:29 you can have custom to order cars,
0:45:31 like all cars that are like, you know,
0:45:33 the same way you choose what T-shirt you wear
0:45:35 and your T-shirt is different than mine,
0:45:37 we also don’t have to drive the same, you know,
0:45:39 Model S or, you know, Model 3.
0:45:41 We can actually have our own customized
0:45:43 Model 3s and Model Ss.
0:45:45 – So what’s the, I mean, is that the,
0:45:48 if you think sort of long-term for Machina,
0:45:49 is like that what you think about?
0:45:53 Like, give me the five-year vision.
0:45:54 – Yeah.
0:45:55 – Or 10-year or whatever.
0:45:58 – Yeah, so I think the long-term motivation
0:46:00 behind our company is,
0:46:02 can it create this democratization
0:46:05 of ideas for people who wanna build anything, right?
0:46:06 Can I express myself?
0:46:09 If I’m a builder, can I go build something
0:46:11 without having to build a factory for it?
0:46:13 So that’s really the long-term goal.
0:46:17 So I imagine in the next five to 10 years,
0:46:20 you can, as a designer, somebody who has an idea,
0:46:25 you can go to a website, get guided through your ideas
0:46:28 on how to make and design a physical product,
0:46:31 hit a button and say, okay, I want 20 of these
0:46:35 and I want them in Chatsworth, California
0:46:38 and the right facility programs the right number of robots
0:46:39 to actually do those operations
0:46:41 without any hardware or investment
0:46:44 that needs to be made for those specific parts
0:46:46 and ship it to you two days later in the right location.
0:46:49 That is the future we’re building towards.
0:46:53 Cars is just one of the products that could be built.
0:46:56 But I imagine that this technology or technology like these,
0:46:58 technologies like these can be used to do
0:46:59 the myriad of designs.
0:47:03 I think the moment you open up this possibility
0:47:05 of any designs could be a reality.
0:47:07 I think so many things will be created
0:47:09 that we’re not even thinking of right now.
0:47:12 You know, the fact that we have cars today
0:47:13 and they’re all looked the same
0:47:15 is limitation of technology.
0:47:18 But the moment you can open up this creativity
0:47:20 of turning ideas into physical reality
0:47:24 without an initial investment or a huge barrier to entry,
0:47:26 then I think we’re gonna have all kinds of drones,
0:47:28 all kinds of satellites, all kinds of rockets,
0:47:31 all kinds of cars, then you’re gonna be just like,
0:47:34 you know, Cambrian explosion of different designs
0:47:35 that’s gonna come into our world.
0:47:37 And I think that’s what future is about.
0:47:38 The future is about, you know,
0:47:39 I call like future is custom.
0:47:41 Like future is about being able to make
0:47:43 these all these ideas in reality.
0:47:45 We had this explosion happening in digital world.
0:47:46 – Yeah.
0:47:48 – You know, now we have even models generating images
0:47:52 and videos and there’s this, you know,
0:47:55 explosion of different ideas and content being created
0:47:58 using the technology, but the link is broken
0:47:59 to the physical world.
0:48:01 And the physical work is still pretty uniform
0:48:03 ’cause it’s very hard to make things in physical work.
0:48:05 Can we bridge that gap?
0:48:08 Can we connect the digital world of creation
0:48:09 to physical world of creation
0:48:12 and create the same variety in the physical world
0:48:14 as we have in the digital world?
0:48:16 I think that’s the goal in our company.
0:48:19 (upbeat music)
0:48:22 We’ll be back in a minute with the lightning round.
0:48:24 (upbeat music)
0:48:28 – Hey, it’s Jacob.
0:48:30 I’m here with Rachel Botsman.
0:48:34 Rachel lectures on trust at Oxford University.
0:48:37 And she is the author of a new Pushkin audio book
0:48:40 called How to Trust and Be Trusted.
0:48:41 Hi, Rachel.
0:48:42 – Hi, Jacob.
0:48:46 – Rachel Botsman, tell me three things
0:48:47 I need to know about trust.
0:48:52 – Number one, do not mistake confidence for competence.
0:48:53 Big trust mistakes.
0:48:55 So when people are making trust decisions,
0:48:59 they often look for confidence versus competence.
0:49:04 Number two, transparency doesn’t equal more trust.
0:49:05 Big myth and misconception.
0:49:08 And a real problem actually in the tech world.
0:49:11 The reason why is because trust
0:49:14 is a confident relationship with the unknown.
0:49:17 So what are you doing if you make things more transparent?
0:49:20 You’re reducing the need for trust.
0:49:25 And number three, become a stellar expectation setter.
0:49:30 Inconsistency with expectations really damages trust.
0:49:31 – I love it.
0:49:33 Say the name of the book again
0:49:35 and why everybody should listen to it.
0:49:38 So it’s called How to Trust and Be Trusted.
0:49:40 Intentionally, it’s a two-way title
0:49:44 because we have to give trust and we have to earn trust.
0:49:47 And the reason why I wrote it is because we often hear
0:49:49 about how trust is in a state of crisis
0:49:51 or how it’s in a state of decline.
0:49:54 But there’s lots of things that you can do
0:49:56 to improve trust in your own lives,
0:49:58 to improve trust in your teams,
0:50:01 trusting yourself to take more risks
0:50:04 or even making smarter trust decisions.
0:50:05 – Rachel Botsman, the new audio book
0:50:08 is called How to Trust and Be Trusted.
0:50:09 Great to talk with you.
0:50:10 – It’s so good to talk with you
0:50:12 and I really hope listeners listen to it
0:50:15 because it can change people’s lives.
0:50:19 – Let’s finish with the lightning round.
0:50:23 Do you drive a customized car?
0:50:26 – I don’t actually yet.
0:50:27 Well, if I am–
0:50:29 – What have I seen on your Instagram?
0:50:31 What’s that truck you keep posting on your Instagram?
0:50:33 – So I have a truck that’s customized.
0:50:36 I don’t drive it around as much.
0:50:38 But maybe this year, I’ll start taking it out this year.
0:50:40 We have been kind of stealth about it.
0:50:41 We’ve been talking about it,
0:50:43 but we haven’t talked about it in a big way
0:50:45 because we have a big release coming soon.
0:50:47 – I mean, you’re literally posting it on Instagram.
0:50:49 It’s not that stealth.
0:50:52 Tell me about that truck you keep posting on Instagram.
0:50:53 What’s going on with that?
0:50:54 – So it’s a truck that’s fully,
0:50:56 the full body is fully customized.
0:50:59 – It says Anvil in the back when you post it.
0:51:00 Is it called Anvil?
0:51:00 – Dumb question.
0:51:02 – We call it Anvil.
0:51:04 I think the idea was that actually the shape,
0:51:05 design of it was inspired by Anvil.
0:51:07 If you look at the front fender,
0:51:10 it actually looks like the front bumper looks like an Anvil.
0:51:12 But also the idea is that like,
0:51:14 we are actually forming sheets on an Anvil.
0:51:16 So it was very fitting.
0:51:17 Tell me about that truck.
0:51:19 Like just tell me what’s it look like?
0:51:20 – Yeah. So for example, like,
0:51:24 we put a lot of form and sharp edges in the hood, right?
0:51:27 Most vehicles have a very hard time.
0:51:30 If you look at most of the hood of the vehicles,
0:51:32 they’re very smooth because it’s very hard
0:51:35 to actually put sharp angles in the hood.
0:51:36 So if you look at this truck,
0:51:37 this truck has a lot of angles,
0:51:41 a lot of sharp detail right in the hood, right?
0:51:44 And that’s very expressive of the type of person,
0:51:46 for example, that I am, right?
0:51:47 I like things that are edgy.
0:51:50 And that truck is certainly edgy, right?
0:51:53 It’s bare metal, right?
0:51:58 There is no blemishes being hidden under the vehicle.
0:52:02 A lot of people, when Cybertruck came out,
0:52:04 we got very excited about, you know,
0:52:06 oh, it’s bare metal, it looks like a metal,
0:52:07 but then there was no form in it
0:52:08 because it’s actually very hard
0:52:11 to make a form metal look nice.
0:52:13 And so that’s one of the things we wanted to show.
0:52:14 We wanted to show that, okay,
0:52:17 you can actually have a form metal with a lot of detail in it
0:52:20 and still keep it bare metal because it will look nice, right?
0:52:24 So yeah, a lot of design features of it for me
0:52:26 kind of represents the type of personality character
0:52:29 that I have, but I think that’s how every car should be.
0:52:31 You know, people should be able to have that freedom
0:52:33 to choose what their cars look like.
0:52:37 – How many skull tattoos do you have?
0:52:39 – I’ve got three.
0:52:41 – Why?
0:52:46 – So yeah, it’s an interesting thing.
0:52:50 So a skull for me represents kind of,
0:52:54 and it’s an abstract for death of ego.
0:52:56 So I have a tattoo on my thumb,
0:52:59 which is a skull that’s holding a microphone
0:53:00 into his ears.
0:53:02 And this was a time where, you know,
0:53:04 I felt like, you know, I had a good platform
0:53:07 and I could talk a lot and people would listen,
0:53:09 but then I realized I should, yes, that’s great,
0:53:12 but I should maybe keep the mic close to my ears
0:53:16 and also listen as opposed to talk all the time, right?
0:53:18 So I think a skull–
0:53:21 – Microphones don’t work that way for the record,
0:53:23 but I like it as a metaphor.
0:53:27 – Exactly, but I think the idea was around, you know,
0:53:28 kind of reminders.
0:53:30 You can see a lot of my tattoos are on my hands,
0:53:34 so it’s really a reminder for myself to know that, you know,
0:53:37 be present and make sure that, you know,
0:53:39 you’re not involved with your ego too much
0:53:42 and you can see other people’s perspective.
0:53:45 – Is there any tension between ego death and custom cars?
0:53:50 – Tension between ego death and custom cars.
0:53:52 – I don’t know, I’m just playing, but like, you know,
0:53:55 custom car kind of seems like, hey, look at me, I’m special.
0:53:58 And ego death seems like, oh, don’t look at me,
0:53:59 I’m not so special.
0:54:01 – Yeah, no, I think the difference is,
0:54:04 I think, yeah, if you have attachment to your custom car,
0:54:06 then maybe there’s tension,
0:54:09 but I more think of it in terms of expression, right?
0:54:10 You know, you can be an artist.
0:54:14 You can design your home the way it expresses you.
0:54:17 You can design the theme of your podcast
0:54:18 the way it expresses you.
0:54:21 You can design your car also the way it expresses you.
0:54:23 I think it’s less so about, oh, look at me, I’m special.
0:54:26 It’s more like, here’s my expression to the world
0:54:28 for the people to see.
0:54:31 But I think that expressiveness is pretty amazing.
0:54:33 I think that’s uniquely one of the unique things
0:54:36 about humans that like, you know, we,
0:54:37 I think all we do when we come to this world
0:54:39 is expressing ourselves, right?
0:54:41 Expressing ourselves through our work,
0:54:43 expressing through ourselves, through our relationships.
0:54:47 And if we can enable people to express themselves better,
0:54:49 better, I think that’s great.
0:54:51 But if you get attached to your expressions
0:54:53 and your ideas and your thoughts,
0:54:55 and you think, oh, I’m better than everybody else,
0:54:56 then I think that that becomes,
0:54:59 that becomes a little bit of an ego-driven trip.
0:55:02 (upbeat music)
0:55:10 – Edward Mayer is the co-founder and CEO of Mockin’ The Labs.
0:55:13 Today’s show was produced by Gabriel Hunter Chang.
0:55:15 It was edited by Lydia Jean Cotte
0:55:18 and engineered by Sarah Bruginger.
0:55:21 You can email us at problem@pushkin.fm.
0:55:24 I’m Jacob Goldstein and we’ll be back next week
0:55:26 with another episode of What’s Your Problem?
0:55:34 – Hey, it’s Jacob.
0:55:35 I’m here with Rachel Botsman.
0:55:39 Rachel lectures on trust at Oxford University
0:55:43 and she is the author of a new Pushkin audiobook
0:55:46 called How to Trust and Be Trusted.
0:55:47 Hi, Rachel.
0:55:48 – Hi, Jacob.
0:55:51 – Rachel Botsman, tell me three things
0:55:53 I need to know about trust.
0:55:58 – Number one, do not mistake confidence for competence.
0:55:59 Big trust mistakes.
0:56:01 So when people are making trust decisions,
0:56:05 they often look for confidence versus competence.
0:56:09 Number two, transparency doesn’t equal more trust.
0:56:11 Big myth and misconception.
0:56:13 And a real problem, actually in the tech world,
0:56:16 the reason why is because trust
0:56:20 is a confident relationship with the unknown.
0:56:23 So what are you doing if you make things more transparent?
0:56:26 You’re reducing the need for trust.
0:56:31 And number three, become a stellar expectation setter.
0:56:36 Inconsistency with expectations really damages trust.
0:56:37 – I love it.
0:56:38 Say the name of the book again
0:56:41 and why everybody should listen to it.
0:56:43 – So it’s called How to Trust and Be Trusted.
0:56:46 Intentionally, it’s a two-way title
0:56:49 because we have to give trust and we have to earn trust.
0:56:53 And the reason why I wrote it is because we often hear
0:56:55 about how trust is in a state of crisis
0:56:57 or how it’s in a state of decline.
0:56:59 But there’s lots of things that you can do
0:57:02 to improve trust in your own lives,
0:57:04 to improve trust in your teams,
0:57:06 trusting yourself to take more risks
0:57:09 or even making smarter trust decisions.
0:57:11 – Rachel Botsman, the new audio book is called
0:57:14 How to Trust and Be Trusted.
0:57:14 Great to talk with you.
0:57:16 – It’s so good to talk with you.
0:57:18 And I really hope listeners listen to it
0:57:21 because it can change people’s lives.
AI works well in the virtual world. That’s partly because the internet provides so much data to train AI models. But there’s no analogous data set for the physical world – and as a result, AI doesn’t work as well there… yet.
Edward Mehr is the co-founder and CEO of Machina Labs. Edward’s problem is this: How can you use AI to turn robots from dumb, inflexible machines into skilled, versatile craftsmen?
See omnystudio.com/listener for privacy information.