Stop comparing yourself to AI

0:00:01 Are you forgetting about that chip in your windshield?
0:00:03 It’s time to fix it.
0:00:05 Come to Speedy Glass before it turns into a crack.
0:00:08 Our experts will repair your windshield in less than an hour,
0:00:09 and it’s free if you’re insured.
0:00:12 Book your appointment today at speedyglass.ca.
0:00:14 Details and conditions at speedyglass.ca.
0:00:20 What do you think about when you think about AI?
0:00:25 Maybe chatbots giving you new lasagna recipes,
0:00:29 research assistants helping you finish that paper.
0:00:33 Do you think about machines taking your job?
0:00:37 Maybe you think of something even more ominous,
0:00:41 like Skynet robots wiping out humanity.
0:00:46 If you’re like me, you probably think of all those things,
0:00:47 depending on the day.
0:00:50 And that’s sort of the point.
0:00:55 AI is not well understood, even by the people creating it.
0:00:58 And even though we all know it’s a technology
0:01:00 that’s going to change our lives,
0:01:03 that’s really all we know at this point.
0:01:10 So how do we confront this uncertainty?
0:01:13 How do we navigate the current moment?
0:01:17 And how do we, the people who have been told
0:01:19 that we will be impacted by AI,
0:01:21 but don’t seem to have much of a say
0:01:23 in how the AI is being built,
0:01:26 engage in the conversation?
0:01:31 I’m Sean Elling, and this is The Gray Area.
0:01:45 Today’s guest is Jaron Lanier.
0:01:48 He’s a virtual reality pioneer,
0:01:50 a digital philosopher,
0:01:54 and the author of several best-selling books on technology.
0:01:57 He’s also one of the most profound critics
0:02:01 of Silicon Valley and the business model driving it.
0:02:04 I wanted to bring Jaron on the show
0:02:07 for the first episode of this special series on AI
0:02:10 because I think he’s uniquely positioned
0:02:14 to speak both to the technological side of AI,
0:02:16 what’s happening, where it’s going,
0:02:20 and also to the human side.
0:02:24 Jaron’s a computer scientist who loves technology.
0:02:29 But at his core, he’s a humanist
0:02:32 who’s always thinking about what technologies are doing to us
0:02:36 and how our understanding of these tools
0:02:39 will inevitably determine how they’re used.
0:02:43 Maybe what Jaron does the best, though,
0:02:45 is offer a different lens
0:02:47 through which to view these technologies.
0:02:51 We’re encouraged to treat these machines
0:02:54 as though they’re godlike,
0:02:56 as though they’re thinking for themselves.
0:03:01 Indeed, they’re designed to make you feel that way
0:03:04 because it adds to the mystique around them
0:03:07 and obscures the truth about how they really work.
0:03:12 But Jaron’s plea is to be careful
0:03:15 about thoughtlessly adopting the language
0:03:17 that the AI creators give us
0:03:18 to describe their creation
0:03:21 because that language structures
0:03:25 not only how we think about these technologies,
0:03:27 but what we do with them.
0:03:35 Jaron Lanier, welcome to the show.
0:03:36 That’s me. Hey.
0:03:39 So look, I have heard
0:03:43 so many of these big picture conversations about AI
0:03:48 and they often begin with a question
0:03:52 about how or whether AI is going to take over the world.
0:03:55 But I discovered very quickly
0:03:57 that you don’t accept the terms of that question,
0:03:59 which is why I’m not going to ask it.
0:04:01 but I thought it would be useful
0:04:03 as a beginning to ask you
0:04:05 why you find questions like that
0:04:07 or claims like that ridiculous.
0:04:10 Oh, well, you know,
0:04:12 when it comes to AI,
0:04:15 the whole technical field
0:04:16 is kind of defined
0:04:19 by an almost metaphysical assertion,
0:04:22 which is we are creating intelligence.
0:04:23 Well, what is intelligence?
0:04:26 Something human.
0:04:28 The whole field was founded
0:04:31 by Alan Turing’s thought experiment
0:04:32 called the Turing test,
0:04:37 where if you can fool a human
0:04:38 into thinking you’ve made a human,
0:04:40 then you might as well have made a human
0:04:42 because what other tests could there be?
0:04:45 Which in a way is fair enough.
0:04:45 On the other hand,
0:04:47 what other scientific field
0:04:50 other than maybe supporting stage magicians
0:04:53 is entirely based on being able to fool people?
0:04:53 I mean, it’s stupid.
0:04:56 Fooling people in itself accomplishes nothing.
0:04:58 There’s no productivity.
0:04:59 There’s no insight
0:05:01 unless you’re studying
0:05:03 the cognition of being fooled, of course.
0:05:06 So there’s an alternative way
0:05:07 to think about what we do
0:05:09 with what we call AI,
0:05:12 which is that there’s no new entity.
0:05:14 There’s nothing intelligent there.
0:05:16 What there is is a new
0:05:17 and in my opinion,
0:05:18 sometimes quite useful
0:05:21 form of collaboration between people.
0:05:23 If you look at something like the Wikipedia,
0:05:25 where people mash up
0:05:27 a lot of their communications into one thing,
0:05:30 you can think of that as a step on the way
0:05:32 to what we call large model AI,
0:05:34 where we take all the data that we have
0:05:35 and we put it together
0:05:39 in a way that allows more interpolation
0:05:43 and more commingling than previous methods.
0:05:47 And I think that can be of great use,
0:05:49 but I don’t think there’s any requirement
0:05:52 that we perceive that as a new entity.
0:05:53 Now, you might say,
0:05:54 well, what’s the harm if we do?
0:05:56 That’s a fair question.
0:05:57 Like, who cares?
0:05:58 If somebody wants to think of it
0:06:00 as a new type of person
0:06:02 or even a new type of God or whatever,
0:06:03 what’s wrong with that?
0:06:06 Potentially nothing.
0:06:08 People believe all kinds of things all the time.
0:06:12 But, in the case of our technology,
0:06:15 let me put it this way.
0:06:19 If you’re a mathematician or a scientist,
0:06:25 you can do what you do
0:06:27 in a kind of an abstract way.
0:06:28 Like, you can say,
0:06:30 I’m furthering math.
0:06:33 And, in a way, that’ll be true
0:06:35 even if nobody else ever even perceives
0:06:36 that I’ve done it.
0:06:37 I’ve written down this proof.
0:06:40 But that’s not true for technologists.
0:06:43 Technologists only make sense
0:06:46 if there’s a designated beneficiary.
0:06:49 Like, you have to make technology for someone.
0:06:52 And, as soon as you say
0:06:56 the technology itself is a new someone,
0:07:00 you stop making sense as a technologist.
0:07:01 Right?
0:07:03 Let me actually take up that question
0:07:04 that you just posed a second ago
0:07:05 with a thought,
0:07:07 I’ve heard from you,
0:07:09 which is something to the effect of,
0:07:11 I think the way you put it is
0:07:13 the easiest way to mismanage a technology
0:07:15 is to misunderstand it.
0:07:17 So, to answer your question…
0:07:18 Sounds like me, I guess.
0:07:19 Yeah. Okay.
0:07:22 If we make the mistake,
0:07:23 which is now common,
0:07:26 to insist that AI is, in fact,
0:07:28 some kind of god or creature
0:07:30 or entity or oracle,
0:07:31 whatever term you prefer,
0:07:33 instead of a tool as you define it,
0:07:34 the implication is that
0:07:37 that would be a consequential mistake, right?
0:07:39 That we will mismanage the technology
0:07:40 by misunderstanding it.
0:07:41 So, is that not quite right?
0:07:42 Am I not quite understanding?
0:07:43 No, I think that’s right.
0:07:46 I think when you treat the technology
0:07:47 as its own beneficiary,
0:07:49 you miss a lot of opportunities
0:07:50 to make it better.
0:07:52 Like, I see this in AI all the time.
0:07:53 I see people saying,
0:07:55 well, if we did this,
0:07:56 it would pass the Turing test better,
0:07:57 and if we did that,
0:07:58 it would seem more like
0:07:59 it was an independent mind.
0:08:01 But those are all goals
0:08:01 that are different
0:08:04 from it being economically useful.
0:08:05 They’re different from it
0:08:08 being useful to any particular user.
0:08:09 They’re just these weird,
0:08:12 to me, almost religious ritual goals
0:08:13 or something.
0:08:15 like they, and so every time
0:08:16 you’re devoting yourself to that,
0:08:18 it means you’re not devoting yourself
0:08:20 to making it better.
0:08:22 Like, an example is,
0:08:25 we have, in my view,
0:08:28 deliberately designed large model AI
0:08:32 to obscure the original human sources
0:08:34 of the data that the AI is trained on
0:08:36 to help create this illusion
0:08:37 of the new entity.
0:08:38 But when we do that,
0:08:41 we make it harder to do quality control.
0:08:43 We make it harder to do authentication
0:08:48 and to detect malicious uses of the model
0:08:52 because we can’t tell what the intent is,
0:08:54 what data it’s drawing upon.
0:08:56 We’re sort of willfully making ourselves
0:08:58 kind of blind in a way
0:09:00 that we probably don’t really need to.
0:09:01 And I really want to emphasize
0:09:03 from a metaphysical point of view,
0:09:05 I can’t prove,
0:09:06 and neither can anyone else,
0:09:08 that a computer is alive or not
0:09:09 or conscious or not or whatever.
0:09:11 I mean, all that stuff
0:09:13 is always going to be a matter of faith.
0:09:15 That’s just the way it is.
0:09:17 That’s what we got around here.
0:09:19 But what I can say
0:09:21 is that this emphasis
0:09:22 on trying to make the models
0:09:25 seem like they’re freestanding new entities
0:09:27 does blind us
0:09:29 to some ways we could make them better.
0:09:30 And so I think, like, why bother?
0:09:32 What do we get out of that?
0:09:32 Not a lot.
0:09:34 So do you think maybe
0:09:35 the cardinal mistake
0:09:37 with a lot of this kind of thinking
0:09:38 is to assume
0:09:42 that artificial intelligence
0:09:43 is something that’s in competition
0:09:45 with human intelligence
0:09:46 and human abilities,
0:09:47 that that kind of misunderstanding
0:09:48 sets us off on a course
0:09:50 for a lot of other kinds
0:09:51 of misunderstandings?
0:09:53 I wouldn’t choose that language
0:09:54 because then the natural thing
0:09:55 somebody’s going to say
0:09:56 who’s a true believer
0:09:57 that the AI is coming alive,
0:09:58 they’re going to say,
0:09:59 yeah, you’re right.
0:10:00 It’s not competition.
0:10:01 We’re going to align them
0:10:02 and they’re going to be
0:10:03 our collaborators
0:10:05 or whatever.
0:10:06 So that, to me,
0:10:07 doesn’t go far enough.
0:10:09 My own way of thinking
0:10:11 is that I’m able
0:10:12 to improve the models
0:10:13 when I say
0:10:14 there’s no new entity there.
0:10:15 I just say they don’t,
0:10:15 they’re not there.
0:10:16 They don’t exist
0:10:17 as separate entities.
0:10:18 They’re just collaborations
0:10:19 of people.
0:10:20 I have to go that far
0:10:22 to get the clarity
0:10:23 to improve them.
0:10:26 It might be a little late
0:10:27 in the language game
0:10:29 to replace a term
0:10:30 like artificial intelligence,
0:10:30 but if you could,
0:10:31 do you have a better one?
0:10:34 I have had the experience
0:10:35 of coming up with terms
0:10:37 that were widely adopted
0:10:37 in society.
0:10:38 I came up with
0:10:39 virtual reality
0:10:40 and some other things
0:10:41 when I was young
0:10:44 and I have seen that
0:10:45 even when you get
0:10:46 to coin the term,
0:10:47 you don’t get to define it
0:10:50 and I don’t love
0:10:51 the way people think
0:10:52 of virtual reality
0:10:53 typically today.
0:10:54 It’s lost a little bit
0:10:55 of its old humanism,
0:10:56 I would say.
0:10:59 So that experience
0:11:00 has led me to feel
0:11:01 that it’s really
0:11:02 younger generations
0:11:03 who should come up
0:11:03 with their own terms.
0:11:04 So what I would prefer
0:11:06 to see is younger people
0:11:07 reject our terms
0:11:09 and come up
0:11:09 with their own.
0:11:11 Fair enough.
0:11:14 I’ve read a lot
0:11:14 of your work
0:11:15 on AI
0:11:17 and I’ve listened
0:11:19 to a lot of your interviews
0:11:21 and I take your point
0:11:22 that AI
0:11:25 is a distillation
0:11:26 of all these human inputs
0:11:27 fundamentally.
0:11:30 but for you at what point
0:11:32 does or can complexity
0:11:35 start looking like autonomy
0:11:37 and what would autonomy
0:11:38 even mean
0:11:39 that the thing starts
0:11:40 making its own decisions
0:11:41 and is that the simple
0:11:42 definition of that?
0:11:43 This is an obsession
0:11:44 that people have
0:11:45 but you have to understand
0:11:46 it’s a religious
0:11:48 and entirely subjective
0:11:50 or sort of cultural obsession
0:11:51 not a scientific one.
0:11:52 It’s your judgment
0:11:54 of how you want to see
0:11:55 the start of autonomy.
0:11:58 So I love complex systems
0:11:59 and I love different levels
0:12:00 of description
0:12:01 and I love the independence
0:12:03 of different levels
0:12:03 of grantedness
0:12:04 in physics
0:12:06 so I’m utterly
0:12:07 as obsessed
0:12:07 as anyone
0:12:08 with that
0:12:10 but it’s important
0:12:10 to distinguish
0:12:12 that fascination
0:12:12 which is a scientific
0:12:13 fascination
0:12:14 with the question
0:12:16 of does crossing
0:12:17 some threshold
0:12:18 make something
0:12:19 human or not?
0:12:21 because the question
0:12:22 of humanness
0:12:24 or of becoming
0:12:24 an entity
0:12:26 that we care about
0:12:27 in our planning
0:12:27 becoming
0:12:28 creating something
0:12:29 that itself
0:12:30 is a beneficiary
0:12:31 of our technology
0:12:32 that question
0:12:33 has to be
0:12:34 a matter of faith
0:12:36 we just have
0:12:36 to accept
0:12:38 that our culture
0:12:39 our law
0:12:40 our ability
0:12:41 to be technologists
0:12:42 ultimately rests
0:12:43 on values
0:12:45 that in a sense
0:12:45 we pull out
0:12:46 of our asses
0:12:47 or if you like
0:12:48 we have to be
0:12:49 a little bit mystical
0:12:50 in order to create
0:12:51 the ground layer
0:12:52 in order to be
0:12:52 then rational
0:12:53 as technologists
0:12:54 in a way
0:12:55 I wish it wasn’t so
0:12:56 it sort of sucks
0:12:57 but it’s just the truth
0:12:57 and the sooner
0:12:58 we accept that
0:12:59 the better off
0:13:00 we’ll be
0:13:00 and the more honest
0:13:01 we’ll be
0:13:02 and I’m okay with it
0:13:03 why?
0:13:05 because
0:13:06 if I’m designing
0:13:07 AI for AI’s sake
0:13:08 I’m talking nonsense
0:13:09 you know
0:13:10 like
0:13:11 right now
0:13:13 it’s very expensive
0:13:13 to compute AI
0:13:14 so what percentage
0:13:16 of that expense
0:13:17 it goes into
0:13:18 creating the illusion
0:13:19 so that you can believe
0:13:20 it’s sort of
0:13:21 another person
0:13:22 when you use chat
0:13:23 how much electricity
0:13:24 is being spent
0:13:25 so that the way
0:13:26 it talks to you
0:13:27 feels like it’s a person
0:13:28 a lot
0:13:28 you know
0:13:29 and it’s a waste
0:13:30 like why are we doing that
0:13:31 why are we doing
0:13:32 why are we creating
0:13:34 a carbon footprint
0:13:36 for the benefit
0:13:38 of some non-entity
0:13:39 in order to fool humans
0:13:40 like it’s
0:13:40 it’s ridiculous
0:13:42 but we don’t see that
0:13:43 because we have this
0:13:45 religious imperative
0:13:46 in the tech
0:13:48 cultural world
0:13:49 to create
0:13:50 this new life
0:13:52 but it’s entirely
0:13:53 a matter of
0:13:54 our own perception
0:13:55 there’s no test
0:13:55 for it
0:13:56 other than the
0:13:56 Turing test
0:13:57 which is no test
0:13:57 at all
0:13:58 I mean
0:13:59 we still don’t even
0:14:01 have a real
0:14:01 definition
0:14:03 of consciousness
0:14:05 and I hear all
0:14:05 these discussions
0:14:07 about machine learning
0:14:09 and human intelligence
0:14:09 and the differences
0:14:11 and I continue
0:14:12 to have no idea
0:14:13 when something
0:14:14 stops being a
0:14:15 simulacrum of intelligence
0:14:16 and becomes the real thing
0:14:17 I still don’t quite know
0:14:18 when something can
0:14:19 reasonably be called
0:14:20 sentient
0:14:21 or intelligent
0:14:22 but maybe the question
0:14:22 doesn’t even matter
0:14:24 maybe it’s enough
0:14:25 for us to think it does
0:14:26 right
0:14:27 so the problem
0:14:28 in what you just
0:14:29 said is the word
0:14:29 still
0:14:32 like it’s a
0:14:33 this
0:14:35 lack of knowledge
0:14:36 is structural
0:14:37 you’re not going
0:14:38 to overcome it
0:14:39 you can pretend
0:14:40 you have
0:14:40 but you’re not going
0:14:41 to
0:14:42 this is genuinely
0:14:43 a matter of faith
0:14:43 you know
0:14:44 and
0:14:46 it’s a very
0:14:46 old discussion
0:14:47 when it comes
0:14:48 to God
0:14:49 but
0:14:50 it’s a new
0:14:50 discussion
0:14:51 when it comes
0:14:52 to each other
0:14:53 or to AIs
0:14:54 and
0:14:54 you know
0:14:55 like
0:14:56 faith is okay
0:14:56 we can live
0:14:57 with faith
0:14:57 we just have
0:14:58 to be honest
0:14:59 about it
0:14:59 and I think
0:15:01 being dishonest
0:15:01 and saying
0:15:02 oh
0:15:03 it’s not faith
0:15:04 I have this
0:15:04 rational proof
0:15:05 of something
0:15:07 it’s not
0:15:08 dishonesty
0:15:08 is probably
0:15:09 not good
0:15:10 especially
0:15:10 if you’re
0:15:10 trying to do
0:15:11 science or technology
0:15:15 maybe we just
0:15:17 maybe we just
0:15:18 hold on
0:15:19 maybe
0:15:20 I’m going to
0:15:21 say this
0:15:23 we probably
0:15:23 just have to
0:15:24 hold on to
0:15:24 some notion
0:15:25 that there’s
0:15:26 something
0:15:26 fundamentally
0:15:27 special
0:15:28 about human
0:15:29 consciousness
0:15:30 and that even
0:15:30 if on some
0:15:31 purely empirical
0:15:31 level
0:15:32 that’s not
0:15:32 even true
0:15:33 maybe believing
0:15:34 that it is
0:15:34 is essential
0:15:36 to our
0:15:36 survival
0:15:37 I don’t
0:15:37 think you
0:15:38 can rationally
0:15:40 proceed
0:15:41 as an
0:15:41 as an
0:15:42 acting
0:15:42 technologist
0:15:44 without
0:15:45 an
0:15:46 irrational
0:15:47 belief
0:15:48 that people
0:15:49 are special
0:15:50 because once again
0:15:50 then you have
0:15:51 no recipient
0:15:52 and if you
0:15:53 say well
0:15:53 there’s going
0:15:54 to be
0:15:54 no belief
0:15:55 all the way
0:15:55 to the bottom
0:15:56 it’s just
0:15:56 going to be
0:15:57 rationality
0:15:57 forever
0:15:58 I mean
0:15:59 it doesn’t
0:15:59 work
0:16:00 rationality
0:16:01 never creates
0:16:01 a total
0:16:02 enclosed
0:16:02 system
0:16:04 we kind
0:16:05 of float
0:16:05 in a sea
0:16:05 of mystery
0:16:06 and we
0:16:06 have like
0:16:07 this belief
0:16:07 that lets
0:16:08 us have
0:16:08 a footing
0:16:09 and it’s
0:16:10 our job
0:16:11 to acknowledge
0:16:11 that even
0:16:12 if we’re
0:16:12 uncomfortable
0:16:13 with it
0:16:15 can I try
0:16:15 another angle
0:16:16 on you
0:16:16 yeah
0:16:17 do you know
0:16:17 my
0:16:17 okay
0:16:18 so there’s
0:16:18 another
0:16:19 argument
0:16:19 about the
0:16:20 turing test
0:16:20 right
0:16:21 turing test
0:16:22 you have a
0:16:23 person on a
0:16:23 computer
0:16:23 they’re each
0:16:24 trying to fool
0:16:24 a judge
0:16:25 and at the
0:16:26 moment the
0:16:26 judge can’t
0:16:26 tell them
0:16:27 apart
0:16:27 you say
0:16:28 well we
0:16:28 might as
0:16:29 well call
0:16:30 the computer
0:16:31 human because
0:16:31 what other
0:16:31 tests can
0:16:32 there be
0:16:32 that’s the
0:16:32 best we’ll
0:16:33 get
0:16:33 okay
0:16:35 so the
0:16:36 problem with
0:16:36 the test
0:16:37 is that it
0:16:38 measures whether
0:16:38 there’s a
0:16:38 differential
0:16:39 but it
0:16:40 doesn’t tell
0:16:40 you whether
0:16:41 the computer
0:16:42 got smarter
0:16:42 or the
0:16:42 human got
0:16:43 stupider
0:16:44 it doesn’t
0:16:45 tell you if
0:16:45 the computer
0:16:46 became more
0:16:47 human or if
0:16:47 the human
0:16:48 became less
0:16:48 human in
0:16:49 any sense
0:16:49 whatever that
0:16:50 might be
0:16:51 so there’s
0:16:52 two humans
0:16:52 the contestant
0:16:53 and the judge
0:16:53 and one
0:16:54 computer
0:16:54 therefore
0:16:56 and this is
0:16:56 meant to be
0:16:57 funny but it’s
0:16:57 also kind of
0:16:57 real
0:16:58 there’s a
0:16:58 two-thirds
0:16:59 chance that
0:16:59 it was a
0:17:00 human that
0:17:00 got stupider
0:17:01 rather than
0:17:01 a computer
0:17:01 that got
0:17:02 smarter
0:17:04 and I
0:17:04 see that
0:17:05 borne out
0:17:05 like when I
0:17:06 look at
0:17:06 social media
0:17:07 and I see
0:17:08 people interacting
0:17:08 with the AI
0:17:09 algorithms that
0:17:10 are supposed to
0:17:10 guide their
0:17:11 attention
0:17:12 I see them
0:17:13 getting stupider
0:17:13 two-thirds
0:17:14 of the time
0:17:14 but then you
0:17:15 know sometimes
0:17:16 really good
0:17:16 stuff happens
0:17:17 so I think
0:17:18 this general
0:17:19 spread of most
0:17:20 of the time
0:17:20 things get
0:17:21 worse but then
0:17:21 there’s some
0:17:22 stuff that’s
0:17:22 really cool
0:17:24 tends to be
0:17:24 true when you
0:17:25 believe in AI
0:17:26 and so
0:17:27 I would
0:17:28 say don’t
0:17:28 believe in
0:17:28 it and
0:17:30 some people
0:17:30 are still
0:17:31 getting it
0:17:31 stupider
0:17:31 because that’s
0:17:32 how we are
0:17:33 but I think
0:17:33 we can get to
0:17:34 the point where
0:17:34 the majority
0:17:35 gets better
0:17:36 instead of
0:17:37 stupider but
0:17:37 right now I
0:17:37 think we’re
0:17:38 at two-thirds
0:17:39 get stupider
0:17:40 yeah that
0:17:41 math checks out
0:17:41 to me
0:17:42 great I
0:17:43 think that’s
0:17:43 a rigorous
0:17:44 argument that’s
0:17:44 what you call
0:17:45 a rigorous
0:17:46 quantitative
0:17:47 theoretically and
0:17:48 empirically supported
0:17:49 argument right
0:17:49 there
0:17:50 so do you
0:17:51 think all
0:17:53 the anxieties
0:17:54 including from
0:17:55 serious people
0:17:56 in in the
0:17:57 world of AI
0:17:58 all the worries
0:18:00 about human
0:18:01 extinction and
0:18:01 mitigating the
0:18:02 risks thereof
0:18:04 does that is
0:18:04 that religious
0:18:06 hysteria to
0:18:06 you or does
0:18:07 that feel
0:18:09 what drives me
0:18:09 crazy about
0:18:10 this I this
0:18:11 is my world
0:18:11 you know so I
0:18:12 talk to the
0:18:12 people who
0:18:13 believe that
0:18:14 stuff all the
0:18:15 time and
0:18:16 increasingly a
0:18:16 lot of them
0:18:17 believe that it
0:18:17 would be good to
0:18:18 wipe out people
0:18:19 and that the AI
0:18:19 future would be a
0:18:20 better one and
0:18:21 that we should
0:18:22 wear a disposable
0:18:24 temporary container
0:18:25 for the birth of
0:18:26 AI I hear that
0:18:27 opinion quite a lot
0:18:27 that’s a real
0:18:28 opinion held by
0:18:29 real people
0:18:32 many many I
0:18:33 mean like the
0:18:34 other day I was
0:18:35 at a lunch in
0:18:36 Palo Alto and
0:18:36 there were some
0:18:37 young AI
0:18:38 scientists there
0:18:39 who were saying
0:18:41 that they would
0:18:42 never have a
0:18:43 bio baby because
0:18:43 as soon as you
0:18:44 have a bio baby
0:18:44 you get the
0:18:46 mind virus of
0:18:48 the bio world
0:18:48 and that when
0:18:49 you have the
0:18:50 bio mind virus
0:18:50 you become
0:18:51 committed to
0:18:52 your human baby
0:18:52 but it’s much
0:18:53 more important to
0:18:54 be committed to
0:18:54 the AI of the
0:18:56 future and so
0:18:57 to have human
0:18:58 babies is
0:18:58 fundamentally
0:18:59 unethical
0:19:01 now okay in
0:19:01 this particular
0:19:03 case this was
0:19:03 a young man
0:19:04 with a female
0:19:05 partner who
0:19:06 wanted a kid
0:19:06 and what I’m
0:19:07 thinking is this
0:19:07 is just another
0:19:08 variation of the
0:19:09 very very old
0:19:10 story of young
0:19:11 men attempting to
0:19:12 put off the baby
0:19:13 thing with their
0:19:14 sexual partner as
0:19:15 long as possible
0:19:16 because I’ve been
0:19:16 there and many of
0:19:16 us have been
0:19:17 there so in a
0:19:18 way I think it’s
0:19:19 not anything new
0:19:19 and it’s just the
0:19:20 old thing but
0:19:21 it’s a very
0:19:23 common attitude
0:19:25 not the dominant
0:19:25 one I would say
0:19:26 the dominant one
0:19:27 is that the
0:19:28 super AI will
0:19:29 turn into this
0:19:30 god thing that’ll
0:19:31 save us and
0:19:32 will either upload
0:19:33 us to be immortal
0:19:34 or solve all our
0:19:34 problems at the
0:19:35 very least or
0:19:36 something create
0:19:37 super abundance at
0:19:38 the very very very
0:19:41 least and I
0:19:45 I have to say
0:19:45 there’s a bit of
0:19:46 an inverse
0:19:47 proportion here
0:19:48 between the people
0:19:49 who directly work
0:19:50 in making AI
0:19:51 systems and then
0:19:51 the people who
0:19:52 are adjacent to
0:19:54 them who have
0:19:54 these various
0:19:57 beliefs my own
0:19:58 opinion is that
0:19:59 the people
0:20:00 how can I put
0:20:02 this the people
0:20:03 who are able to
0:20:04 be skeptical and
0:20:05 a little bored and
0:20:06 dismissive of the
0:20:07 technology they’re
0:20:08 working on tend to
0:20:09 improve it more than
0:20:09 the people kind of
0:20:10 worship it too much
0:20:13 like I’ve seen that
0:20:14 a lot in a lot of
0:20:15 different things not
0:20:16 not just computer
0:20:17 science and I think
0:20:18 I think you have to
0:20:19 have a kind of
0:20:20 like you can’t drink
0:20:21 your own whiskey too
0:20:22 much when you’re a
0:20:24 technologist you have
0:20:25 to kind of be ready
0:20:26 to say oh maybe
0:20:27 this thing’s a bit
0:20:28 overhyped I’m not
0:20:29 going to tell that
0:20:30 to the people buying
0:20:31 shares in my company
0:20:31 but you know what
0:20:32 like just between us
0:20:35 you know and but
0:20:35 that attitude is
0:20:37 exactly the one that
0:20:38 puts you over the
0:20:38 threshold to then
0:20:39 start improving it
0:20:40 more and that’s one
0:20:41 of the dangers of
0:20:42 this kind of
0:20:43 mythologizing of it
0:20:44 oh it’s about to
0:20:45 become this god
0:20:45 that’ll take over
0:20:46 everything but
0:20:48 that what follows
0:20:49 from that is this
0:20:50 very curious thing
0:20:51 which is that the
0:20:52 way of thinking
0:20:53 about it where it’s
0:20:54 about to turn into
0:20:55 this god that’ll
0:20:56 run everything and
0:20:57 either kill us all
0:20:57 or fix all our
0:20:58 problems that
0:21:00 attitude in itself
0:21:02 makes you not
0:21:04 only a little bit
0:21:05 of a lesser
0:21:06 improver of the
0:21:07 technology by any
0:21:08 like real measurable
0:21:10 metric but it
0:21:11 also makes you a
0:21:12 bad steward of it
0:21:15 part of part of
0:21:15 what makes this
0:21:16 very confusing
0:21:17 especially to you
0:21:19 know non-technical
0:21:20 normie outsiders
0:21:21 like me and like
0:21:22 most people frankly
0:21:24 is that it is it’s
0:21:25 just moving and
0:21:26 changing and evolving
0:21:27 really quickly and
0:21:28 the terms and
0:21:29 concepts are very
0:21:30 slippery if you’re
0:21:32 not deep in it and
0:21:32 you know you’re
0:21:33 talking about super
0:21:34 super AI and godlike
0:21:36 powers one example
0:21:37 is and you’ll bear
0:21:38 with me for a second
0:21:39 so I can bring people
0:21:41 along we have this
0:21:42 dichotomy between
0:21:44 AI versus AGI
0:21:45 artificial intelligence
0:21:46 versus artificial
0:21:47 general intelligence and
0:21:48 my understanding is
0:21:50 that AI is a term for
0:21:51 the general set of
0:21:52 tools that people
0:21:53 are building chat
0:21:54 bots and that sort
0:21:54 of thing and that
0:21:56 AGI is still sort of
0:21:57 a theoretical thing
0:21:58 where this tech is
0:22:00 basically as good at
0:22:01 everything as a
0:22:03 normal regular person
0:22:03 is and it can also
0:22:04 learn and grow and
0:22:05 apply that knowledge
0:22:07 just like we can and
0:22:08 we’ve got AI now
0:22:09 clearly but we don’t
0:22:11 have AGI yet and if
0:22:13 we get it and there
0:22:13 are people who think
0:22:14 we’re maybe closer
0:22:15 than we thought
0:22:16 recently that it’ll be
0:22:18 a real Rubicon
0:22:20 crossing moment for
0:22:21 us what’s your
0:22:22 feeling on that do
0:22:23 you think AGI is
0:22:24 even possible in the
0:22:25 way most people
0:22:26 have you not
0:22:26 listened to a word
0:22:28 I said that’s a
0:22:28 religious question
0:22:30 that’s like asking
0:22:30 if I think the
0:22:31 rapture is coming
0:22:33 soon I mean it’s
0:22:33 yeah but you can
0:22:34 have an opinion
0:22:34 about religious
0:22:35 questions I guess
0:22:38 that’s true I mean
0:22:40 there are those who
0:22:41 say we have AGI
0:22:42 already and their
0:22:43 opinion is as
0:22:44 legitimate as
0:22:45 anybody else’s I
0:22:46 mean I just think
0:22:47 the moment you’ve
0:22:48 put the question
0:22:48 that way you’ve
0:22:49 already confused
0:22:50 yourself and made
0:22:50 yourself kind of
0:22:51 useless in talking
0:22:52 about what to do
0:22:53 with the technology
0:22:54 so I have to reject
0:22:55 your question as
0:22:56 being like poorly
0:22:56 framed and
0:22:57 ill-informed I’m
0:22:59 sorry I was hoping
0:22:59 to get through this
0:23:00 fucking conversation
0:23:01 without you having
0:23:02 to beat back at
0:23:03 one of my ill-informed
0:23:04 questions and I
0:23:05 did make it I made
0:23:06 it almost 20 minutes
0:23:07 in yeah good luck
0:23:08 with that my friend
0:23:12 all right sir
0:23:13 it was a valiant
0:23:14 effort you win that
0:23:17 you really I mean
0:23:19 look I mean this
0:23:20 is silly this is
0:23:21 like I’m also
0:23:21 trying to speak for
0:23:22 concerns that I
0:23:23 know a lot of
0:23:24 people I know
0:23:25 because we broadcast
0:23:26 that way of thinking
0:23:27 about it so yeah
0:23:31 look there’s a
0:23:31 thing all right
0:23:33 look I’m I
0:23:35 benefit from people
0:23:36 believing in AI
0:23:37 professionally and
0:23:39 there’s a way that
0:23:39 the whole economy
0:23:40 runs on attention
0:23:42 getting and in a
0:23:44 funny way the way
0:23:45 digital attention
0:23:46 economy works
0:23:51 is it rewards
0:23:52 anxieties and
0:23:54 terror as much
0:23:54 or maybe a
0:23:56 little more than
0:23:59 optimism or you
0:24:01 know goodwill and
0:24:02 so you have this
0:24:03 weird situation where
0:24:05 somebody can play
0:24:06 the villain on
0:24:06 social media and
0:24:08 do very well and
0:24:09 similar things
0:24:10 happening in the
0:24:11 rhetoric of computer
0:24:12 science so when we
0:24:13 say oh our stuff
0:24:14 might be about to
0:24:15 come alive and
0:24:16 it’s about to get
0:24:17 smarter than you
0:24:18 it generates this
0:24:19 little anxiety in
0:24:20 people and then that
0:24:21 actually benefits us
0:24:22 because it keeps it
0:24:24 keeps the attention
0:24:27 on us and so
0:24:28 there’s a funny way
0:24:29 that we’re
0:24:30 incentivized to put
0:24:31 things in the most
0:24:33 alarming way what I
0:24:34 what I will say is
0:24:36 that I like the
0:24:37 idea of models being
0:24:38 useful so I think
0:24:40 of the models that
0:24:41 we’re building as
0:24:42 being wonderful
0:24:43 mashup models so
0:24:44 like for instance
0:24:46 I love being able
0:24:47 to use large models
0:24:48 to go through the
0:24:48 scientific literature
0:24:51 and find correlations
0:24:51 between different
0:24:52 papers that might not
0:24:53 use the same
0:24:54 terminology that would
0:24:54 have been a pain in
0:24:55 the butt to detect
0:24:57 before that’s great
0:24:58 if you present that
0:24:59 with a chat
0:25:00 interface it seems
0:25:01 like a smart
0:25:02 scientist if people
0:25:03 like that I mean I
0:25:04 guess whatever it’s
0:25:05 not my job to judge
0:25:06 everybody but the
0:25:08 thing is you don’t
0:25:09 need to present it
0:25:09 that way you’d
0:25:10 still get the
0:25:11 same value but
0:25:11 that’s the way we
0:25:13 do it we we add
0:25:14 in personhood
0:25:16 fooling to what
0:25:17 would otherwise be
0:25:19 really in a way
0:25:20 more clear
0:25:21 freestanding value I
0:25:23 think but we like
0:25:24 to present the
0:25:24 fantasy
0:25:37 there’s over 500
0:25:38 thousand small
0:25:40 businesses in bc and
0:25:40 no two are alike
0:25:42 i’m a carpenter i’m a
0:25:43 graphic designer i sell
0:25:45 dog socks online that’s
0:25:47 why bcaa created one
0:25:48 size doesn’t fit all
0:25:49 insurance it’s
0:25:51 customizable based on
0:25:52 your unique needs so
0:25:53 whether you manage
0:25:54 rental properties or
0:25:55 paint pet portraits you
0:25:56 can protect your small
0:25:58 business with bc’s most
0:25:59 trusted insurance brand
0:26:01 visit bcaa.com slash
0:26:03 small business and use
0:26:04 promo code radio to
0:26:05 receive fifty dollars
0:26:06 off conditions apply
0:26:16 all right let me try to
0:26:17 pull away a little bit
0:26:18 from religious questions
0:26:22 okay so look i i’m i’m
0:26:23 not worried about the
0:26:23 matrix and the
0:26:25 terminator um i am
0:26:27 worried about a much
0:26:28 more boring and
0:26:30 unsexy scenario but i
0:26:31 think equally bad
0:26:34 possibility is that these
0:26:37 emergent technologies will
0:26:39 accelerate a trend that
0:26:41 i think digital tech in
0:26:42 general and social media
0:26:43 in particular has already
0:26:47 started which is to pull
0:26:49 us away more and more
0:26:50 from the physical world
0:26:52 and encourage us to
0:26:54 perform versions of
0:26:55 ourselves in the virtual
0:26:56 world and because of how
0:26:58 it’s designed it has this
0:27:00 habit of reducing other
0:27:02 people to crude avatars
0:27:03 which is why it’s so easy
0:27:05 to be cruel and vicious
0:27:07 online and why people who
0:27:08 are on social media too
0:27:10 much start to become
0:27:12 mutually unintelligible
0:27:13 to each other and i
0:27:16 worry about ai super
0:27:17 charging some of this
0:27:18 stuff i mean do you even
0:27:19 accept that framing am i
0:27:20 right to be thinking of ai
0:27:23 as a potential accelerant of
0:27:26 these trends yeah i mean i
0:27:29 i think you are correct
0:27:36 so it’s arguable and
0:27:37 actually consistent with the
0:27:38 way the community speaks
0:27:41 internally to say that the
0:27:43 algorithms that have been
0:27:44 driving social media up to
0:27:49 now are a form of ai if you
0:27:52 if you unlike me wish to use
0:27:55 the term ai and what the
0:27:59 algorithms do is they
0:28:01 attempt to predict human
0:28:03 behavior based on the
0:28:05 stimulus given to the
0:28:07 human and by putting that
0:28:08 in an adaptive loop they
0:28:11 hope to drive attention and
0:28:13 sort of an obsessive
0:28:15 attachment to a platform
0:28:18 because these algorithms
0:28:21 can’t tell whether
0:28:23 something’s being driven
0:28:25 because of things that we
0:28:25 might think are positive
0:28:26 or things that we might
0:28:28 think are negative so i
0:28:29 call this the life of the
0:28:30 parody the this notion
0:28:32 that you can’t tell like
0:28:33 if a bid is one or zero
0:28:34 doesn’t matter because it’s
0:28:36 an arbitrary designation in
0:28:38 a digital system so if
0:28:39 somebody’s getting
0:28:40 attention by being a dick
0:28:42 that works just as well as
0:28:43 if they’re offering
0:28:44 life-saving information or
0:28:45 helping people improve
0:28:46 themselves but then the
0:28:47 peaks that are good are
0:28:48 really good and i don’t
0:28:49 want to deny that i love
0:28:50 dance culture on tiktok
0:28:53 science bloggers on on
0:28:54 youtube have achieved a
0:28:55 level that’s like
0:28:57 astonishingly good and so
0:28:58 on like there’s all these
0:29:00 really really positive good
0:29:01 spots but then overall
0:29:03 there’s this loss of truth
0:29:06 and political paranoia and
0:29:09 unnecessary confrontation
0:29:11 between arbitrarily created
0:29:13 cultural groups and so on
0:29:15 that’s really doing damage
0:29:18 um and as is often pointed
0:29:20 out especially to young
0:29:21 girls and so on and so
0:29:22 forth uh not not great
0:29:25 and so uh yeah could
0:29:27 better ai algorithms make
0:29:27 that worse
0:29:31 plausibly i mean it’s
0:29:32 possible that it’s already
0:29:34 bottomed out that it’s kind
0:29:37 of the the badness just
0:29:37 comes from the overall
0:29:38 structure and if the
0:29:39 algorithms themselves get
0:29:41 more sophisticated it won’t
0:29:42 really push it that much
0:29:43 further but i think
0:29:45 actually kind of can i’m
0:29:46 i’m worried about it i
0:29:48 because we so much want to
0:29:49 pass the turing test and
0:29:50 make people think our
0:29:51 programs are people
0:29:55 we’re moving to this um
0:29:56 so-called agentic era where
0:29:59 it’s not just that you have a
0:30:00 chat interface with with the
0:30:01 thing but the chat interface
0:30:04 gets to know you for years at
0:30:06 a time and gets a so-called
0:30:08 personality and but and all
0:30:09 this and then the idea is that
0:30:10 people then fall in love with
0:30:11 these and we’re already
0:30:13 seeing examples of this
0:30:15 here and there um and this
0:30:16 notion of a whole generation
0:30:17 of young people falling in
0:30:20 love with fake avatars i mean
0:30:24 people people talk about ai as
0:30:25 if it’s just like this yeast in
0:30:26 the air it’s like oh ai will
0:30:27 appear and people will fall in
0:30:29 love with ai avatars but it’s
0:30:30 not ai is always run by
0:30:32 companies so like they’re going
0:30:33 to be falling in love with
0:30:35 something from google or meta or
0:30:39 whatever and like that notion
0:30:41 that your love life becomes
0:30:44 owned by some company or even
0:30:45 worse tiktok or a chinese thing
0:30:49 eek eek eek eek i think that’ll
0:30:51 create a a a new centralization
0:30:56 or or or xai eek eek eek eek i’ll
0:30:57 add some more eeks to that and so
0:30:59 this centralization of power and
0:31:02 influence could be even worse and
0:31:04 that might be a breaking point
0:31:06 event and so that kind of thing
0:31:07 ending civilization or ending up
0:31:09 killing all the people does seem
0:31:11 plausible to me and some of my
0:31:12 colleagues would interpret that as
0:31:15 ai become coming alive and killing
0:31:16 everybody but i would just
0:31:17 interpret it as people being
0:31:20 making terrible choices it all
0:31:21 amounts to the same thing in the
0:31:23 end anyway it does at the end of
0:31:25 the day in terms of actual events
0:31:27 the same so jaron from your point
0:31:29 of view is it even possible to have
0:31:33 good algorithms nudging us around
0:31:35 online or are all algorithms bad yes
0:31:37 of course it is okay what does that
0:31:39 look like course it is of course it
0:31:41 is yes yes yes yes give me the good
0:31:42 stuff here give me the good
0:31:44 algorithms well i mean look in the
0:31:49 scientific community we do it like i
0:31:51 mean like okay here’s an example um
0:31:55 deep research from open ai is a great
0:31:57 tool it does a literature search on some
0:31:59 topic and assembles a little report
0:32:03 it has unnecessary chatbot elements
0:32:05 to try to make it seem like there’s
0:32:07 somebody there i view that as a waste
0:32:10 of time and a waste of energy and i i
0:32:11 would be happy without it but but
0:32:13 whatever okay it’s it’s not terrible
0:32:16 though what it does is it saves
0:32:18 scientists a ton of time it makes a lot
0:32:20 of sense i get a lot out of it it’s
0:32:21 great and now there’s some new
0:32:25 competitors to it great that stuff’s
0:32:27 fabulous i really really really it’s
0:32:28 good because the scientific literature
0:32:30 has become impossible to use without
0:32:33 it i do a lot of work that’s pretty
0:32:35 mathematical and the problem is that
0:32:37 every time somebody comes across
0:32:38 similar math they don’t realize
0:32:39 somebody else has done it so they come
0:32:41 up with their own terms for things and
0:32:43 then you have the same ideas or
0:32:45 similar ones with different terms and
0:32:46 all these scattered papers in totally
0:32:47 different communities at different
0:32:48 conferences and different journals
0:32:53 yeah but with a tool like this you
0:32:55 can capture all that and get it into
0:32:59 place it’s like what what what ai is is
0:33:01 it’s a way of improving collaboration
0:33:03 between people it’s a way of gathering
0:33:06 what people have done in a more unified
0:33:09 way that can notice multiple hops of
0:33:12 different terms and similar structures it’s
0:33:15 it’s a better way of using statistics to
0:33:17 connect what we’ve all done together to
0:33:21 get more use out of it it’s great i love
0:33:25 it and the amount of avatar illusion
0:33:27 nonsense is kept to a minimum because
0:33:29 our job is not to fall in love with our
0:33:31 research our fake research assistant our
0:33:35 job is to make progress efficiently on
0:33:37 whatever we’re doing right and so that
0:33:39 that’s great what is wrong with that
0:33:41 nothing it’s fabulous so yeah there’s
0:33:43 wonderful uses if i didn’t think those
0:33:46 things existed i’d quit what i do
0:33:49 professionally in the industry of course
0:33:51 there’s wonderful uses and i think we
0:33:52 need those things i think they really
0:33:53 matter
0:33:56 i guess what i’m hovering around is the
0:33:58 business model right i mean uh the
0:34:00 advertising model was sort of the
0:34:02 original sin of the internet yeah yeah i
0:34:02 think it is
0:34:06 um how do we not fuck this up how do we
0:34:07 not repeat those mistakes what’s a better
0:34:09 model i mean you talk a lot about data
0:34:11 dignity so you’re saying we can say fuck
0:34:13 on this podcast oh you can say whatever
0:34:15 you want if i had known that there would
0:34:17 be a lot of fuckery up to now in my in my
0:34:18 speech it’s not too late anyway it’s not
0:34:23 too late we got plenty of time okay but no
0:34:25 but seriously what how do we get it right
0:34:26 this time how do we not make the same
0:34:29 mistakes what is a better model yeah well
0:34:32 um this is actually more important this
0:34:34 question is the central question of our
0:34:36 time in my view like the central
0:34:39 question of our time isn’t um being able
0:34:42 to scale ai more is is an important
0:34:45 question and i get that and most people
0:34:47 are focused on that and dealing with the
0:34:49 climate is an important question but in
0:34:51 terms of our own survival coming up with
0:34:53 a business model for civilization that
0:34:56 isn’t self-destructive is in a way our
0:34:59 most primary problem and challenge right
0:35:01 now because the way we’re doing it what
0:35:04 we kind of we went through this thing in
0:35:06 the earlier phase of the internet like
0:35:08 information should be free and then the
0:35:09 only business model that’s left is paying
0:35:12 for influence uh and so then all the
0:35:16 platforms look free or very cheap to the
0:35:17 user but then actually the real customer
0:35:19 trying to influence the user and you end
0:35:23 up with what’s essentially a stealthy form
0:35:26 of um manipulation being the central
0:35:30 project of civilization and we can only
0:35:31 get away with that for so long at some
0:35:33 point that bites us and we become too
0:35:36 crazy to survive so we must change the
0:35:38 business model of civilization and so
0:35:41 exactly how to get from here to there is
0:35:44 a bit of a mystery but i continue to work
0:35:46 on it like i think we should incentivize
0:35:48 people to put great data into the ai
0:35:51 programs of the future uh and i’d like
0:35:53 people to be paid for data used
0:35:55 ai models and also to be celebrated and
0:35:56 made visible and known because i think
0:35:58 it’s just a big collaboration and our
0:36:01 collaborators should be valued how easy
0:36:02 would it be to do that do you think we
0:36:05 can or will there’s still some unsolved
0:36:07 technical questions about how to do it
0:36:09 i’m very very actively working on those
0:36:10 and i believe it’s doable and there’s a
0:36:12 whole you know research community devoted
0:36:14 to exactly that distributed around the
0:36:16 world and i think it’ll make better
0:36:18 models i mean better data makes better
0:36:20 models and there’s a lot of people who
0:36:21 dispute that and they say no it’s just
0:36:22 better algorithms and we already have
0:36:25 enough data for the rest of all time but
0:36:28 i disagree with that i think i don’t
0:36:29 think we’re the smartest people who will
0:36:31 ever live and there might be new creative
0:36:33 things that happen in the future that we
0:36:35 don’t foresee and the models we’ve
0:36:37 currently built might not extend into
0:36:39 those things and having some open system
0:36:41 where people can contribute to new models
0:36:44 in new ways is a more expansive and
0:36:47 creative and you know open-minded and
0:36:51 and just you know kind of spiritually
0:36:53 optimistic way of thinking about the deep
0:36:53 future
0:37:15 today explained here with eric levitt senior
0:37:17 correspondent at vox.com to talk about the
0:37:21 2024 election that can’t be right eric i thought
0:37:22 we were done with that i feel like i’m pacino
0:37:24 in three just when i thought i was out
0:37:28 they pull me back in why are we talking about
0:37:30 the 2024 election again the reason why we’re
0:37:33 still looking back is that it takes a while
0:37:36 after an election to get all of the most high
0:37:40 quality data on what exactly happened so the
0:37:42 full picture is starting to just come into view
0:37:45 now and you wrote a piece about the full
0:37:49 picture for vox recently and it did bonkers business
0:37:53 on the internet what did it say what struck a
0:37:56 chord yeah so this was my interview with
0:38:00 david shore of blue rose research he’s one of
0:38:04 the biggest sort of democratic data gurus in
0:38:08 the party and basically the big picture headline
0:38:12 takeaways are on today explained you’ll have to go listen
0:38:15 to them there find the show wherever you listen to shows bro
0:38:35 i think i’m a humanist like you in the end and what i want fundamentally is just the
0:38:39 elevation of human agency not the diminishment of it and part of what that means to borrow your
0:38:45 language is creating more creative classes and less dependent classes yep uh you’ve convinced me
0:38:50 that that’s at least possible i don’t know if it’s likely but i hope it is and and maybe some
0:38:55 some kind of data dignity type model is the most promising thing i’ve heard
0:39:06 no i sort of feel like the human project our our survival is simultaneously both certain and
0:39:11 unlikely if you know what i mean like i i feel like if we just follow the immediate trend lines
0:39:13 and what we see we’re probably gonna
0:39:16 buck ourselves up to use the word i’m
0:39:22 encouraged to say here there you go but i also just have this feeling we’ve made it through a lot of
0:39:26 stuff in the past and i just have this feeling we’re gonna rise to the occasion and figure this
0:39:32 one out really i don’t know exactly how we will but i think we will i don’t know what the
0:39:40 alternative is the alternative is in 200 million years there’ll be smart cephalopods to take over
0:39:45 the planet uh and maybe they’ll do that i mean that’s the alternative but i think we can do it i
0:39:56 really do i really i we just we just have to be a little less full of ourselves and not believe we’re
0:40:02 making a new god no more golden calves that’s really our problem still yeah good luck with
0:40:09 that i mean i i i like i’m constantly thinking more about the the social and political and cultural
0:40:14 dynamics because that’s just my background um and you know i mean i i guess speaking of dependent
0:40:23 classes i a very common concern is is this fear that ai is going to create a lot of social instability by
0:40:29 taking all of our jobs it’s a widespread fear it’s scary as hell and it feels like
0:40:35 the latest iteration of a very old story about new technologies like automation displacing workers
0:40:39 i mean how do you speak to these sorts of fears when you hear them because surely you hear them a lot
0:40:41 yeah and they concern me i mean
0:40:54 look um there’s not a perfect solution to that problem uh there i’ll give you an example of one
0:41:00 that i find tricky to think about uh my mom died in a car accident and i’ve always believed from when
0:41:05 i was very young that cars should drive themselves that it was manifestly obvious that we could create
0:41:13 a digital system that would save many many lives so we have tens of thousands of people killed by cars
0:41:16 every year still in the us and i think it’s over a million worldwide or something like that i mean
0:41:24 it’s like crazy it’s like and so um there are a lot of reasons for it and a self-driving car is never
0:41:28 going to be perfect because it’s not a task that can be done perfectly there’ll be circumstances where
0:41:34 there’s no optimal solution you know in the instant but overall we ought to be able to save a lot of
0:41:42 life so i’m really supportive of that project at the same time an incredibly large number of blue collar
0:41:49 people around the world get by behind a wheel whether it’s truck drivers or rideshare drivers these days
0:42:00 or etc you know and so like how do you reconcile those two things uh and i i don’t think there’s any way to do it perfectly i think there’s
0:42:10 there’s two things that should be true one is that we need to find an intermediate way to love
0:42:15 a social safety net that isn’t all the way to universal basic income because the universal basic
0:42:21 income idea gives people this idea that they’re not worth anything and they’re just being supported by
0:42:27 the tech titans as a hobby and it doesn’t feel very secure or very dignified and or stable there’s like
0:42:33 just a lot of reasons why i’m skeptical of that uh in the long term and i don’t think people like it
0:42:40 or want it but on the other hand um just telling people well you’re thrown out into the mix and in
0:42:45 the u.s you have no health insurance and just figure something out that’s also just too cruel and not viable
0:42:51 if it’s a lot of people at once so we have to find our way to a very unfashionable intermediate
0:43:00 sense of social safety network or uh to help people through transitions and right now the accounting for
0:43:04 that is very very difficult to sort out and especially in the united states there’s a deep
0:43:13 hostility to it and i just don’t see logically any other way but then beyond that um i do think new roles
0:43:18 will appear like the the story that well new things will happen and new new things will be possible
0:43:25 i do believe that like there’s a kind of a vague and uncomfortable sense that surely new things will
0:43:31 come along and i i actually think that’s true i don’t feel comfortable making that claim for all
0:43:35 those drivers like we’re not going to retrain them to be programmers because low-level programming
0:43:43 itself is also getting automated right i don’t know exactly how that’ll work um i have thought a great
0:43:49 great deal about it but that’s who i am for the moment i believe that there could be all kinds of
0:43:57 things we don’t foresee and that within that explosion of new sectors of creativity there will be enough new
0:44:04 needs for people to do things if only to train ai’s that it’ll keep up with human needs and support
0:44:08 some kind of a world of economics that’s more distributed than just a central authority
0:44:13 distributing income to everybody which i think would be corrupted yeah yeah i agree with that
0:44:20 do you think we’re being sufficiently intentional about the development of this technology do you
0:44:27 think we’re asking the right questions as a society now well i mean the questions are dominated by
0:44:33 a certain internal technical culture which is and the mainstream of technical culture is very
0:44:39 obsessed with ai as a new god or some kind of new entity and so i think that that does make the
0:44:47 whole conversation go askew and that said we’re almost like if you go to ai conferences
0:44:54 there might there’s usually more talk where somebody is saying we’re going to talk about how to talk the
0:45:01 ai into not killing us you know and that kind of conversation which to me is not well grounded and
0:45:09 i think it kind of loses itself in loops but that kind of conversation can take up as much time and space
0:45:14 as like a serious conversation of like how can we optimize this algorithm or how can we you know like the
0:45:20 the actual work that we should be doing as technologists um i was at one conference i was
0:45:25 kind of funny where i forget what there were these different factions there’s the artificial general
0:45:30 intelligence and there’s the super intelligence and there’s all these different people who have
0:45:34 slightly different ideas about how awesome ai will be and help might kill us all in different ways
0:45:42 and they were so conflicted that they got into a fist fight um a not very competent fist fight it must be
0:45:48 said but i’m shocked it’s kind of funny anyway i sort of wish i had a film of that that was really funny but
0:45:54 i don’t know i mean i love my world i love the people i do kind of make fun of us a little bit
0:46:00 sometimes because i just think it’s important too you know okay so if we just let’s just set aside for
0:46:06 the moment that the more common fears about ai the alignment problem and taking our jobs and
0:46:12 flattening human creativity all that stuff all that is there all of that um is there is there a fear of
0:46:18 yours something you think we could get terribly wrong that’s not currently something we hear much about
0:46:26 uh god i don’t even know where to start yeah there’s like a lot lot lot lot lot lot lot
0:46:38 lot i’m i mean one of the things i worry about is we’re gradually moving education into an ai model
0:46:46 and the motivations for that are often very good because in a lot of places on earth it’s just been
0:46:50 impossible to come up with an economics of supporting and training enough human teachers
0:46:58 and a lot of cultural issues in changing societies make it very very hard to make schools that work
0:47:06 and so on like there’s a lot of issues and in theory a sort of uh client self-adapting ai tutor
0:47:13 could solve a lot of problems at a low cost in a lot of situations but then the issue with that is
0:47:19 once again creativity how do you keep people who learn in a system like that
0:47:24 how do you train them so that they’re able to step outside of what the system was trained on
0:47:29 you know like there’s this funny way that you’re always retreading and recombining the training data
0:47:35 in any ai system and you can address that to a degree with constant fresh input and this and that but
0:47:40 i am a little worried about people being trained in a closed system that makes them a little less than
0:47:46 they might otherwise have been and have a little less faith in themselves i’m a little concerned about
0:47:53 sort of defining the nature of life and education downward you know and the thing is the history
0:47:59 of education is filled with doing exactly that thing like education has been filled with overly
0:48:08 reductive ideas or overly idealistic and and um biased ideas of different kinds i mean so it’s not like
0:48:14 we’re entering this perfect system messing it up we’re entering a messed up system and trying to figure out
0:48:22 how to not perpetuate it it’s messed up itness i think in the case of education um challenging really
0:48:28 challenging i think i just ask just because i’m just curious what you would say i i have a five-year-old
0:48:37 son and he’s already started asking questions about you know like what kind of skills should he learn what
0:48:42 what should he what should he aspire to do in the world oh man that’s a hard one right and i don’t know
0:48:49 what to tell him because i have no idea what the world is going to look like by the time he’s 18 or 20 or 15
0:48:54 hell you know i what would you what would you tell him if uncle jaron came over oh yeah and he asked
0:48:59 you that what would you say well i have a teen daughter now and when she was younger uh she went
0:49:07 to coding camp you know and loved it and then when uh copilot for github came out and now some of the
0:49:12 other ones that are out she was like well you know the kinds of programs i’d write i can just ask for now
0:49:17 so why did you send me to all this thing why did i waste all my time at these things and i said uh remember
0:49:24 you loved coding camp remember you liked it you liked it it’s like well yeah but i would have
0:49:30 i could have liked spelunking camp or something too like why coding camp and um i i mean
0:49:36 i don’t have a perfect answer for all that right now i really don’t i do
0:49:44 i do think there are new things that will emerge i have a feeling there’ll be a lot of new professions
0:49:51 related to adaptive biology and modifications and helping people deal with weird changes to their
0:49:56 bodies that will become possible i think that’ll become a big thing i don’t know exactly how it’s
0:50:04 too early to say like i there’s a subtle point here i want to make which is um i am very far from being
0:50:12 anti-futuristic or disliking extreme change in the future but what i what i have to insist upon
0:50:18 is continuity so in this idea there’s a term called the singularity uh applied to ai sometimes that
0:50:23 there’ll be this rush of change so fast that nobody can learn anything nobody can know anything and it
0:50:29 just is beyond us beyond us beyond us the problem with the singularity whether it’s in a black hole or in
0:50:35 the big bang or in technology is that it’s very hard to have you know like by definition even if
0:50:41 you don’t technically lose information you lose the ability to access the information in the in the
0:50:46 original context or with any kind of structure so it’s essentially a form of massive forgetting and
0:50:54 massive loss of context and massive loss of meaning therefore and so however radical we get if in the future
0:51:00 we’re all going to evolve into massive distributed colonies of space bacteria flying around
0:51:07 and intergalactically or something whatever we turn into i’m all for it i’m in i’m in i’m in but
0:51:12 the line from here to there has to have memory it has to be continuous enough that we’re learning
0:51:19 lessons and we we remember if we break that because we want the thrill of polpot’s year zero where from
0:51:24 now on we’re the smartest people and everybody else was wrong and we start over if we want that break
0:51:29 we must resist it we must oppose people who want that break year zero never works out well it’s a
0:51:38 really really bad idea and so that to me i’m like pro extreme futures but anti discontinuity into the
0:51:43 future and and so that’s a an in-between place to be that’s a little subtle and hard to get across but
0:51:48 i think that that’s the right place to be well i always try to end these conversations with as much
0:51:55 optimism as possible so do you have any other good news or uh rosy scenarios you can you can paint for
0:52:01 us uh before we get out of here about how things are going to be awesome in the future right now we’re
0:52:09 in a very hard to parse moment things are strange things are scary and what i keep on telling myself
0:52:16 there’s always hope in chaos as much as someone might someone driving chaos might be certain that
0:52:25 it’s under their command but it never is and those of us who watch unfolding chaos looking for signs of
0:52:33 hope looking for optimism looking for little openings in which to do something good we will find them if we
0:52:40 stay alert and so i’d urge everybody to do that during this period jaron lanier i’m a fan of your
0:52:45 work i’m a fan of you as a human being as well i appreciate you coming in oh well that’s very kind
0:52:52 of you thank you so much and i really appreciate all the effort and also just the goodwill and warmth
0:53:03 you put into this interview i really do appreciate it so much
0:53:12 all right i hope you enjoyed this episode there was a lot going on in this one jaron is a unique mind
0:53:22 and i appreciate the way he thinks about all of this this conversation did force me to reflect on the
0:53:31 language i use to make sense of ai and all the assumptions buried in that language so i hope you
0:53:39 found his insights useful but either way as always we want to know what you think so drop us a line
0:53:52 at the gray area at vox.com or leave us a message on our new voicemail line at 1-800-214-5749
0:53:58 and once you’re finished with that if you have a second please go ahead and rate and review and
0:54:09 subscribe to the podcast this episode was produced by beth morrissey edited by jorge just engineered by erica
0:54:17 wong fact check by melissa hirsch and alex overington wrote our theme music new episodes of the gray area
0:54:25 drop on mondays listen and subscribe the show is part of vox support vox’s journalism by joining our
0:54:33 membership program today go to vox.com slash members to sign up and if you decide to sign up because of this show
0:54:51 let us know you