Transcript
00:00:00 The following is a conversation with Michael I. Jordan, a professor at Berkeley and one
00:00:05 of the most influential people in the history of machine learning, statistics, and artificial
00:00:10 intelligence.
00:00:11 He has been cited over 170,000 times and he has mentored many of the world class researchers
00:00:17 defining the field of AI today, including Andrew Ng, Zubin Garamani, Ben Taskar, and
00:00:25 Yoshua Bengio.
00:00:27 All this, to me, is as impressive as the over 32,000 points in the six NBA championships
00:00:34 of the Michael J. Jordan of basketball fame.
00:00:38 There’s a nonzero probability that I talked to the other Michael Jordan given my connection
00:00:43 to and love of the Chicago Bulls of the 90s, but if I had to pick one, I’m going with
00:00:48 the Michael Jordan of statistics and computer science, or as Yann LeCun calls him, the Miles
00:00:54 Davis of machine learning.
00:00:56 In his blog post titled Artificial Intelligence, the Revolution Hasn’t Happened Yet, Michael
00:01:01 argues for broadening the scope of the artificial intelligence field.
00:01:05 In many ways, the underlying spirit of this podcast is the same, to see artificial intelligence
00:01:12 as a deeply human endeavor, to not only engineer algorithms and robots, but to understand and
00:01:18 empower human beings at all levels of abstraction, from the individual to our civilization as
00:01:25 a whole.
00:01:26 This is the Artificial Intelligence Podcast.
00:01:29 If you enjoy it, subscribe and YouTube, give it five stars at Apple Podcast, support it
00:01:34 on Patreon, or simply connect with me on Twitter at Lex Friedman spelled F R I D M A N.
00:01:42 As usual, I’ll do one or two minutes of ads now and never any ads in the middle that
00:01:46 can break the flow of the conversation.
00:01:48 I hope that works for you and doesn’t hurt the listening experience.
00:01:54 This show is presented by Cash App, the number one finance app in the App Store.
00:01:58 When you get it, use code LEX PODCAST.
00:02:02 Cash App lets you send money to friends, buy Bitcoin, and invest in the stock market with
00:02:06 as little as $1.
00:02:08 Since Cash App does fractional share trading, let me mention that the order execution algorithm
00:02:13 that worked behind the scenes to create the abstraction of the fractional orders is to
00:02:18 me an algorithmic marvel.
00:02:21 Great props for the Cash App engineers for solving a hard problem that in the end provides
00:02:26 an easy interface that takes a step up to the next layer of abstraction over the stock
00:02:31 market, making trading more accessible for new investors and diversification much easier.
00:02:38 So once again, if you get Cash App from the App Store or Google Play and use the code
00:02:42 LEX PODCAST, you’ll get $10 and Cash App will also donate $10 to First, one of my favorite
00:02:49 organizations that is helping to advance robotics and STEM education for young people around
00:02:55 the world.
00:02:57 And now, here’s my conversation with Michael I. Jordan.
00:03:02 Given that you’re one of the greats in the field of AI, machine learning, computer science,
00:03:06 and so on, you’re trivially called the Michael Jordan of machine learning, although as you
00:03:14 know, you were born first, so technically MJ is the Michael I. Jordan of basketball.
00:03:19 But anyway, my favorite is Yann LeCun calling you the Miles Davis of machine learning because
00:03:25 as he says, you reinvent yourself periodically and sometimes leave fans scratching their
00:03:30 heads after you change direction.
00:03:32 So can you put at first your historian hat on and give a history of computer science
00:03:38 and AI as you saw it, as you experienced it, including the four generations of AI successes
00:03:46 that I’ve seen you talk about?
00:03:47 Sure.
00:03:48 Yeah, first of all, I much prefer Yann’s metaphor.
00:03:54 Miles Davis was a real explorer in jazz and he had a coherent story.
00:04:00 So I think I have one, but it’s not just the one you lived, it’s the one you think about
00:04:04 later.
00:04:05 What the historian does is they look back and they revisit.
00:04:09 I think what’s happening right now is not AI, that was an intellectual aspiration that’s
00:04:16 still alive today as an aspiration.
00:04:18 But I think this is akin to the development of chemical engineering from chemistry or
00:04:22 electrical engineering from electromagnetism.
00:04:25 So if you go back to the 30s or 40s, there wasn’t yet chemical engineering.
00:04:31 There was chemistry, there was fluid flow, there was mechanics and so on.
00:04:35 But people pretty clearly viewed interesting goals to try to build factories that make
00:04:41 chemicals products and do it viably, safely, make good ones, do it at scale.
00:04:48 So people started to try to do that, of course, and some factories worked, some didn’t, some
00:04:52 were not viable, some exploded, but in parallel, developed a whole field called chemical engineering.
00:04:58 Electrical engineering is a field, it’s no bones about it, it has theoretical aspects
00:05:02 to it, it has practical aspects.
00:05:04 It’s not just engineering, quote unquote, it’s the real thing, real concepts are needed.
00:05:09 Same thing with electrical engineering.
00:05:11 There was Maxwell’s equations, which in some sense were everything you know about electromagnetism,
00:05:16 but you needed to figure out how to build circuits, how to build modules, how to put
00:05:19 them together, how to bring electricity from one point to another safely and so on and
00:05:22 so forth.
00:05:23 So a whole field that developed called electrical engineering.
00:05:26 I think that’s what’s happening right now, is that we have a proto field, which is statistics,
00:05:33 more of the theoretical side of it, algorithmic side of computer science, that was enough
00:05:36 to start to build things, but what things?
00:05:39 Systems that bring value to human beings and use human data and mix in human decisions.
00:05:44 The engineering side of that is all ad hoc.
00:05:47 That’s what’s emerging.
00:05:48 In fact, if you wanna call machine learning a field, I think that’s what it is, that it’s
00:05:51 a proto form of engineering based on statistical and computational ideas of previous generations.
00:05:56 But do you think there’s something deeper about AI in his dreams and aspirations as
00:06:01 compared to chemical engineering and electrical engineering?
00:06:03 Well the dreams and aspirations maybe, but those are 500 years from now.
00:06:07 I think that that’s like the Greeks sitting there and saying, it would be neat to get
00:06:10 to the moon someday.
00:06:12 I think we have no clue how the brain does computation.
00:06:16 We’re just a clueless.
00:06:17 We’re even worse than the Greeks on most anything interesting scientifically of our era.
00:06:23 Can you linger on that just for a moment because you stand not completely unique, but a little
00:06:29 bit unique in the clarity of that.
00:06:31 Can you elaborate your intuition of why we’re, like where we stand in our understanding of
00:06:36 the human brain?
00:06:37 And a lot of people say, you know, scientists say we’re not very far in understanding human
00:06:41 brain, but you’re like, you’re saying we’re in the dark here.
00:06:44 Well, I know I’m not unique.
00:06:45 I don’t even think in the clarity, but if you talk to real neuroscientists that really
00:06:49 study real synapses or real neurons, they agree, they agree.
00:06:53 It’s a hundreds of year task and they’re building it up slowly and surely.
00:06:58 What the signal is there is not clear.
00:07:00 We think we have all of our metaphors.
00:07:02 We think it’s electrical, maybe it’s chemical, it’s a whole soup, it’s ions and proteins
00:07:08 and it’s a cell.
00:07:09 And that’s even around like a single synapse.
00:07:11 If you look at a electron micrograph of a single synapse, it’s a city of its own.
00:07:15 And that’s one little thing on a dendritic tree, which is extremely complicated electrochemical
00:07:20 thing.
00:07:22 And it’s doing these spikes and voltages are flying around and then proteins are taking
00:07:25 that and taking it down into the DNA and who knows what.
00:07:29 So it is the problem of the next few centuries.
00:07:31 It is fantastic.
00:07:33 But we have our metaphors about it.
00:07:34 Is it an economic device?
00:07:36 Is it like the immune system or is it like a layered set of, you know, arithmetic computations?
00:07:42 We have all these metaphors and they’re fun.
00:07:44 But that’s not real science per se.
00:07:48 There is neuroscience.
00:07:49 That’s not neuroscience.
00:07:50 All right.
00:07:51 That’s like the Greek speculating about how to get to the moon, fun, right?
00:07:55 And I think that I like to say this fairly strongly because I think a lot of young people
00:07:59 think we’re on the verge because a lot of people who don’t talk about it clearly let
00:08:03 it be understood that, yes, we kind of, this is a brain inspired, we’re kind of close,
00:08:07 you know, breakthroughs are on the horizon.
00:08:10 And that’s scrupulous people sometimes who need money for their labs.
00:08:13 That’s what I’m saying, scrupulous, but people will oversell, I need money for my lab, I’m
00:08:18 studying computational neuroscience, I’m going to oversell it.
00:08:23 And so there’s been too much of that.
00:08:25 So I’ll step into the gray area between metaphor and engineering with, I’m not sure if you’re
00:08:32 familiar with brain computer interfaces.
00:08:35 So a company like Elon Musk has Neuralink that’s working on putting electrodes into
00:08:42 the brain and trying to be able to read, both read and send electrical signals.
00:08:46 Just as you said, even the basic mechanism of communication in the brain is not something
00:08:54 we understand.
00:08:55 But do you hope without understanding the fundamental principles of how the brain works,
00:09:00 we’ll be able to do something interesting at that gray area of metaphor?
00:09:06 It’s not my area.
00:09:07 So I hope in the sense, like anybody else hopes for some interesting things to happen
00:09:11 from research, I would expect more something like Alzheimer’s will get figured out from
00:09:15 modern neuroscience.
00:09:16 There’s a lot of human suffering based on brain disease and we throw things like lithium
00:09:22 at the brain, it kind of works, no one has a clue why.
00:09:25 That’s not quite true, but mostly we don’t know.
00:09:28 And that’s even just about the biochemistry of the brain and how it leads to mood swings
00:09:31 and so on.
00:09:33 How thought emerges from that, we were really, really completely dim.
00:09:38 So that you might want to hook up electrodes and try to do some signal processing on that
00:09:41 and try to find patterns, fine, by all means, go for it.
00:09:45 It’s just not scientific at this point.
00:09:48 So it’s like kind of sitting in a satellite and watching the emissions from a city and
00:09:53 trying to infer things about the microeconomy, even though you don’t have microeconomic concepts.
00:09:57 It’s really that kind of thing.
00:09:59 And so yes, can you find some signals that do something interesting or useful?
00:10:02 Can you control a cursor or mouse with your brain?
00:10:06 Yeah, absolutely, and then I can imagine business models based on that and even medical applications
00:10:13 of that.
00:10:14 But from there to understanding algorithms that allow us to really tie in deeply from
00:10:19 the brain to computer, I just, no, I don’t agree with Elon Musk.
00:10:22 I don’t think that’s even, that’s not for our generations, not even for the century.
00:10:26 So just in hopes of getting you to dream, you’ve mentioned Kolmogorov and Turing might
00:10:33 pop up, do you think that there might be breakthroughs that will get you to sit back in five, 10
00:10:41 years and say, wow?
00:10:43 Oh, I’m sure there will be, but I don’t think that there’ll be demos that impress me.
00:10:49 I don’t think that having a computer call a restaurant and pretend to be a human is
00:10:55 a breakthrough.
00:10:56 Right.
00:10:57 And people, you know, some people present it as such.
00:10:59 It’s imitating human intelligence.
00:11:01 It’s even putting coughs in the thing to make a bit of a PR stunt.
00:11:07 And so fine that the world runs on those things too.
00:11:11 And I don’t want to diminish all the hard work and engineering that goes behind things
00:11:14 like that and the ultimate value to the human race.
00:11:17 But that’s not scientific understanding.
00:11:20 And I know the people that work on these things, they are after scientific understanding.
00:11:23 In the meantime, they’ve got to kind of, you know, the trains got to run and they got mouths
00:11:26 to feed and they got things to do and there’s nothing wrong with all that.
00:11:30 I would call that though, just engineering.
00:11:32 And I want to distinguish that between an engineering field, like electrical engineering
00:11:35 and chemical engineering that originally emerged, that had real principles and you really know
00:11:39 what you’re doing and you had a little scientific understanding, maybe not even complete.
00:11:43 So it became more predictable and it really gave value to human life because it was understood.
00:11:49 And so we don’t want to muddle too much these waters of, you know, what we’re able to do
00:11:54 versus what we really can’t do in a way that’s going to impress the next.
00:11:58 So I don’t need to be wowed, but I think that someone comes along in 20 years, a younger
00:12:02 person who’s absorbed all the technology and for them to be wowed, I think they have to
00:12:08 be more deeply impressed.
00:12:09 A young Kolmogorov would not be wowed by some of the stunts that you see right now coming
00:12:13 from the big companies.
00:12:14 The demos, but do you think the breakthroughs from Kolmogorov would be, and give this question
00:12:19 a chance, do you think there’ll be in the scientific fundamental principles arena or
00:12:24 do you think it’s possible to have fundamental breakthroughs in engineering?
00:12:28 Meaning, you know, I would say some of the things that Elon Musk is working with SpaceX
00:12:33 and then others sort of trying to revolutionize the fundamentals of engineering, of manufacturing,
00:12:39 of saying, here’s a problem we know how to do a demo of and actually taking it to scale.
00:12:44 Yeah.
00:12:45 So there’s going to be all kinds of breakthroughs.
00:12:46 I just don’t like that terminology.
00:12:48 I’m a scientist and I work on things day in and day out and things move along and eventually
00:12:52 you say, wow, something happened, but I don’t like that language very much.
00:12:56 Also I don’t like to prize theoretical breakthroughs over practical ones.
00:13:01 I tend to be more of a theoretician and I think there’s lots to do in that arena right
00:13:05 now.
00:13:06 And so I wouldn’t point to the Kolmogorovs, I might point to the Edisons of the era and
00:13:09 maybe Musk is a bit more like that.
00:13:11 But you know, Musk, God bless him, also will say things about AI that he knows very little
00:13:17 about and he leads people astray when he talks about things he doesn’t know anything about.
00:13:23 Trying to program a computer to understand natural language, to be involved in a dialogue
00:13:27 we’re having right now, that ain’t going to happen in our lifetime.
00:13:30 You could fake it, you can mimic, sort of take old sentences that humans use and retread
00:13:35 them, but the deep understanding of language, no, it’s not going to happen.
00:13:38 And so from that, I hope you can perceive that the deeper, yet deeper kind of aspects
00:13:42 and intelligence are not going to happen.
00:13:44 Now will there be breakthroughs?
00:13:45 No, I think that Google was a breakthrough, I think Amazon is a breakthrough, you know,
00:13:49 I think Uber is a breakthrough, you know, that bring value to human beings at scale
00:13:53 in new, brand new ways based on data flows and so on.
00:13:56 A lot of these things are slightly broken because there’s not kind of an engineering
00:14:01 field that takes economic value in context of data and, you know, planetary scale and
00:14:06 worries about all the externalities, the privacy, you know, we don’t have that field so we don’t
00:14:11 think these things through very well.
00:14:13 I see that as emerging and that will be, you know, looking back from 100 years, that will
00:14:17 be a constituted breakthrough in this era, just like electrical engineering was a breakthrough
00:14:21 in the early part of the last century and chemical engineering was a breakthrough.
00:14:24 So the scale, the markets that you talk about and we’ll get to will be seen as sort of breakthrough
00:14:30 and we’re in the very early days of really doing interesting stuff there and we’ll get
00:14:34 to that, but just taking a quick step back, can you give, kind of throw off the historian
00:14:40 hat.
00:14:41 I mean, you briefly said that the history of AI kind of mimics the history of chemical
00:14:47 engineering, but…
00:14:49 I keep saying machine learning.
00:14:50 You keep wanting to say AI, just to let you know, I don’t, you know, I resist that.
00:14:54 I don’t think this is about AI really was John McCarthy as almost a philosopher saying,
00:15:01 wouldn’t it be cool if we could put thought in a computer?
00:15:03 If we could mimic the human capability to think or put intelligence in, in some sense
00:15:08 into a computer.
00:15:09 That’s an interesting philosophical question and he wanted to make it more than philosophy.
00:15:13 He wanted to actually write down a logical formula and algorithms that would do that.
00:15:17 And that is a perfectly valid, reasonable thing to do.
00:15:20 That’s not what’s happening in this era.
00:15:23 So the reason I keep saying AI actually, and I’d love to hear what you think about it.
00:15:27 Machine learning has a very particular set of methods and tools.
00:15:34 Maybe your version of it is that mine doesn’t, it’s very, very open.
00:15:37 It does optimization, it does sampling, it does…
00:15:40 So systems that learn is what machine learning is.
00:15:42 Systems that learn and make decisions.
00:15:44 And make decisions.
00:15:45 So it’s not just pattern recognition and, you know, finding patterns, it’s all about
00:15:49 making decisions in real worlds and having close feedback loops.
00:15:52 So something like symbolic AI, expert systems, reasoning systems, knowledge based representation,
00:15:58 all of those kinds of things, search, does that neighbor fit into what you think of as
00:16:03 machine learning?
00:16:04 So I don’t even like the word machine learning, I think that what the field you’re talking
00:16:07 about is all about making large collections of decisions under uncertainty by large collections
00:16:11 of entities.
00:16:12 Right?
00:16:13 And there are principles for that, at that scale.
00:16:16 You don’t have to say the principles are for a single entity that’s making decisions, single
00:16:19 agent or single human.
00:16:20 It really immediately goes to the network of decisions.
00:16:22 Is a good word for that or no?
00:16:24 No, there’s no good words for any of this.
00:16:25 That’s kind of part of the problem.
00:16:27 So we can continue the conversation to use AI for all that.
00:16:29 I just want to kind of raise the flag here that this is not about, we don’t know what
00:16:35 intelligence is and real intelligence.
00:16:38 We don’t know much about abstraction and reasoning at the level of humans.
00:16:41 We don’t have a clue.
00:16:42 We’re not trying to build that because we don’t have a clue.
00:16:44 Eventually it may emerge.
00:16:45 They’ll make, I don’t know if there’ll be breakthroughs, but eventually we’ll start
00:16:48 to get glimmers of that.
00:16:50 It’s not what’s happening right now.
00:16:51 Okay.
00:16:52 We’re taking data.
00:16:53 We’re trying to make good decisions based on that.
00:16:54 We’re trying to scale.
00:16:55 We’re trying to economically viably, we’re trying to build markets.
00:16:58 We’re trying to keep value at that scale and aspects of this will look intelligent.
00:17:04 Computers were so dumb before, they will seem more intelligent.
00:17:08 We will use that buzzword of intelligence so we can use it in that sense.
00:17:12 So machine learning, you can scope it narrowly as just learning from data and pattern recognition.
00:17:17 But when I talk about these topics, maybe data science is another word you could throw
00:17:22 in the mix, it really is important that the decisions are as part of it.
00:17:26 It’s consequential decisions in the real world.
00:17:28 Am I going to have a medical operation?
00:17:30 Am I going to drive down the street?
00:17:33 Things where there’s scarcity, things that impact other human beings or other environments
00:17:38 and so on.
00:17:39 How do I do that based on data?
00:17:40 How do I do that adaptively?
00:17:41 How do I use computers to help those kinds of things go forward?
00:17:44 Whatever you want to call that.
00:17:45 So let’s call it AI.
00:17:46 Let’s agree to call it AI, but let’s not say that the goal of that is intelligence.
00:17:52 The goal of that is really good working systems at planetary scale that we’ve never seen before.
00:17:56 So reclaim the word AI from the Dartmouth conference from many decades ago of the dream
00:18:00 of humans.
00:18:01 I don’t want to reclaim it.
00:18:02 I want a new word.
00:18:03 I think it was a bad choice.
00:18:04 I mean, if you read one of my little things, the history was basically that McCarthy needed
00:18:09 a new name because cybernetics already existed and he didn’t like, no one really liked Norbert
00:18:14 Wiener.
00:18:15 Norbert Wiener was kind of an island to himself and he felt that he had encompassed all this
00:18:19 and in some sense he did.
00:18:21 You look at the language of cybernetics, it was everything we’re talking about.
00:18:24 It was control theory and signal processing and some notions of intelligence and closed
00:18:28 feedback loops and data.
00:18:29 It was all there.
00:18:30 It’s just not a word that lived on partly because of the maybe the personalities.
00:18:34 But McCarthy needed a new word to say, I’m different from you.
00:18:36 I’m not part of your show.
00:18:38 I got my own.
00:18:40 Invented this word and again, thinking forward about the movies that would be made about
00:18:46 it, it was a great choice.
00:18:48 But thinking forward about creating a sober academic and real world discipline, it was
00:18:52 a terrible choice because it led to promises that are not true that we understand.
00:18:56 We understand artificial perhaps, but we don’t understand intelligence.
00:18:58 It’s a small tangent because you’re one of the great personalities of machine learning,
00:19:03 whatever the heck you call the field.
00:19:06 Do you think science progresses by personalities or by the fundamental principles and theories
00:19:11 and research that’s outside of personalities?
00:19:15 Both.
00:19:16 And I wouldn’t say there should be one kind of personality.
00:19:17 I have mine and I have my preferences and I have a kind of network around me that feeds
00:19:23 me and some of them agree with me and some of them disagree, but all kinds of personalities
00:19:26 are needed.
00:19:28 Right now, I think the personality that it’s a little too exuberant, a little bit too ready
00:19:31 to promise the moon is a little bit too much in ascendance.
00:19:35 And I do think that there’s some good to that.
00:19:38 It certainly attracts lots of young people to our field, but a lot of those people come
00:19:41 in with strong misconceptions and they have to then unlearn those and then find something
00:19:47 to do.
00:19:48 And so I think there’s just got to be some multiple voices and I wasn’t hearing enough
00:19:52 of the more sober voice.
00:19:54 So as a continuation of a fun tangent and speaking of vibrant personalities, what would
00:20:02 you say is the most interesting disagreement you have with Jan Lacune?
00:20:07 So Jan’s an old friend and I just say that I don’t think we disagree about very much
00:20:12 really.
00:20:13 He and I both kind of have a let’s build it kind of mentality and does it work kind of
00:20:18 mentality and kind of concrete.
00:20:21 We both speak French and we speak French more together and we have a lot in common.
00:20:27 And so if one wanted to highlight a disagreement, it’s not really a fundamental one.
00:20:31 I think it’s just kind of what we’re emphasizing.
00:20:35 Jan has emphasized pattern recognition and has emphasized prediction.
00:20:43 And it’s interesting to try to take that as far as you can.
00:20:45 If you could do perfect prediction, what would that give you kind of as a thought experiment?
00:20:50 And I think that’s way too limited.
00:20:55 We cannot do perfect prediction.
00:20:56 We will never have the data sets that allow me to figure out what you’re about ready to
00:20:59 do, what question you’re going to ask next.
00:21:00 I have no clue.
00:21:01 I will never know such things.
00:21:03 Moreover, most of us find ourselves during the day in all kinds of situations we had
00:21:07 no anticipation of that are kind of very, very novel in various ways.
00:21:13 And in that moment, we want to think through what we want.
00:21:16 And also there’s going to be market forces acting on us.
00:21:19 I’d like to go down that street, but now it’s full because there’s a crane in the street.
00:21:22 I got it.
00:21:23 I got to think about that.
00:21:24 I got to think about what I might really want here.
00:21:26 And I got to sort of think about how much it costs me to do this action versus this
00:21:29 action.
00:21:30 I got to think about the risks involved.
00:21:32 A lot of our current pattern recognition and prediction systems don’t do any risk evaluations.
00:21:37 They have no error bars, right?
00:21:39 I got to think about other people’s decisions around me.
00:21:41 I got to think about a collection of my decisions, even just thinking about like a medical treatment,
00:21:45 you know, I’m not going to take a, the prediction of a neural net about my health, about something
00:21:50 consequential.
00:21:51 I’m not about ready to have a heart attack because some number is over 0.7.
00:21:54 Even if you had all the data in the world that ever been collected about heart attacks
00:21:58 better than any doctor ever had, I’m not going to trust the output of that neural net to
00:22:02 predict my heart attack.
00:22:03 I’m going to want to ask what if questions around that.
00:22:06 I’m going to want to look at some us or other possible data I didn’t have, causal things.
00:22:10 I’m going to want to have a dialogue with a doctor about things we didn’t think about
00:22:13 when he gathered the data.
00:22:15 You know, I could go on and on.
00:22:16 I hope you can see.
00:22:17 And I don’t, I think that if you say predictions, everything that, that, that you’re missing
00:22:21 all of this stuff.
00:22:23 And so prediction plus decision making is everything, but both of them are equally important.
00:22:28 And so the field has emphasized prediction, Jan rightly so has seen how powerful that
00:22:32 is.
00:22:33 But at the cost of people not being aware that decision making is where the rubber really
00:22:37 hits the road, where human lives are at stake, where risks are being taken, where you got
00:22:41 to gather more data.
00:22:42 You got to think about the error bars.
00:22:43 You got to think about the consequences of your decisions on others.
00:22:45 You got to think about the economy around your decisions, blah, blah, blah, blah.
00:22:48 I’m not the only one working on those, but we’re a smaller tribe.
00:22:52 And right now we’re not the one that people talk about the most.
00:22:56 But you know, if you go out in the real world and industry, you know, at Amazon, I’d say
00:23:00 half the people there are working on decision making and the other half are doing, you know,
00:23:03 the pattern recognition.
00:23:04 It’s important.
00:23:05 And the words of pattern recognition and prediction, I think the distinction there, not to linger
00:23:10 on words, but the distinction there is more a constrained sort of in the lab data set
00:23:16 versus decision making is talking about consequential decisions in the real world, under the messiness
00:23:21 and the uncertainty of the real world.
00:23:23 And just the whole of it, the whole mess of it that actually touches human beings and
00:23:27 scale.
00:23:28 And the forces, that’s the distinction.
00:23:31 It helps add those, that perspective, that broader perspective.
00:23:33 You’re right.
00:23:34 I totally agree.
00:23:35 On the other hand, if you’re a real prediction person, of course, you want it to be in the
00:23:38 real world.
00:23:39 You want to predict real world events.
00:23:40 I’m just saying that’s not possible with just data sets.
00:23:43 That it has to be in the context of, you know, strategic things that someone’s doing, data
00:23:47 they might gather, things they could have gathered, the reasoning process around data.
00:23:50 It’s not just taking data and making predictions based on the data.
00:23:53 So one of the things that you’re working on, I’m sure there’s others working on it, but
00:23:58 I don’t hear often it talked about, especially in the clarity that you talk about it, and
00:24:04 I think it’s both the most exciting and the most concerning area of AI in terms of decision
00:24:10 making.
00:24:11 So you’ve talked about AI systems that help make decisions that scale in a distributed
00:24:15 way, millions, billions decisions, sort of markets of decisions.
00:24:19 Can you, as a starting point, sort of give an example of a system that you think about
00:24:24 when you’re thinking about these kinds of systems?
00:24:27 Yeah, so first of all, you’re absolutely getting into some territory, which I will be beyond
00:24:31 my expertise.
00:24:32 And there are lots of things that are going to be very not obvious to think about.
00:24:35 Just like, again, I like to think about history a little bit, but think about put yourself
00:24:39 back in the sixties.
00:24:40 There was kind of a banking system that wasn’t computerized really.
00:24:43 There was database theory emerging and database people had to think about how do I actually
00:24:48 not just move data around, but actual money and have it be, you know, valid and have transactions
00:24:53 that ATMs happen that are actually, you know, all valid and so on and so forth.
00:24:57 So that’s the kind of issues you get into when you start to get serious about sorts
00:25:01 of things like this.
00:25:02 I like to think about as kind of almost a thought experiment to help me think something
00:25:07 simpler, which is the music market.
00:25:11 And because there is, to first order, there is no music market in the world right now
00:25:16 and in our country, for sure.
00:25:18 There are something called things called record companies and they make money and they prop
00:25:23 up a few really good musicians and make them superstars and they all make huge amounts
00:25:29 of money.
00:25:30 But there’s a long tail of huge numbers of people that make lots and lots of really good
00:25:33 music that is actually listened to by more people than the famous people.
00:25:40 They are not in a market.
00:25:41 They cannot have a career.
00:25:42 They do not make money.
00:25:43 The creators, the creators, the creators, the so called influencers or whatever that
00:25:47 diminishes who they are.
00:25:49 So there are people who make extremely good music, especially in the hip hop or Latin
00:25:53 world these days.
00:25:55 They do it on their laptop.
00:25:56 That’s what they do on the weekend and they have another job during the week and they
00:26:01 put it up on SoundCloud or other sites.
00:26:03 Eventually it gets streamed.
00:26:04 It now gets turned into bits.
00:26:06 It’s not economically valuable.
00:26:07 The information is lost.
00:26:08 It gets put up there.
00:26:10 People stream it.
00:26:11 You walk around in a big city, you see people with headphones, especially young kids listening
00:26:16 to music all the time.
00:26:17 If you look at the data, very little of the music they are listening to is the famous
00:26:21 people’s music and none of it’s old music.
00:26:23 It’s all the latest stuff.
00:26:24 But the people who made that latest stuff are like some 16 year old somewhere who will
00:26:27 never make a career out of this, who will never make money.
00:26:29 Of course there will be a few counter examples.
00:26:31 The record companies incentivize to pick out a few and highlight them.
00:26:35 Long story short, there’s a missing market there.
00:26:37 There is not a consumer producer relationship at the level of the actual creative acts.
00:26:43 The pipelines and Spotify’s of the world that take this stuff and stream it along, they
00:26:48 make money off of subscriptions or advertising and those things.
00:26:51 They’re making the money.
00:26:52 All right.
00:26:53 And then they will offer bits and pieces of it to a few people again to highlight that
00:26:55 they simulate a market.
00:26:58 Anyway, a real market would be if you’re a creator of music that you actually are somebody
00:27:03 who’s good enough that people want to listen to you, you should have the data available
00:27:07 to you.
00:27:08 There should be a dashboard showing a map of the United States.
00:27:11 So in last week, here’s all the places your songs were listened to.
00:27:14 It should be transparent, vetable, so that if someone down in Providence sees that you’re
00:27:20 being listened to 10,000 times in Providence, that they know that’s real data.
00:27:24 You know it’s real data.
00:27:25 They will have you come give a show down there.
00:27:27 They will broadcast to the people who’ve been listening to you that you’re coming.
00:27:30 If you do this right, you could go down there and make $20,000.
00:27:34 You do that three times a year, you start to have a career.
00:27:37 So in this sense, AI creates jobs.
00:27:39 It’s not about taking away human jobs.
00:27:40 It’s creating new jobs because it creates a new market.
00:27:43 Once you’ve created a market, you’ve now connected up producers and consumers.
00:27:46 The person who’s making the music can say to someone who comes to their shows a lot,
00:27:50 hey, I’ll play at your daughter’s wedding for $10,000.
00:27:53 You’ll say 8,000.
00:27:54 They’ll say 9,000.
00:27:55 Then again, you can now get an income up to $100,000.
00:27:59 You’re not going to be a millionaire.
00:28:01 And now even think about really the value of music is in these personal connections,
00:28:06 even so much so that a young kid wants to wear a tshirt with their favorite musician’s
00:28:13 signature on it.
00:28:14 So if they listen to the music on the internet, the internet should be able to provide them
00:28:18 with a button that they push and the merchandise arrives the next day.
00:28:21 We can do that.
00:28:23 And now why should we do that?
00:28:24 Well, because the kid who bought the shirt will be happy, but more the person who made
00:28:27 the music will get the money.
00:28:29 There’s no advertising needed.
00:28:32 So you can create markets between producers and consumers, take 5% cut.
00:28:36 Your company will be perfectly sound.
00:28:39 It’ll go forward into the future and it will create new markets and that raises human happiness.
00:28:45 Now this seems like, well, this is easy, just create this dashboard, kind of create some
00:28:48 connections and all that.
00:28:49 But if you think about Uber or whatever, you think about the challenges in the real world
00:28:52 of doing things like this, and there are actually new principles going to be needed.
00:28:56 You’re trying to create a new kind of two way market at a different scale that’s ever
00:28:59 been done before.
00:29:00 There’s going to be unwanted aspects of the market.
00:29:04 There’ll be bad people.
00:29:05 There’ll be the data will get used in the wrong ways, it’ll fail in some ways, it won’t
00:29:10 deliver about.
00:29:11 You have to think that through.
00:29:12 Just like anyone who ran a big auction or ran a big matching service in economics will
00:29:17 think these things through.
00:29:18 And so that maybe doesn’t get at all the huge issues that can arise when you start to create
00:29:22 markets, but it starts to, at least for me, solidify my thoughts and allow me to move
00:29:26 forward in my own thinking.
00:29:28 Yeah.
00:29:29 So I talked to the head of research at Spotify actually, and I think their longterm goal,
00:29:32 they’ve said, is to have at least one million creators make a comfortable living putting
00:29:39 on Spotify.
00:29:41 So I think you articulate a really nice vision of the world and the digital and the cyberspace
00:29:52 of markets.
00:29:53 What do you think companies like Spotify or YouTube or Netflix can do to create such markets?
00:30:04 Is it an AI problem?
00:30:05 Is it an interface problem for interface design?
00:30:08 Is it some other kind of, is it an economics problem?
00:30:13 Who should they hire to solve these problems?
00:30:15 Well, part of it’s not just top down.
00:30:17 So the Silicon Valley has this attitude that they know how to do it.
00:30:20 They will create the system just like Google did with the search box that will be so good
00:30:23 that they’ll just, everyone will adopt that.
00:30:27 It’s everything you said, but really I think missing that kind of culture.
00:30:31 So it’s literally that 16 year old who’s able to create the songs.
00:30:34 You don’t create that as a Silicon Valley entity.
00:30:37 You don’t hire them per se.
00:30:39 You have to create an ecosystem in which they are wanted and that they belong.
00:30:44 And so you have to have some cultural credibility to do things like this.
00:30:47 Netflix, to their credit, wanted some of that credibility and they created shows, content.
00:30:53 They call it content.
00:30:54 It’s such a terrible word, but it’s culture.
00:30:56 And so with movies, you can kind of go give a large sum of money to somebody graduating
00:31:01 from the USC film school.
00:31:03 It’s a whole thing of its own, but it’s kind of like rich white people’s thing to do.
00:31:07 And American culture has not been so much about rich white people.
00:31:11 It’s been about all the immigrants, all the Africans who came and brought that culture
00:31:16 and those rhythms to this world and created this whole new thing.
00:31:23 American culture.
00:31:24 And so companies can’t artificially create that.
00:31:26 They can’t just say, hey, we’re here.
00:31:28 We’re going to buy it up.
00:31:29 You’ve got a partner.
00:31:31 And so anyway, not to denigrate, these companies are all trying and they should, and I’m sure
00:31:37 they’re asking these questions and some of them are even making an effort.
00:31:40 But it is partly a respect the culture as a technology person.
00:31:44 You’ve got to blend your technology with cultural meaning.
00:31:49 How much of a role do you think the algorithm, so machine learning has in connecting the
00:31:54 consumer to the creator, sort of the recommender system aspect of this?
00:31:59 Yeah.
00:32:00 It’s a great question.
00:32:01 I think pretty high.
00:32:02 There’s no magic in the algorithms, but a good recommender system is way better than
00:32:07 a bad recommender system.
00:32:09 And recommender systems is a billion dollar industry back even 10, 20 years ago.
00:32:15 And it continues to be extremely important going forward.
00:32:17 What’s your favorite recommender system, just so we can put something, well, just historically
00:32:20 I was one of the, when I first went to Amazon, I first didn’t like Amazon because they put
00:32:24 the book people out of business or the library, the local booksellers went out of business.
00:32:30 I’ve come to accept that there probably are more books being sold now and poor people
00:32:34 reading them than ever before.
00:32:36 And then local book stores are coming back.
00:32:39 So that’s how economics sometimes work.
00:32:41 You go up and you go down.
00:32:44 But anyway, when I finally started going there and I bought a few books, I was really pleased
00:32:48 to see another few books being recommended to me that I never would have thought of.
00:32:52 And I bought a bunch of them.
00:32:53 So they obviously had a good business model.
00:32:55 But I learned things and I still to this day kind of browse using that service.
00:33:00 And I think lots of people get a lot, that is a good aspect of a recommendation system.
00:33:05 I’m learning from my peers in an indirect way.
00:33:10 And their algorithms are not meant to have them impose what we learn.
00:33:13 It really is trying to find out what’s in the data.
00:33:16 It doesn’t work so well for other kinds of entities, but that’s just the complexity of
00:33:19 human life.
00:33:20 Like shirts, I’m not going to get recommendations on shirts, but that’s interesting.
00:33:26 If you try to recommend restaurants, it’s hard.
00:33:32 It’s hard to do it at scale.
00:33:35 But a blend of recommendation systems with other economic ideas, matchings and so on
00:33:42 is really, really still very open research wise.
00:33:45 And there’s new companies that are going to emerge that do that well.
00:33:48 What do you think is going to the messy, difficult land of say politics and things like that,
00:33:54 that YouTube and Twitter have to deal with in terms of recommendation systems?
00:33:58 Being able to suggest, I think Facebook just launched Facebook news.
00:34:03 So recommend the kind of news that are most likely for you to be interesting.
00:34:08 Do you think this is AI solvable, again, whatever term we want to use, do you think it’s a solvable
00:34:14 problem for machines or is it a deeply human problem that’s unsolvable?
00:34:18 So I don’t even think about it at that level.
00:34:20 I think that what’s broken with some of these companies, it’s all monetization by advertising.
00:34:25 They’re not, at least Facebook, I want to critique them, but they didn’t really try
00:34:29 to connect a producer and a consumer in an economic way, right?
00:34:32 No one wants to pay for anything.
00:34:34 And so they all, you know, starting with Google and Facebook, they went back to the playbook
00:34:38 of, you know, the television companies back in the day.
00:34:41 No one wanted to pay for this signal.
00:34:43 They will pay for the TV box, but not for the signal, at least back in the day.
00:34:47 And so advertising kind of filled that gap and advertising was new and interesting and
00:34:50 it somehow didn’t take over our lives quite, right?
00:34:54 Fast forward, Google provides a service that people don’t want to pay for.
00:34:59 And so somewhat surprisingly in the nineties, they made, they ended up making huge amounts
00:35:04 so they cornered the advertising market.
00:35:05 It didn’t seem like that was going to happen, at least to me.
00:35:08 These little things on the right hand side of the screen just did not seem all that economically
00:35:11 interesting, but that companies had maybe no other choice.
00:35:14 The TV market was going away and billboards and so on.
00:35:17 So they’ve, they got it.
00:35:19 And I think that sadly that Google just has, it was doing so well with that at making such
00:35:24 money.
00:35:25 They didn’t think much more about how, wait a minute, is there a producer consumer relationship
00:35:28 to be set up here?
00:35:29 Not just between us and the advertisers market to be created.
00:35:32 Is there an actual market between the producer consumer?
00:35:35 They’re the producers, the person who created that video clip, the person that made that
00:35:38 website, the person who could make more such things, the person who could adjust it as
00:35:42 a function of demand, the person on the other side who’s asking for different kinds of things,
00:35:46 you know?
00:35:47 So you see glimmers of that now there’s influencers and there’s kind of a little glimmering of
00:35:51 a market, but it should have been done 20 years ago.
00:35:53 It should have been thought about.
00:35:54 It should have been created in parallel with the advertising ecosystem.
00:35:58 And then Facebook inherited that.
00:35:59 And I think they also didn’t think very much about that.
00:36:03 So fast forward and now they are making huge amounts of money off of advertising.
00:36:07 And the news thing and all these clicks is just feeding the advertising.
00:36:11 It’s all connected up to the advertiser.
00:36:13 So you want more people to click on certain things because that money flows to you, Facebook.
00:36:18 You’re very much incentivized to do that.
00:36:20 And when you start to find it’s breaking, people are telling you, well, we’re getting
00:36:23 into some troubles.
00:36:24 You try to adjust it with your smart AI algorithms, right?
00:36:27 And figure out what are bad clicks.
00:36:28 So maybe it shouldn’t be click through rate, it should be something else.
00:36:31 I find that pretty much hopeless.
00:36:34 It does get into all the complexity of human life and you can try to fix it.
00:36:37 You should, but you could also fix the whole business model.
00:36:40 And the business model is that really, what are, are there some human producers and consumers
00:36:44 out there?
00:36:45 Is there some economic value to be liberated by connecting them directly?
00:36:48 Is it such that it’s so valuable that people will be able to pay for it?
00:36:52 All right.
00:36:53 And micro payments, like small payments.
00:36:54 Micro, but even have to be micro.
00:36:56 So I like the example, suppose I’m going, next week I’m going to India.
00:37:00 Never been to India before.
00:37:01 Right?
00:37:02 I have a couple of days in Mumbai, I have no idea what to do there.
00:37:06 Right?
00:37:07 And I could go on the web right now and search.
00:37:08 It’s going to be kind of hopeless.
00:37:10 I’m not going to find, you know, I have lots of advertisers in my face.
00:37:14 Right?
00:37:15 What I really want to do is broadcast to the world that I am going to Mumbai and have someone
00:37:19 on the other side of a market look at me and, and there’s a recommendation system there.
00:37:24 So I’m not looking at all possible people coming to Mumbai.
00:37:26 They’re looking at the people who are relevant to them.
00:37:27 So someone in my age group, someone who kind of knows me in some level, I give up a little
00:37:32 privacy by that, but I’m happy because what I’m going to get back is this person can make
00:37:35 a little video for me, or they’re going to write a little two page paper on here’s the
00:37:39 cool things that you want to do and move by this week, especially, right?
00:37:43 I’m going to look at that.
00:37:44 I’m not going to pay a micro payment.
00:37:45 I’m going to pay, you know, a hundred dollars or whatever for that.
00:37:48 It’s real value.
00:37:49 It’s like journalism.
00:37:50 Um, and as an honest subscription, it’s that I’m going to pay that person in that moment.
00:37:54 Company’s going to take 5% of that.
00:37:56 And that person has now got it.
00:37:57 It’s a gig economy, if you will, but you know, done for it, you know, thinking about a little
00:38:01 bit behind YouTube, there was actually people who could make more of those things.
00:38:05 If they were connected to a market, they would make more of those things independently.
00:38:07 You don’t have to tell them what to do.
00:38:08 You don’t have to incentivize them any other way.
00:38:11 Um, and so, yeah, these companies, I don’t think have thought long and hard about that.
00:38:15 So I do distinguish on Facebook on the one side, who just not thought about these things
00:38:20 at all.
00:38:21 I think, uh, thinking that AI will fix everything, uh, and Amazon thinks about them all the time
00:38:25 because they were already out in the real world.
00:38:26 They were delivering packages, people’s doors.
00:38:28 They were, they were worried about a market.
00:38:29 They were worried about sellers and, you know, they worry and some things they do are great.
00:38:32 Some things maybe not so great, but you know, they’re in that business model.
00:38:36 And then I’d say Google sort of hovers somewhere in between.
00:38:38 I don’t, I don’t think for a long, long time they got it.
00:38:41 I think they probably see that YouTube is more pregnant with possibility than, than,
00:38:45 than they might’ve thought and that they’re probably heading that direction.
00:38:49 Um, but uh, you know, Silicon Valley has been dominated by the Google Facebook kind of mentality
00:38:54 and the subscription and advertising and that is, that’s the core problem, right?
00:38:58 The fake news actually rides on top of that because it means that you’re monetizing with
00:39:03 clip through rate and that is the core problem.
00:39:05 You got to remove that.
00:39:06 So advertisement, if we’re going to linger on that, I mean, that’s an interesting thesis.
00:39:11 I don’t know if everyone really deeply thinks about that.
00:39:15 So you’re right.
00:39:16 The thought is the advertising model is the only thing we have, the only thing we’ll ever
00:39:20 have.
00:39:21 We have to fix, we have to build algorithms that despite that business model, you know,
00:39:30 find the better angels of our nature and do good by society and by the individual.
00:39:34 But you think we can slowly, you think, first of all, there’s a difference between should
00:39:40 and could.
00:39:42 So you’re saying we should slowly move away from the advertising model and have a direct
00:39:46 connection between the consumer and the creator.
00:39:49 The question I also have is, can we, because the advertising model is so successful now
00:39:55 in terms of just making a huge amount of money and therefore being able to build a big company
00:40:00 that provides, has really smart people working that create a good service.
00:40:03 Do you think it’s possible?
00:40:05 And just to clarify, you think we should move away?
00:40:07 Well, I think we should.
00:40:08 Yeah.
00:40:09 But we is the, you know, me.
00:40:10 So society.
00:40:11 Yeah.
00:40:12 Well, the companies, I mean, so first of all, full disclosure, I’m doing a day a week at
00:40:16 Amazon because I kind of want to learn more about how they do things.
00:40:18 So, you know, I’m not speaking for Amazon in any way, but, you know, I did go there
00:40:22 because I actually believe they get a little bit of this or trying to create these markets.
00:40:26 And they don’t really use, advertising is not a crucial part of it.
00:40:29 Well, that’s a good question.
00:40:30 So it has become not crucial, but it’s become more and more present if you go to Amazon
00:40:34 website.
00:40:35 And, you know, without revealing too many deep secrets about Amazon, I can tell you
00:40:38 that, you know, a lot of people in the company question this and there’s a huge questioning
00:40:42 going on.
00:40:43 You do not want a world where there’s zero advertising.
00:40:45 That actually is a bad world.
00:40:47 Okay.
00:40:48 So here’s a way to think about it.
00:40:49 You’re a company that like Amazon is trying to bring products to customers, right?
00:40:55 And the customer, at any given moment, you want to buy a vacuum cleaner, say, you want
00:40:58 to know what’s available for me.
00:40:59 And, you know, it’s not going to be that obvious.
00:41:00 You have to do a little bit of work at it.
00:41:02 The recommendation system will sort of help, right?
00:41:04 But now suppose this other person over here has just made the world, you know, they spent
00:41:08 a huge amount of energy.
00:41:09 They had a great idea.
00:41:10 They made a great vacuum cleaner.
00:41:11 They know they really did it.
00:41:12 They nailed it.
00:41:13 It’s an MIT, you know, whiz kid that made a great new vacuum cleaner, right?
00:41:16 It’s not going to be in the recommendation system.
00:41:18 No one will know about it.
00:41:19 The algorithms will not find it and AI will not fix that.
00:41:22 Okay.
00:41:23 At all.
00:41:24 Right.
00:41:25 How do you allow that vacuum cleaner to start to get in front of people, be sold well advertising.
00:41:30 And here, what advertising is, it’s a signal that you’re, you believe in your product enough
00:41:35 that you’re willing to pay some real money for it.
00:41:37 And to me as a consumer, I look at that signal.
00:41:39 I say, well, first of all, I know these are not just cheap little ads cause we have now
00:41:43 right now there.
00:41:44 I know that, you know, these are super cheap, you know, pennies.
00:41:47 If I see an ad where it’s actually, I know the company is only doing a few of these and
00:41:51 they’re making, you know, real money is kind of flowing and I see an ad, I may pay more
00:41:54 attention to it.
00:41:55 And I actually might want that because I see, Hey, that guy spent money on his vacuum cleaner.
00:42:01 Maybe there’s something good there.
00:42:02 So I will look at it.
00:42:03 And so that’s part of the overall information flow in a good market.
00:42:06 So advertising has a role, but the problem is of course that that signal is now completely
00:42:11 gone because it just, you know, dominant by these tiny little things that add up to big
00:42:15 money for the company, you know?
00:42:17 So I think it will just, I think it will change because the societies just don’t, you know,
00:42:22 stick with things that annoy a lot of people and advertising currently annoys people more
00:42:26 than it provides information.
00:42:28 And I think that a Google probably is smart enough to figure out that this is a dead,
00:42:32 this is a bad model, even though it’s a hard, huge amount of money and they’ll have to figure
00:42:35 out how to pull it away from it slowly.
00:42:38 And I’m sure the CEO there will figure it out, but they need to do it.
00:42:42 And they needed it to, so if you reduce advertising, not to zero, but you reduce it at the same
00:42:47 time you bring up producer, consumer, actual real value being delivered.
00:42:51 So real money is being paid and they take a 5% cut that 5% could start to get big enough
00:42:56 to cancel out the lost revenue from the kind of the poor kind of advertising.
00:43:00 And I think that a good company will do that, will realize that.
00:43:04 And Facebook, you know, again, God bless them.
00:43:08 They bring, you know, grandmothers, they bring children’s pictures into grandmothers lives.
00:43:14 It’s fantastic.
00:43:17 But they need to think of a new business model and that’s the core problem there.
00:43:22 Until they start to connect producer consumer, I think they will just continue to make money
00:43:26 and then buy the next social network company and then buy the next one and the innovation
00:43:30 level will not be high and the health issues will not go away.
00:43:34 So I apologize that we kind of returned to words, I don’t think the exact terms matter,
00:43:41 but in sort of defense of advertisement, don’t you think the kind of direct connection between
00:43:49 consumer and creator producer is what advertisement strives to do, right?
00:44:00 So that is best advertisement is literally now Facebook is listening to our conversation
00:44:06 and heard that you’re going to India and will be able to actually start automatically for
00:44:11 you making these connections and start giving this offer.
00:44:14 So like, I apologize if it’s just a matter of terms, but just to draw a distinction,
00:44:19 is it possible to make advertisements just better and better and better algorithmically
00:44:23 to where it actually becomes a connection, almost a direct connection?
00:44:26 That’s a good question.
00:44:27 So let’s component on that.
00:44:28 First of all, what we just talked about, I was defending advertising.
00:44:32 Okay.
00:44:33 So I was defending it as a way to get signals into a market that don’t come any other way,
00:44:36 especially algorithmically.
00:44:37 It’s a sign that someone spent money on it, it’s a sign they think it’s valuable.
00:44:41 And if I think that if other things, someone else thinks it’s valuable, and if I trust
00:44:45 other people, I might be willing to listen.
00:44:47 I don’t trust that Facebook though, who’s an intermediary between this.
00:44:51 I don’t think they care about me.
00:44:54 Okay.
00:44:55 I don’t think they do.
00:44:56 And I find it creepy that they know I’m going to India next week because of our conversation.
00:45:00 Why do you think that is?
00:45:02 So what, could you just put your PR hat on?
00:45:07 Why do you think you find Facebook creepy and not trust them as do majority of the population?
00:45:14 So they’re out of the Silicon Valley companies, I saw like not approval rate, but there’s
00:45:19 ranking of how much people trust companies and Facebook is in the gutter.
00:45:23 In the gutter, including people inside of Facebook.
00:45:25 So what do you attribute that to?
00:45:28 Because when I…
00:45:29 Come on, you don’t find it creepy that right now we’re talking that I might walk out on
00:45:31 the street right now that some unknown person who I don’t know kind of comes up to me and
00:45:35 says, I hear you’re going to India.
00:45:37 I mean, that’s not even Facebook.
00:45:38 That’s just, I want transparency in human society.
00:45:42 I want to have, if you know something about me, there’s actually some reason you know
00:45:45 something about me.
00:45:47 That’s something that if I look at it later and audit it kind of, I approve.
00:45:51 You know something about me because you care in some way.
00:45:54 There’s a caring relationship even, or an economic one or something.
00:45:58 Not just that you’re someone who could exploit it in ways I don’t know about or care about
00:46:02 or I’m troubled by or whatever.
00:46:05 We’re in a world right now where that happens way too much and that Facebook knows things
00:46:09 about a lot of people and could exploit it and does exploit it at times.
00:46:14 I think most people do find that creepy.
00:46:16 It’s not for them.
00:46:17 It’s not that Facebook is not doing it because they care about them in a real sense.
00:46:23 And they shouldn’t.
00:46:24 They should not be a big brother caring about us.
00:46:26 That is not the role of a company like that.
00:46:28 Why not?
00:46:29 Wait, not the big brother part, but the caring, the trusting.
00:46:32 I mean, don’t those companies, just to link on it because a lot of companies have a lot
00:46:37 of information about us.
00:46:38 I would argue that there’s companies like Microsoft that has more information about
00:46:42 us than Facebook does and yet we trust Microsoft more.
00:46:46 Well, Microsoft is pivoting.
00:46:47 Microsoft, you know, under Satya Nadella has decided this is really important.
00:46:51 We don’t want to do creepy things.
00:46:53 Really want people to trust us to actually only use information in ways that they really
00:46:56 would approve of, that we don’t decide, right?
00:47:00 And I’m just kind of adding that the health of a market is that when I connect to someone
00:47:06 who produces a consumer, it’s not just a random producer or consumer, it’s people who see
00:47:10 each other.
00:47:11 They don’t like each other, but they sense that if they transact, some happiness will
00:47:14 go up on both sides.
00:47:15 If a company helps me to do that in moments that I choose of my choosing, then fine.
00:47:22 So, and also think about the difference between, you know, browsing versus buying, right?
00:47:28 There are moments in my life I just want to buy, you know, a gadget or something.
00:47:31 I need something for that moment.
00:47:33 I need some ammonia for my house or something because I got a problem with a spill.
00:47:37 I want to just go in.
00:47:38 I don’t want to be advertised at that moment.
00:47:40 I don’t want to be led down various, you know, that’s annoying.
00:47:43 I want to just go and have it be extremely easy to do what I want.
00:47:49 Other moments I might say, no, it’s like today I’m going to the shopping mall.
00:47:52 I want to walk around and see things and see people and be exposed to stuff.
00:47:55 So I want control over that though.
00:47:56 I don’t want the company’s algorithms to decide for me, right?
00:48:00 I think that’s the thing.
00:48:01 There’s a total loss of control if Facebook thinks they should take the control from us
00:48:04 of deciding when we want to have certain kinds of information, when we don’t, what information
00:48:08 that is, how much it relates to what they know about us that we didn’t really want them
00:48:11 to know about us.
00:48:13 I don’t want them to be helping me in that way.
00:48:15 I don’t want them to be helping them by they decide they have control over what I want
00:48:21 and when.
00:48:22 I totally agree.
00:48:23 Facebook, by the way, I have this optimistic thing where I think Facebook has the kind
00:48:28 of personal information about us that could create a beautiful thing.
00:48:32 So I’m really optimistic of what Facebook could do.
00:48:36 It’s not what it’s doing, but what it could do.
00:48:38 So I don’t see that.
00:48:39 I think that optimism is misplaced because there’s not a bit, you have to have a business
00:48:43 model behind these things.
00:48:44 Create a beautiful thing is really, let’s be, let’s be clear.
00:48:48 It’s about something that people would value.
00:48:51 And I don’t think they have that business model and I don’t think they will suddenly
00:48:55 discover it by what, you know, a long hot shower.
00:48:58 I disagree.
00:48:59 I disagree in terms of, you can discover a lot of amazing things in a shower.
00:49:04 So I didn’t say that.
00:49:05 I said, they won’t come, they won’t do it, but in the shower, I think a lot of other
00:49:10 people will discover it.
00:49:11 I think that this guy, so I should also, full disclosure, there’s a company called United
00:49:15 Masters, which I’m on their board and they’ve created this music market and I have a hundred
00:49:18 thousand artists now signed on and they’ve done things like gone to the NBA and the NBA,
00:49:23 the music you find behind NBA clips right now is their music, right?
00:49:26 That’s a company that had the right business model in mind from the get go, right?
00:49:31 Executed on that.
00:49:32 And from day one, there was value brought to, so here you have a kid who made some songs
00:49:37 who suddenly their songs are on the NBA website, right?
00:49:41 That’s real economic value to people.
00:49:43 And so, you know, so you and I differ on the optimism of being able to sort of change the
00:49:51 direction of the Titanic, right?
00:49:54 So I, yeah, I’m older than you, so I’ve seen some Titanic’s crash, got it.
00:50:01 But and just to elaborate, cause I totally agree with you and I just want to know how
00:50:05 difficult you think this problem is of, so for example, I want to read some news and
00:50:11 I would, there’s a lot of times in the day where something makes me either smile or think
00:50:16 in a way where I like consciously think this really gave me value.
00:50:20 Like I sometimes listen to the daily podcasts in the New York times, way better than the
00:50:26 New York times themselves, by the way, for people listening.
00:50:29 That’s like real journalism is happening for some reason in the podcast space.
00:50:32 It doesn’t make sense to me, but often I listen to it 20 minutes and I would be willing to
00:50:37 pay for that, like $5, $10 for that experience.
00:50:41 And how difficult, that’s kind of what you’re getting at is that little transaction.
00:50:48 How difficult is it to create a frictionless system like Uber has, for example, for other
00:50:52 things?
00:50:53 What’s your intuition there?
00:50:55 So I, first of all, I pay little bits of money to, you know, to send, there’s something
00:50:58 called courts that does financial things.
00:51:00 I like medium as a site, I don’t pay there, but I would.
00:51:04 You had a great post on medium.
00:51:06 I would have loved to pay you a dollar and not others.
00:51:10 I wouldn’t have wanted it per se because there should be also sites where that’s not actually
00:51:15 the goal.
00:51:16 The goal is to actually have a broadcast channel that I monetize in some other way if I chose
00:51:20 to.
00:51:21 I mean, I could now people know about it.
00:51:23 I could, I’m not doing it, but that’s fine with me.
00:51:26 Also the musicians who are making all this music, I don’t think the right model is that
00:51:29 you pay a little subscription fee to them, right?
00:51:32 Because people can copy the bits too easily and it’s just not that somewhere the value
00:51:35 is.
00:51:36 The value is that a connection was made between real human beings, then you can follow up
00:51:39 on that.
00:51:40 All right.
00:51:41 And create yet more value.
00:51:42 So no, I think there’s a lot of open questions here, hot open questions, but also, yeah,
00:51:47 I do want good recommendation systems that recommend cool stuff to me.
00:51:51 But it’s pretty hard, right?
00:51:52 I don’t like them to recommend stuff just based on my browsing history.
00:51:55 I don’t like the based on stuff they know about me, quote unquote.
00:51:59 What’s unknown about me is the most interesting.
00:52:00 So this is the, this is the really interesting question.
00:52:03 We may disagree, maybe not.
00:52:05 I think that I love recommender systems and I want to give them everything about me in
00:52:12 a way that I trust.
00:52:13 Yeah.
00:52:14 But you, but you don’t, because, so for example, this morning I clicked on a, you know, I was
00:52:17 pretty sleepy this morning.
00:52:19 I clicked on a story about the queen of England.
00:52:23 Yes.
00:52:24 Right.
00:52:25 I do not give a damn about the queen of England.
00:52:26 I really do not.
00:52:27 But it was clickbait.
00:52:28 It kind of looked funny and I had to say, what the heck are they talking about?
00:52:31 I don’t want to have my life, you know, heading that direction.
00:52:34 Now that’s in my browsing history.
00:52:36 The system in any reasonable system will think that I care about the queen of England.
00:52:39 That’s browsing history.
00:52:40 Right.
00:52:41 But, but you’re saying all the trace, all the digital exhaust or whatever, that’s been
00:52:44 kind of the models.
00:52:45 If you collect all this stuff, you’re going to figure all of us out.
00:52:48 Well, if you’re trying to figure out like kind of one person like Trump or something,
00:52:51 maybe you could figure him out.
00:52:52 But if you’re trying to figure out, you know, 500 million people, you know, no way, no way.
00:52:58 You think so?
00:52:59 No, I do.
00:53:00 I think so.
00:53:01 I think we are, humans are just amazingly rich and complicated.
00:53:02 Every one of us has our little quirks, every one of us has our little things that could
00:53:05 intrigue us that we don’t even know it will intrigue us.
00:53:08 And there’s no sign of it in our past, but by God, there it comes and you know, you fall
00:53:12 in love with it.
00:53:13 And I don’t want a company trying to figure that out for me and anticipate that I want
00:53:16 them to provide a forum, a market, a place that I kind of go and by hook or by crook,
00:53:22 this happens, you know, I I’m walking down the street and I hear some Chilean music being
00:53:26 played and I never knew I liked Chilean music, but wow.
00:53:28 So there is that side and I want them to provide a limited, but you know, interesting place
00:53:33 to go.
00:53:34 Right.
00:53:35 And so don’t try to use your AI to kind of, you know, figure me out and then put me in
00:53:39 a world where you figured me out, you know, no, create huge spaces for human beings where
00:53:45 our creativity and our style will be enriched and come forward and it’ll be a lot of more
00:53:50 transparency.
00:53:51 I won’t have people randomly, anonymously putting comments up and I’ll special based
00:53:55 on stuff they know about me, facts that, you know, we are so broken right now.
00:54:00 If you’re, you know, especially if you’re a celebrity, but you know, it’s about anybody
00:54:02 that anonymous people are hurting lots and lots of people right now.
00:54:06 That’s part of this thing that Silicon Valley is thinking that, you know, just collect all
00:54:10 this information and use it in a great way.
00:54:12 So no, I’m not, I’m not a pessimist, I’m very much an optimist by nature, but I think that’s
00:54:16 just been the wrong path for the whole technology to take.
00:54:19 Be more limited, create, let humans rise up.
00:54:24 Don’t try to replace them.
00:54:25 That’s the AI mantra.
00:54:26 Don’t try to anticipate them.
00:54:28 Don’t try to predict them because you’re, you’re, you’re not going to, you’re not going
00:54:32 to be able to do those things.
00:54:33 You’re going to make things worse.
00:54:34 Okay.
00:54:35 So right now, just give this a chance.
00:54:38 Right now, the recommender systems are the creepy people in the shadow watching your
00:54:43 every move.
00:54:45 So they’re looking at traces of you.
00:54:47 They’re not directly interacting with you, sort of the, your close friends and family,
00:54:53 the way they know you is by having conversation, by actually having interactions back and forth.
00:54:57 Do you think there’s a place for recommender systems sort of to step, cause you, you just
00:55:02 emphasize the value of human to human connection, but yeah, just give it a chance, AI human
00:55:06 connection.
00:55:07 Is there a role for an AI system to have conversations with you in terms of, to try to figure out
00:55:13 what kind of music you like, not by just watching what you listening to, but actually having
00:55:17 a conversation, natural language or otherwise.
00:55:19 Yeah, no, I’m, I’m, so I’m not against it.
00:55:21 I just wanted to push back against the, maybe you’re saying you have options for Facebook.
00:55:25 So there I think it’s misplaced, but, but I think that distributing, yeah, no, so good
00:55:31 for you.
00:55:33 Go for it.
00:55:34 That’s a hard spot to be in.
00:55:35 Yeah, no, good.
00:55:36 Human interaction, like on our daily, the context around me in my own home is something
00:55:39 that I don’t want some big company to know about at all, but I would be more than happy
00:55:42 to have technology help me with it.
00:55:44 Which kind of technology?
00:55:45 Well, you know, just, Alexa, Amazon, well, a good, Alexa’s done right.
00:55:49 And I think Alexa is a research platform right now more than anything else.
00:55:52 But Alexa done right, you know, could do things like I, I leave the water running in my garden
00:55:56 and I say, Hey, Alexa, the water’s running in my garden.
00:55:59 And even have Alexa figure out that that means when my wife comes home, that she should be
00:56:02 told about that.
00:56:03 That’s a little bit of a reasoning.
00:56:04 I would call that AI and by any kind of stretch, it’s a little bit of reasoning and it actually
00:56:08 kind of would make my life a little easier and better.
00:56:11 And you know, I don’t, I wouldn’t call this a wow moment, but I kind of think that overall
00:56:14 rises human happiness up to have that kind of thing.
00:56:18 But not when you’re lonely, Alexa, knowing loneliness.
00:56:20 No, no, I don’t want Alexa to be, feel intrusive.
00:56:25 And I don’t want just the designer of the system to kind of work all this out.
00:56:28 I really want to have a lot of control and I want transparency and control.
00:56:32 And if a company can stand up and give me that in the context of new technology, I think
00:56:36 they’re good.
00:56:37 First of all, be way more successful than our current generation.
00:56:39 And like I said, I was mentioning Microsoft, I really think they’re, they’re pivoting to
00:56:43 kind of be the trusted old uncle, but you know, I think that they get that this is a
00:56:47 way to go, that if you let people find technology, empowers them to have more control and have
00:56:51 and have control, not just over privacy, but over this rich set of interactions, that that
00:56:56 people are going to like that a lot more.
00:56:58 And that’s, that’s the right business model going forward.
00:57:00 What does control over privacy look like?
00:57:02 Do you think you should be able to just view all the data that?
00:57:04 No, it’s much more than that.
00:57:05 I mean, first of all, it should be an individual decision.
00:57:07 Some people don’t want privacy.
00:57:09 They want their whole life out there.
00:57:10 Other people’s want it.
00:57:13 Privacy is not a zero one.
00:57:16 It’s not a legal thing.
00:57:17 It’s not just about which data is available, which is not.
00:57:20 I like to recall to people that, you know, a couple hundred years ago, everyone, there
00:57:24 was not really big cities, everyone lived in on the countryside and villages and villages.
00:57:29 Everybody knew everything about you.
00:57:30 Very, you didn’t have any privacy.
00:57:32 Is that bad?
00:57:33 Are we better off now?
00:57:34 Well, you know, arguably no, because what did you get for that loss of certain kinds
00:57:39 of privacy?
00:57:40 Well, people help each other if they, because they know everything about you.
00:57:44 They know something’s bad’s happening, they will help you with that.
00:57:46 Right.
00:57:47 And now you live in a big city, no one knows about that.
00:57:48 You get no help.
00:57:50 So it kind of depends the answer.
00:57:52 I want certain people who I trust and there should be relationships.
00:57:56 I should kind of manage all those, but who knows what about me?
00:57:59 I should have some agency there.
00:58:00 It shouldn’t, I shouldn’t be a drift in a sea of technology where I have no agency.
00:58:04 I don’t want to go reading things and checking boxes.
00:58:08 So I don’t know how to do that.
00:58:09 And I’m not a privacy researcher per se.
00:58:11 I just, I recognize the vast complexity of this.
00:58:14 It’s not just technology.
00:58:15 It’s not just legal scholars meeting technologists.
00:58:18 There’s gotta be kind of a whole layers around it.
00:58:20 And so I, when I alluded to this emerging engineering field, this is a big part of it.
00:58:26 When electrical engineering came, I’m not one around at the time, but you just didn’t
00:58:31 plug electricity into walls and all kinds of work.
00:58:34 You don’t have to have like underwriters laboratory that reassured you that that plug’s not going
00:58:37 to burn up your house and that that machine will do this and that and everything.
00:58:41 There’ll be whole people who can install things.
00:58:44 There’ll be people who can watch the installers.
00:58:46 There’ll be a whole layers, you know, an onion of these kinds of things.
00:58:49 And for things as deep and interesting as privacy, which is as least as interesting
00:58:53 as electricity, that’s going to take decades to kind of work out, but it’s going to require
00:58:58 a lot of new structures that we don’t have right now.
00:59:00 So it’s kind of hard to talk about it.
00:59:02 And you’re saying there’s a lot of money to be made if you get it right.
00:59:04 So something you should look at.
00:59:05 A lot of money to be made in all these things that provide human services and people recognize
00:59:09 them as useful parts of their lives.
00:59:12 So yeah.
00:59:14 So yeah, the dialogue sometimes goes from the exuberant technologists to the no technology
00:59:19 is good, kind of.
00:59:20 And that’s, you know, in our public discourse, you know, and as far as you see too much of
00:59:24 this kind of thing and the sober discussions in the middle, which are the challenge he
00:59:28 wants to have or where we need to be having our conversations.
00:59:31 And you know, there’s just not actually, there’s not many forum fora for those.
00:59:36 You know, there’s, that’s, that’s kind of what I would look for.
00:59:39 Maybe I could go and I could read a comment section of something and it would actually
00:59:42 be this kind of dialogue going back and forth.
00:59:44 You don’t see much of this, right?
00:59:45 Which is why actually there’s a resurgence of podcasts out of all, because people are
00:59:49 really hungry for conversation, but there’s technology is not helping much.
00:59:55 So comment sections of anything, including YouTube is not hurting and not helping.
01:00:01 Yeah.
01:00:02 And you think technically speaking, it’s possible to help.
01:00:07 I don’t know the answers, but it’s a, it’s a, it’s a less anonymity, a little more locality,
01:00:13 you know, worlds that you kind of enter in and you trust the people there in those worlds
01:00:17 so that when you start having a discussion, you know, not only is that people are not
01:00:20 going to hurt you, but it’s not going to be a total waste of your time because there’s
01:00:23 a lot of wasting of time that, you know, a lot of us, I pulled out of Facebook early
01:00:26 on cause it was clearly going to waste a lot of my time even though there was some value.
01:00:31 And so, yeah, worlds that are somehow you enter in and you know what you’re getting
01:00:34 and it’s kind of appeals to you and you might, new things might happen, but you kind of have
01:00:38 some, some trust in that world.
01:00:40 And there’s some deep, interesting, complex psychological aspects around anonymity, how
01:00:46 that changes human behavior that’s quite dark.
01:00:49 Quite dark.
01:00:50 Yeah.
01:00:51 I think a lot of us are, especially those of us who really loved the advent of technology.
01:00:55 I love social networks when they came out.
01:00:56 I was just, I didn’t see any negatives there at all.
01:00:59 But then I started seeing comment sections.
01:01:01 I think it was maybe, you know, with the CNN or something.
01:01:04 And I started to go, wow, this, this darkness I just did not know about and, and our technology
01:01:10 is now amplifying it.
01:01:11 So sorry for the big philosophical question, but on that topic, do you think human beings,
01:01:15 cause you’ve also, out of all things, had a foot in psychology too, the, do you think
01:01:21 human beings are fundamentally good?
01:01:23 Like all of us have good intent that could be mind or is it depending on context and
01:01:32 environment, everybody could be evil.
01:01:34 So my answer is fundamentally good.
01:01:37 But fundamentally limited.
01:01:39 All of us have very, you know, blinkers on.
01:01:41 We don’t see the other person’s pain that easily.
01:01:43 We don’t see the other person’s point of view that easily.
01:01:46 We’re very much in our own head, in our own world.
01:01:49 And on my good days, I think the technology could open us up to, you know, more perspectives
01:01:53 and more less blinkered and more understanding, you know, a lot of wars in human history happened
01:01:58 because of just ignorance.
01:01:59 They didn’t, they, they thought the other person was doing this while their person wasn’t
01:02:02 doing this.
01:02:03 And we have a huge amounts of that.
01:02:05 But in my lifetime, I’ve not seen technology really help in that way yet.
01:02:09 And I do, I do, I do believe in that, but you know, no, I think fundamentally humans
01:02:13 are good.
01:02:14 The people suffer, people have grievances because you have grudges and those things
01:02:17 cause them to do things they probably wouldn’t want.
01:02:20 They regret it often.
01:02:22 So no, I, I think it’s a, you know, part of the progress of technology is to indeed allow
01:02:28 it to be a little easier to be the real good person you actually are.
01:02:31 Well, but do you think individual human life or society could be modeled as an optimization
01:02:39 problem?
01:02:40 Not the way I think typically, I mean, that’s, you’re talking about one of the most complex
01:02:45 phenomenon in the whole, you know, in all of which the individual human life or society
01:02:49 as a whole.
01:02:50 Both, both.
01:02:51 I mean, individual human life is amazingly complex.
01:02:54 And so you know, optimization is kind of just one branch of mathematics that talks about
01:02:58 certain kinds of things.
01:02:59 And it just feels way too limited for the complexity of such things.
01:03:04 What properties of optimization problems do you think, so do you think most interesting
01:03:09 problems that could be solved through optimization, what kind of properties does that surface
01:03:13 have non convexity, convexity, linearity, all those kinds of things, saddle points?
01:03:19 Well, so optimization is just one piece of mathematics.
01:03:22 You know, there’s like, you just, even in our era, we’re aware that say sampling is
01:03:27 coming up, examples of something coming up with a distribution.
01:03:31 What’s optimization?
01:03:32 What’s sampling?
01:03:33 Well, they, you can, if you’re a kind of a certain kind of mathematician, you can try
01:03:35 to blend them and make them seem to be sort of the same thing.
01:03:38 But optimization is roughly speaking, trying to find a point that, a single point that
01:03:44 is the optimum of a criterion function of some kind.
01:03:48 And sampling is trying to, from that same surface, treat that as a distribution or density
01:03:53 and find points that have high density.
01:03:56 So I want the entire distribution in a sampling paradigm and I want the, you know, the single
01:04:03 point, that’s the best point in the optimization paradigm.
01:04:07 Now if you were optimizing in the space of probability measures, the output of that could
01:04:11 be a whole probability distribution.
01:04:13 So you can start to make these things the same.
01:04:15 But in mathematics, if you go too high up that kind of abstraction hierarchy, you start
01:04:18 to lose the, you know, the ability to do the interesting theorems.
01:04:22 So you kind of don’t try that.
01:04:23 You don’t try to overly over abstract.
01:04:26 So as a small tangent, what kind of worldview do you find more appealing?
01:04:31 One that is deterministic or stochastic?
01:04:35 Well, that’s easy.
01:04:36 I mean, I’m a statistician.
01:04:38 You know, the world is highly stochastic.
01:04:40 I don’t know what’s going to happen in the next five minutes, right?
01:04:42 Because what you’re going to ask, what we’re going to do, what I’ll say.
01:04:44 Due to the uncertainty.
01:04:45 Due to the…
01:04:46 Massive uncertainty.
01:04:47 Yeah.
01:04:48 You know, massive uncertainty.
01:04:49 And so the best I can do is have come rough sense or probability distribution on things
01:04:53 and somehow use that in my reasoning about what to do now.
01:04:58 So how does the distributed at scale when you have multi agent systems look like?
01:05:07 So optimization can optimize sort of, it makes a lot more sense, sort of at least from my
01:05:13 from robotics perspective, for a single robot, for a single agent, trying to optimize some
01:05:18 objective function.
01:05:21 When you start to enter the real world, this game theoretic concept starts popping up.
01:05:27 That’s how do you see optimization in this?
01:05:30 Because you’ve talked about markets in a scale.
01:05:32 What does that look like?
01:05:33 Do you see it as optimization?
01:05:34 Do you see it as sampling?
01:05:36 Do you see like, how should you mark?
01:05:38 These all blend together.
01:05:39 And a system designer thinking about how to build an incentivized system will have a blend
01:05:44 of all these things.
01:05:45 So, you know, a particle in a potential well is optimizing a functional called a Lagrangian,
01:05:49 right?
01:05:50 The particle doesn’t know that.
01:05:51 There’s no algorithm running that does that.
01:05:54 It just happens.
01:05:55 And so it’s a description mathematically of something that helps us understand as analysts
01:05:59 what’s happening, right?
01:06:00 And so the same thing will happen when we talk about, you know, mixtures of humans and
01:06:03 computers and markets and so on and so forth, there’ll be certain principles that allow
01:06:07 us to understand what’s happening, whether or not the actual algorithms are being used
01:06:10 by any sense is not clear.
01:06:13 Now at some point, I may have set up a multi agent or market kind of system.
01:06:19 And I’m now thinking about an individual agent in that system.
01:06:22 And they’re asked to do some task and they’re incentivized in some way, they get certain
01:06:25 signals and they have some utility.
01:06:28 What they will do at that point is they just won’t know the answer, they may have to optimize
01:06:31 to find an answer.
01:06:32 Okay, so an artist could be embedded inside of an overall market.
01:06:36 You know, and game theory is very, very broad.
01:06:39 It is often studied very narrowly for certain kinds of problems.
01:06:44 But it’s roughly speaking, this is just the, I don’t know what you’re going to do.
01:06:47 So I kind of anticipate that a little bit, and you anticipate what I’m anticipating.
01:06:51 And we kind of go back and forth in our own minds.
01:06:53 We run kind of thought experiments.
01:06:55 You’ve talked about this interesting point in terms of game theory, you know, most optimization
01:07:00 problems really hate saddle points, maybe you can describe what saddle points are.
01:07:04 But I’ve heard you kind of mentioned that there’s a there’s a branch of optimization
01:07:09 that you could try to explicitly look for saddle points as a good thing.
01:07:14 Oh, not optimization.
01:07:15 That’s just game theory that that so there’s all kinds of different equilibria in game
01:07:19 theory.
01:07:20 And some of them are highly explanatory behavior.
01:07:23 They’re not attempting to be algorithmic.
01:07:24 They’re just trying to say, if you happen to be at this equilibrium, you would see certain
01:07:29 kind of behavior.
01:07:30 And we see that in real life.
01:07:31 That’s what an economist wants to do, especially behavioral economists in continuous differential
01:07:39 game theory, you’re in continuous spaces, a some of the simplest equilibria are saddle
01:07:44 points and Nash equilibrium as a saddle point.
01:07:46 It’s a special kind of saddle point.
01:07:48 So classically, in game theory, you were trying to find Nash equilibria and an algorithmic
01:07:53 game theory, you’re trying to find algorithms that would find them.
01:07:56 And so you’re trying to find saddle points.
01:07:57 I mean, so that’s literally what you’re trying to do.
01:08:00 But you know, any economist knows that Nash equilibria have their limitations.
01:08:04 They are definitely not that explanatory in many situations.
01:08:08 They’re not what you really want.
01:08:10 There’s other kind of equilibria.
01:08:12 And there’s names associated with these because they came from history with certain people
01:08:15 working on them, but there will be new ones emerging.
01:08:18 So you know, one example is a Stackelberg equilibrium.
01:08:21 So you know, Nash, you and I are both playing this game against each other or for each other,
01:08:25 maybe it’s cooperative, and we’re both going to think it through and then we’re going to
01:08:29 decide and we’re going to do our thing simultaneously.
01:08:32 You know, in a Stackelberg, no, I’m going to be the first mover.
01:08:34 I’m going to make a move.
01:08:35 You’re going to look at my move and then you’re going to make yours.
01:08:38 Now since I know you’re going to look at my move, I anticipate what you’re going to do.
01:08:42 And so I don’t do something stupid, but then I know that you are also anticipating me.
01:08:46 So we’re kind of going back and forth on why, but there is then a first mover thing.
01:08:51 And so those are different equilibria, right?
01:08:54 And so just mathematically, yeah, these things have certain topologies and certain shapes
01:08:59 that are like, what’s it, algorithmically or dynamically, how do you move towards them?
01:09:02 How do you move away from things?
01:09:05 You know, so some of these questions have answers, they’ve been studied, others do not.
01:09:09 And especially if it becomes stochastic, especially if there’s large numbers of decentralized
01:09:13 things, there’s just, you know, young people get in this field who kind of think it’s all
01:09:17 done because we have, you know, TensorFlow.
01:09:19 Well, no, these are all open problems and they’re really important and interesting.
01:09:23 And it’s about strategic settings.
01:09:25 How do I collect data?
01:09:26 Suppose I don’t know what you’re going to do because I don’t know you very well, right?
01:09:29 Well, I got to collect data about you.
01:09:31 So maybe I want to push you into a part of the space where I don’t know much about you
01:09:34 so I can get data.
01:09:35 Cause, and then later I’ll realize that you’ll never, you’ll never go there because of the
01:09:38 way the game is set up.
01:09:39 You know, that’s part of the overall, you know, data analysis context is that.
01:09:44 Even the game of poker is fascinating space, whenever there’s any uncertainty, a lack of
01:09:47 information, it’s a super exciting space.
01:09:52 Just to linger on optimization for a second.
01:09:55 So when we look at deep learning, it’s essentially minimization of a complicated loss function.
01:10:01 So is there something insightful or hopeful that you see in the kinds of function surface
01:10:07 that loss functions, the deep learning and in the real world is trying to optimize over?
01:10:13 Is there something interesting as it’s just the usual kind of problems of optimization?
01:10:20 I think from an optimization point of view, that surface, first of all, it’s pretty smooth.
01:10:25 And secondly, if there’s over, if it’s over parameterized, there’s kind of lots of paths
01:10:29 down to reasonable Optima.
01:10:31 And so kind of the getting downhill to the, to an optimum is viewed as not as hard as
01:10:35 you might’ve expected in high dimensions.
01:10:39 The fact that some Optima tend to be really good ones and others not so good.
01:10:43 And you tend to, it’s not, sometimes you find the good ones is sort of still needs explanation.
01:10:48 Yeah.
01:10:49 But, but the particular surface is coming from the particular generation of neural nets.
01:10:53 I kind of suspect those will, those will change in 10 years.
01:10:56 It will not be exactly those surfaces.
01:10:58 There’ll be some others that are an optimization theory will help contribute to why other surfaces
01:11:02 or why other algorithms.
01:11:05 Years of arithmetic operations with a little bit of nonlinearity, that’s not, that didn’t
01:11:09 come from neuroscience per se.
01:11:10 I mean, maybe in the minds of some of the people working on it, they were thinking about
01:11:13 brains, but they were arithmetic circuits in all kinds of fields, computer science control
01:11:19 theory and so on.
01:11:20 And that layers of these could transform things in certain ways.
01:11:23 And that if it’s smooth, maybe you could find parameter values is a sort of big discovery
01:11:32 that it’s working, it’s able to work at this scale.
01:11:35 But I don’t think that we’re stuck with that and we’re, we’re certainly not stuck with
01:11:39 that cause we’re understanding the brain.
01:11:42 So in terms of on the algorithm side sort of gradient descent, do you think we’re stuck
01:11:46 with gradient descent as a variance of it?
01:11:49 What variance do you find interesting or do you think there’ll be something else invented
01:11:53 that is able to walk all over these optimization spaces in more interesting ways?
01:11:59 So there’s a co design of the surface and the, or the architecture and the algorithm.
01:12:04 So if you just ask if we stay with the kind of architectures that we have now and not
01:12:08 just neural nets, but you know, phase retrieval architectures or matrix completion architectures
01:12:13 and so on.
01:12:15 You know, I think we’ve kind of come to a place where yeah, a stochastic gradient algorithms
01:12:19 are dominant and there are versions that are a little better than others.
01:12:25 They have more guarantees, they’re more robust and so on.
01:12:29 And there’s ongoing research to kind of figure out which is the best arm for which situation.
01:12:34 But I think that that’ll start to co evolve, that that’ll put pressure on the actual architecture.
01:12:37 And so we shouldn’t do it in this particular way, we should do it in a different way because
01:12:40 this other algorithm is now available if you do it in a different way.
01:12:45 So that I can’t really anticipate that co evolution process, but you know, gradients
01:12:51 are amazing mathematical objects.
01:12:54 They have a lot of people who start to study them more deeply mathematically are kind of
01:13:01 shocked about what they are and what they can do.
01:13:05 Think about it this way, suppose that I tell you if you move along the x axis, you go uphill
01:13:11 in some objective by three units, whereas if you move along the y axis, you go uphill
01:13:15 by seven units, right?
01:13:18 Now I’m going to only allow you to move a certain unit distance, right?
01:13:22 What are you going to do?
01:13:23 Well, most people will say that I’m going to go along the y axis, I’m getting the biggest
01:13:27 bang for my buck, you know, and my buck is only one unit, so I’m going to put all of
01:13:31 it in the y axis, right?
01:13:33 And why should I even take any of my strength, my step size and put any of it in the x axis
01:13:39 because I’m getting less bang for my buck.
01:13:41 That seems like a completely clear argument and it’s wrong because the gradient direction
01:13:47 is not to go along the y axis, it’s to take a little bit of the x axis.
01:13:51 And to understand that, you have to know some math and so even a trivial so called operator
01:13:59 like gradient is not trivial and so, you know, exploiting its properties is still very important.
01:14:04 Now we know that just pervading descent has got all kinds of problems, it gets stuck in
01:14:06 many ways and it had never, you know, good dimension dependence and so on.
01:14:10 So my own line of work recently has been about what kinds of stochasticity, how can we get
01:14:15 dimension dependence, how can we do the theory of that and we’ve come up pretty favorable
01:14:20 results with certain kinds of stochasticity.
01:14:22 We have sufficient conditions generally.
01:14:25 We know if you do this, we will give you a good guarantee.
01:14:28 We don’t have necessary conditions that it must be done a certain way in general.
01:14:32 So stochasticity, how much randomness to inject into the walking along the gradient?
01:14:38 And what kind of randomness?
01:14:40 Why is randomness good in this process?
01:14:42 Why is stochasticity good?
01:14:44 Yeah, so I can give you simple answers but in some sense again, it’s kind of amazing.
01:14:49 Stochasticity just, you know, particular features of a surface that could have hurt you if you
01:14:55 were doing one thing deterministically won’t hurt you because by chance, there’s very little
01:15:02 chance that you would get hurt.
01:15:04 So here stochasticity, it just kind of saves you from some of the particular features of
01:15:12 surfaces.
01:15:13 In fact, if you think about surfaces that are discontinuous in our first derivative,
01:15:19 like an absolute value function, you will go down and hit that point where there’s nondifferentiability.
01:15:25 And if you’re running a deterministic algorithm at that point, you can really do something
01:15:28 bad.
01:15:29 Whereas stochasticity just means it’s pretty unlikely that’s going to happen, that you’re
01:15:32 going to hit that point.
01:15:35 So it’s again, nontrivial to analyze but especially in higher dimensions, also stochasticity,
01:15:41 our intuition isn’t very good about it but it has properties that kind of are very appealing
01:15:45 in high dimensions for a lot of large number of reasons.
01:15:49 So it’s all part of the mathematics to kind of, that’s what’s fun to work in the field
01:15:52 is that you get to try to understand this mathematics.
01:15:57 But long story short, you know, partly empirically, it was discovered stochastic gradient is very
01:16:01 effective and theory kind of followed, I’d say, that but I don’t see that we’re getting
01:16:06 clearly out of that.
01:16:09 What’s the most beautiful, mysterious, a profound idea to you in optimization?
01:16:15 I don’t know the most.
01:16:17 But let me just say that Nesterov’s work on Nesterov acceleration to me is pretty surprising
01:16:23 and pretty deep.
01:16:26 Can you elaborate?
01:16:27 Well Nesterov acceleration is just that, suppose that we are going to use gradients
01:16:32 to move around in a space.
01:16:33 For the reasons I’ve alluded to, they’re nice directions to move.
01:16:37 And suppose that I tell you that you’re only allowed to use gradients, you’re not going
01:16:40 to be allowed to use this local person that can only sense kind of the change in the surface.
01:16:47 But I’m going to give you kind of a computer that’s able to store all your previous gradients.
01:16:50 And so you start to learn some something about the surface.
01:16:55 And I’m going to restrict you to maybe move in the direction of like a linear span of
01:16:58 all the gradients.
01:16:59 So you can’t kind of just move in some arbitrary direction, right?
01:17:02 So now we have a well defined mathematical complexity model.
01:17:05 There’s certain classes of algorithms that can do that and others that can’t.
01:17:09 And we can ask for certain kinds of surfaces, how fast can you get down to the optimum?
01:17:13 So there’s answers to these.
01:17:14 So for a smooth convex function, there’s an answer, which is one over the number of steps
01:17:21 squared.
01:17:22 You will be within a ball of that size after k steps.
01:17:29 Gradient descent in particular has a slower rate, it’s one over k.
01:17:35 So you could ask, is gradient descent actually, even though we know it’s a good algorithm,
01:17:38 is it the best algorithm?
01:17:39 And the answer is no.
01:17:41 Well, not clear yet, because one over k squared is a lower bound.
01:17:47 That’s probably the best you can do.
01:17:49 Gradient is one over k, but is there something better?
01:17:52 And so I think as a surprise to most, Nesterov discovered a new algorithm that has got two
01:17:59 pieces to it.
01:18:00 It’s two gradients and puts those together in a certain kind of obscure way.
01:18:06 And the thing doesn’t even move downhill all the time.
01:18:09 It sometimes goes back uphill.
01:18:10 And if you’re a physicist, that kind of makes some sense.
01:18:13 You’re building up some momentum and that is kind of the right intuition, but that intuition
01:18:17 is not enough to understand kind of how to do it and why it works.
01:18:22 But it does.
01:18:23 It achieves one over k squared and it has a mathematical structure and it’s still kind
01:18:27 of to this day, a lot of us are writing papers and trying to explore that and understand
01:18:31 it.
01:18:32 So there are lots of cool ideas and optimization, but just kind of using gradients, I think
01:18:36 is number one that goes back, you know, 150 years.
01:18:40 And then Nesterov, I think has made a major contribution with this idea.
01:18:43 So like you said, gradients themselves are in some sense, mysterious.
01:18:47 They’re not as trivial as…
01:18:50 Not as trivial.
01:18:52 Coordinate descent is more of a trivial one.
01:18:54 You just pick one of the coordinates.
01:18:55 That’s how we think.
01:18:56 That’s how our human mind thinks.
01:18:57 That’s how our human minds think.
01:18:58 And gradients are not that easy for our human mind to grapple with.
01:19:03 An absurd question, but what is statistics?
01:19:08 So here it’s a little bit, it’s somewhere between math and science and technology.
01:19:12 It’s somewhere in that convex hole.
01:19:13 So it’s a set of principles that allow you to make inferences that have got some reason
01:19:17 to be believed and also principles that allow you to make decisions where you can have some
01:19:22 reason to believe you’re not going to make errors.
01:19:25 So all of that requires some assumptions about what do you mean by an error?
01:19:27 What do you mean by the probabilities?
01:19:31 But after you start making some of those assumptions, you’re led to conclusions that, yes, I can
01:19:38 guarantee that if you do this in this way, your probability of making an error will be
01:19:42 small.
01:19:43 Your probability of continuing to not make errors over time will be small.
01:19:47 And the probability that you found something that’s real will be small, will be high.
01:19:52 So decision making is a big part of that.
01:19:54 Decision making is a big part.
01:19:55 Yeah.
01:19:56 So statistics, short history was that, it goes back as a formal discipline, 250 years
01:20:03 or so.
01:20:04 It was called inverse probability because around that era, probability was developed
01:20:09 sort of especially to explain gambling situations.
01:20:12 Of course, interesting.
01:20:15 So you would say, well, given the state of nature is this, there’s a certain roulette
01:20:18 board that has a certain mechanism and what kind of outcomes do I expect to see?
01:20:23 And especially if I do things long amounts of time, what outcomes will I see?
01:20:27 And the physicists started to pay attention to this.
01:20:30 And then people said, well, let’s turn the problem around.
01:20:33 What if I saw certain outcomes, could I infer what the underlying mechanism was?
01:20:37 That’s an inverse problem.
01:20:38 And in fact, for quite a while, statistics was called inverse probability.
01:20:41 That was the name of the field.
01:20:44 And I believe that it was Laplace who was working in Napoleon’s government who needed
01:20:50 to do a census of France, learn about the people there.
01:20:54 So he went and gathered data and he analyzed that data to determine policy and said, well,
01:21:01 let’s call this field that does this kind of thing statistics because the word state
01:21:06 is in there.
01:21:07 In French, that’s etat, but it’s the study of data for the state.
01:21:12 So anyway, that caught on and it’s been called statistics ever since.
01:21:18 But by the time it got formalized, it was sort of in the 30s.
01:21:23 And around that time, there was game theory and decision theory developed nearby.
01:21:28 People in that era didn’t think of themselves as either computer science or statistics or
01:21:31 control or econ.
01:21:32 They were all the above.
01:21:34 And so Von Neumann is developing game theory, but also thinking of that as decision theory.
01:21:39 Wald is an econometrician developing decision theory and then turning that into statistics.
01:21:45 And so it’s all about, here’s not just data and you analyze it, here’s a loss function.
01:21:50 Here’s what you care about.
01:21:51 Here’s the question you’re trying to ask.
01:21:53 Here is a probability model and here’s the risk you will face if you make certain decisions.
01:21:59 And to this day, in most advanced statistical curricula, you teach decision theory as the
01:22:04 starting point and then it branches out into the two branches of Bayesian and frequentist.
01:22:08 But that’s all about decisions.
01:22:11 In statistics, what is the most beautiful, mysterious, maybe surprising idea that you’ve
01:22:19 come across?
01:22:20 Yeah, good question.
01:22:21 I mean, there’s a bunch of surprising ones.
01:22:27 There’s something that’s way too technical for this thing, but something called James
01:22:30 Stein estimation, which is kind of surprising and really takes time to wrap your head around.
01:22:36 Can you try to maybe…
01:22:37 I think I don’t want to even want to try.
01:22:39 Let me just say a colleague at Steven Stigler at University of Chicago wrote a really beautiful
01:22:44 paper on James Stein estimation, which helps to…
01:22:47 It’s views a paradox.
01:22:48 It kind of defeats the mind’s attempts to understand it, but you can and Steve has a
01:22:52 nice perspective on that.
01:22:56 So one of the troubles with statistics is that it’s like in physics that are in quantum
01:23:00 physics, you have multiple interpretations.
01:23:02 There’s a wave and particle duality in physics and you get used to that over time, but it
01:23:07 still kind of haunts you that you don’t really quite understand the relationship.
01:23:11 The electron’s a wave and electron’s a particle.
01:23:15 Well the same thing happens here.
01:23:16 There’s Bayesian ways of thinking and frequentist, and they are different.
01:23:21 They sometimes become sort of the same in practice, but they are physically different.
01:23:25 And then in some practice, they are not the same at all.
01:23:27 They give you rather different answers.
01:23:30 And so it is very much like wave and particle duality, and that is something that you have
01:23:33 to kind of get used to in the field.
01:23:35 Can you define Bayesian and frequentist?
01:23:37 Yeah in decision theory you can make, I have a video that people could see.
01:23:41 It’s called are you a Bayesian or a frequentist and kind of help try to make it really clear.
01:23:46 It comes from decision theory.
01:23:47 So you know, decision theory, you’re talking about loss functions, which are a function
01:23:51 of data X and parameter theta.
01:23:54 They’re a function of two arguments.
01:23:57 Okay.
01:23:58 Neither one of those arguments is known.
01:23:59 You don’t know the data a priori.
01:24:01 It’s random and the parameters unknown.
01:24:03 All right.
01:24:04 So you have a function of two things you don’t know, and you’re trying to say, I want that
01:24:07 function to be small.
01:24:08 I want small loss, right?
01:24:10 Well what are you going to do?
01:24:13 So you sort of say, well, I’m going to average over these quantities or maximize over them
01:24:17 or something so that, you know, I turn that uncertainty into something certain.
01:24:23 So you could look at the first argument and average over it, or you could look at the
01:24:25 second argument and average over it.
01:24:27 That’s Bayesian and frequentist.
01:24:28 So the frequentist says, I’m going to look at the X, the data, and I’m going to take
01:24:32 that as random and I’m going to average over the distribution.
01:24:35 So I take the expectation loss under X. Theta is held fixed, right?
01:24:40 That’s called the risk.
01:24:42 And so it’s looking at other, all the data sets you could get, right?
01:24:46 And say, how well will a certain procedure do under all those data sets?
01:24:50 That’s called a frequentist guarantee, right?
01:24:52 So I think it is very appropriate when like you’re building a piece of software and you’re
01:24:56 shipping it out there and people are using it on all kinds of data sets.
01:24:59 You want to have a stamp, a guarantee on it that as people run it on many, many data sets
01:25:02 that you never even thought about that 95% of the time it will do the right thing.
01:25:07 Perfectly reasonable.
01:25:09 The Bayesian perspective says, well, no, I’m going to look at the other argument of the
01:25:13 loss function, the theta part, okay?
01:25:15 That’s unknown and I’m uncertain about it.
01:25:17 So I could have my own personal probability for what it is, you know, how many tall people
01:25:21 are there out there?
01:25:22 I’m trying to infer the average height of the population while I have an idea roughly
01:25:25 what the height is.
01:25:27 So I’m going to average over the theta.
01:25:32 So now that loss function as only now, again, one argument’s gone, now it’s a function of
01:25:37 X and that’s what a Bayesian does is they say, well, let’s just focus on the particular
01:25:41 X we got, the data set we got, we condition on that.
01:25:45 Conditional on the X, I say something about my loss.
01:25:48 That’s a Bayesian approach to things.
01:25:50 And the Bayesian will argue that it’s not relevant to look at all the other data sets
01:25:54 you could have gotten and average over them, the frequentist approach.
01:25:58 It’s really only the data sets you got, right?
01:26:02 And I do agree with that, especially in situations where you’re working with a scientist, you
01:26:06 can learn a lot about the domain and you’re really only focused on certain kinds of data
01:26:09 and you gathered your data and you make inferences.
01:26:13 I don’t agree with it though, that, you know, in the sense that there are needs for frequentist
01:26:17 guarantees, you’re writing software, people are using it out there, you want to say something.
01:26:20 So these two things have to got to fight each other a little bit, but they have to blend.
01:26:24 So long story short, there’s a set of ideas that are right in the middle that are called
01:26:27 empirical Bayes.
01:26:29 And empirical Bayes sort of starts with the Bayesian framework.
01:26:34 It’s kind of arguably philosophically more, you know, reasonable and kosher.
01:26:40 Write down a bunch of the math that kind of flows from that, and then realize there’s
01:26:44 a bunch of things you don’t know because it’s the real world and you don’t know everything.
01:26:48 So you’re uncertain about certain quantities.
01:26:50 At that point, ask, is there a reasonable way to plug in an estimate for those things?
01:26:54 Okay.
01:26:55 And in some cases, there’s quite a reasonable thing to do, to plug in, there’s a natural
01:27:00 thing you can observe in the world that you can plug in and then do a little bit more
01:27:04 mathematics and assure yourself it’s really good.
01:27:06 So based on math or based on human expertise, what’s, what, what are good?
01:27:09 Oh, they’re both going in.
01:27:10 The Bayesian framework allows you to put a lot of human expertise in, but the math kind
01:27:16 of guides you along that path and then kind of reassures you the end, you could put that
01:27:19 stamp of approval under certain assumptions, this thing will work.
01:27:22 So you asked the question, what’s my favorite, you know, or what’s the most surprising, nice
01:27:25 idea.
01:27:26 So one that is more accessible is something called false discovery rate, which is, you
01:27:31 know, you’re making not just one hypothesis test or making one decision, you’re making
01:27:35 a whole bag of them.
01:27:37 And in that bag of decisions, you look at the ones where you made a discovery, you announced
01:27:41 that something interesting had happened.
01:27:43 All right.
01:27:44 That’s going to be some subset of your big bag.
01:27:47 In the ones you made a discovery, which subset of those are bad?
01:27:50 Or false, false discoveries.
01:27:53 You’d like the fraction of your false discoveries among your discoveries to be small.
01:27:57 That’s a different criterion than accuracy or precision or recall or sensitivity and
01:28:02 specificity.
01:28:03 It’s a different quantity.
01:28:04 Those latter ones are almost all of them have more of a frequentist flavor.
01:28:09 They say, given the truth is that the null hypothesis is true.
01:28:13 Here’s what accuracy I would get, or given that the alternative is true, here’s what
01:28:17 I would get.
01:28:18 So it’s kind of going forward from the state of nature to the data.
01:28:22 The Bayesian goes the other direction from the data back to the state of nature.
01:28:25 And that’s actually what false discovery rate is.
01:28:28 It says, given you made a discovery, okay, that’s conditioned on your data.
01:28:32 What’s the probability of the hypothesis?
01:28:34 It’s going the other direction.
01:28:36 And so the classical frequency look at that, well, I can’t know that there’s some priors
01:28:41 needed in that.
01:28:42 And the empirical Bayesian goes ahead and plows forward and starts writing down these formulas
01:28:47 and realizes at some point, some of those things can actually be estimated in a reasonable
01:28:51 way.
01:28:52 And so it’s kind of, it’s a beautiful set of ideas.
01:28:54 So I, this kind of line of argument has come out.
01:28:56 It’s not certainly mine, but it sort of came out from Robbins around 1960.
01:29:02 Brad Efron has written beautifully about this in various papers and books.
01:29:07 And the FDR is, you know, Benjamin in Israel, John Story did this Bayesian interpretation
01:29:14 and so on.
01:29:15 And he used to absorb these things over the years and find it a very healthy way to think
01:29:18 about statistics.
01:29:21 Let me ask you about intelligence to jump slightly back out into philosophy, perhaps.
01:29:28 You said that maybe you can elaborate, but you said that defining just even the question
01:29:33 of what is intelligence is a very difficult question.
01:29:38 Is it a useful question?
01:29:39 Do you think we’ll one day understand the fundamentals of human intelligence and what
01:29:45 it means, you know, have good benchmarks for general intelligence that we put before our
01:29:51 machines?
01:29:53 So I don’t work on these topics so much that you’re really asking the question for a psychologist
01:29:58 really.
01:29:59 And I studied some, but I don’t consider myself at least an expert at this point.
01:30:04 You know, a psychologist aims to understand human intelligence, right?
01:30:07 And I think many psychologists I know are fairly humble about this.
01:30:10 They might try to understand how a baby understands, you know, whether something’s a solid or liquid
01:30:15 or whether something’s hidden or not.
01:30:18 And maybe how a child starts to learn the meaning of certain words, what’s a verb, what’s
01:30:24 a noun and also, you know, slowly but surely trying to figure out things.
01:30:30 But humans ability to take a really complicated environment, reason about it, abstract about
01:30:35 it, find the right abstractions, communicate about it, interact and so on is just, you
01:30:41 know, really staggeringly rich and complicated.
01:30:46 And so, you know, I think in all humility, we don’t think we’re kind of aiming for that
01:30:51 in the near future.
01:30:52 A certain psychologist doing experiments with babies in the lab or with people talking has
01:30:56 a much more limited aspiration.
01:30:58 And you know, Kahneman and Tversky would look at our reasoning patterns and they’re not
01:31:02 deeply understanding all the how we do our reasoning, but they’re sort of saying, hey,
01:31:05 here’s some oddities about the reasoning and some things you should think about it.
01:31:09 But also, as I emphasize in some things I’ve been writing about, you know, AI, the revolution
01:31:14 hasn’t happened yet.
01:31:15 Yeah.
01:31:16 Great blog post.
01:31:17 I’ve been emphasizing that, you know, if you step back and look at intelligent systems
01:31:22 of any kind and whatever you mean by intelligence, it’s not just the humans or the animals or,
01:31:26 you know, the plants or whatever, you know, so a market that brings goods into a city,
01:31:31 you know, food to restaurants or something every day is a system.
01:31:35 It’s a decentralized set of decisions.
01:31:37 Looking at it from far enough away, it’s just like a collection of neurons.
01:31:40 Every neuron is making its own little decisions, presumably in some way.
01:31:44 And if you step back enough, every little part of an economic system is making all of
01:31:48 its decisions.
01:31:49 And just like with the brain, who knows what an individual neuron does and what the overall
01:31:53 goal is, right?
01:31:54 But something happens at some aggregate level, same thing with the economy.
01:31:58 People eat in a city and it’s robust.
01:32:01 It works at all scales, small villages to big cities.
01:32:04 It’s been working for thousands of years.
01:32:07 It works rain or shine, so it’s adaptive.
01:32:10 So all the kind of, you know, those are adjectives one tends to apply to intelligent systems.
01:32:14 Robust, adaptive, you know, you don’t need to keep adjusting it, self healing, whatever.
01:32:19 Plus not perfect.
01:32:20 You know, intelligences are never perfect and markets are not perfect.
01:32:24 But I do not believe in this era that you cannot, that you can say, well, our computers
01:32:28 are, our humans are smart, but you know, no markets are not, more markets are.
01:32:31 So they are intelligent.
01:32:34 Now we humans didn’t evolve to be markets.
01:32:38 We’ve been participating in them, right?
01:32:40 But we are not ourselves a market per se.
01:32:43 The neurons could be viewed as the market.
01:32:45 There’s economic, you know, neuroscience kind of perspective.
01:32:48 That’s interesting to pursue all that.
01:32:50 The point though is, is that if you were to study humans and really be the world’s best
01:32:54 psychologist studied for thousands of years and come up with the theory of human intelligence,
01:32:57 you might have never discovered principles of markets, you know, supply demand curves
01:33:01 and you know, matching and auctions and all that.
01:33:05 Those are real principles and they lead to a form of intelligence that’s not maybe human
01:33:08 intelligence.
01:33:09 It’s arguably another kind of intelligence.
01:33:11 There probably are third kinds of intelligence or fourth that none of us are really thinking
01:33:14 too much about right now.
01:33:16 So if you really, and then all of those are relevant to computer systems in the future.
01:33:20 Certainly the market one is relevant right now.
01:33:23 Whereas the understanding of human intelligence is not so clear that it’s relevant right now.
01:33:27 Probably not.
01:33:29 So if you want general intelligence, whatever one means by that, or, you know, understanding
01:33:33 intelligence in a deep sense and all that, it is definitely has to be not just human
01:33:37 intelligence.
01:33:38 It’s gotta be this broader thing.
01:33:39 And that’s not a mystery.
01:33:40 Markets are intelligent.
01:33:41 So, you know, it’s definitely not just a philosophical stance to say we’ve got to move beyond intelligence.
01:33:46 That sounds ridiculous.
01:33:47 Yeah.
01:33:48 But it’s not.
01:33:49 And in that blog post, you define different kinds of like intelligent infrastructure,
01:33:52 AI, which I really like is some of the concepts you’ve just been describing.
01:33:58 Do you see ourselves, if we see earth, human civilization as a single organism, do you
01:34:02 think the intelligence of that organism, when you think from the perspective of markets
01:34:06 and intelligence infrastructure is increasing, is it increasing linearly?
01:34:12 Is it increasing exponentially?
01:34:14 What do you think the future of that intelligence?
01:34:16 Yeah, I don’t know.
01:34:17 I don’t tend to think, I don’t tend to answer questions like that because you know, that’s
01:34:20 science fiction.
01:34:21 I’m hoping to catch you off guard.
01:34:25 Well again, because you said it’s so far in the future, it’s fun to ask and you’ll probably,
01:34:31 you know, like you said, predicting the future is really nearly impossible.
01:34:36 But say as an axiom, one day we create a human level, a superhuman level intelligent, not
01:34:43 the scale of markets, but the scale of an individual.
01:34:47 What do you think it is, what do you think it would take to do that?
01:34:51 Or maybe to ask another question is how would that system be different than the biological
01:34:58 human beings that we see around us today?
01:35:01 Is it possible to say anything interesting to that question or is it just a stupid question?
01:35:06 It’s not a stupid question, but it’s science fiction.
01:35:08 Science fiction.
01:35:09 And so I’m totally happy to read science fiction and think about it from time in my own life.
01:35:13 I loved, there was this like brain in a vat kind of, you know, little thing that people
01:35:17 were talking about when I was a student, I remember, you know, imagine that, you know,
01:35:22 between your brain and your body, there’s a, you know, there’s a bunch of wires, right?
01:35:26 And suppose that every one of them was replaced with a literal wire.
01:35:31 And then suppose that wire was turned in actually a little wireless, you know, there’s a receiver
01:35:35 and sender.
01:35:36 So the brain has got all the senders and receiver, you know, on all of its exiting, you know,
01:35:41 axons and all the dendrites down to the body have replaced with senders and receivers.
01:35:45 Now you could move the body off somewhere and put the brain in a vat, right?
01:35:50 And then you could do things like start killing off those senders and receivers one by one.
01:35:54 And after you’ve killed off all of them, where is that person?
01:35:56 You know, they thought they were out in the body walking around the world and they moved
01:35:59 on.
01:36:00 So those are science fiction things.
01:36:01 Those are fun to think about.
01:36:02 It’s just intriguing about where is, what is thought, where is it and all that.
01:36:05 And I think every 18 year old should take philosophy classes and think about these things.
01:36:10 And I think that everyone should think about what could happen in society that’s kind of
01:36:13 bad and all that.
01:36:14 But I really don’t think that’s the right thing for most of us that are my age group
01:36:17 to be doing and thinking about.
01:36:19 I really think that we have so many more present, you know, first challenges and dangers and
01:36:26 real things to build and all that such that, you know, spending too much time on science
01:36:32 fiction, at least in public for like this, I think is not what we should be doing.
01:36:36 Maybe over beers in private.
01:36:37 That’s right.
01:36:38 Well, I’m not going to broadcast where I have beers because this is going to go on Facebook
01:36:43 and I don’t want a lot of people showing up there.
01:36:45 But yeah, I’ll, I love Facebook, Twitter, Amazon, YouTube.
01:36:51 I have I’m optimistic and hopeful, but maybe, maybe I don’t have grounds for such optimism
01:36:58 and hope.
01:36:59 But let me ask, you’ve mentored some of the brightest sort of some of the seminal figures
01:37:07 in the field.
01:37:08 Can you give advice to people who are undergraduates today?
01:37:14 What does it take to take, you know, advice on their journey if they’re interested in
01:37:17 machine learning and in the ideas of markets from economics and psychology and all the
01:37:23 kinds of things that you’ve exploring?
01:37:25 What steps should they take on that journey?
01:37:27 Well, yeah, first of all, the door is open and second, it’s a journey.
01:37:30 I like your language there.
01:37:33 It is not that you’re so brilliant and you have great, brilliant ideas and therefore
01:37:37 that’s just, you know, that’s how you have success or that’s how you enter into the field.
01:37:42 It’s that you apprentice yourself, you spend a lot of time, you work on hard things, you
01:37:48 try and pull back and you be as broad as you can, you talk to lots of people.
01:37:53 And it’s like entering in any kind of a creative community.
01:37:57 There’s years that are needed and human connections are critical to it.
01:38:01 So, you know, I think about, you know, being a musician or being an artist or something,
01:38:06 you don’t just, you know, immediately from day one, you know, you’re a genius and therefore
01:38:10 you do it.
01:38:11 No, you, you know, practice really, really hard on basics and you be humble about where
01:38:18 you are and then, and you realize you’ll never be an expert on everything.
01:38:22 So you kind of pick and there’s a lot of randomness and a lot of kind of luck, but luck just kind
01:38:29 of picks out which branch of the tree you go down, but you’ll go down some branch.
01:38:33 So yeah, it’s a community.
01:38:35 So the graduate school is, I still think is one of the wonderful phenomena that we have
01:38:39 in our, in our world.
01:38:40 It’s very much about apprenticeship with an advisor.
01:38:43 It’s very much about a group of people you belong to.
01:38:45 It’s a four or five year process.
01:38:47 So it’s plenty of time to start from kind of nothing to come up to something, you know,
01:38:51 more, more expertise, and then to start to have your own creativity start to flower,
01:38:54 even surprising your own self.
01:38:58 And it’s a very cooperative endeavor.
01:38:59 I think a lot of people think of science as highly competitive and I think in some other
01:39:05 fields it might be more so.
01:39:08 Here it’s way more cooperative than you might imagine.
01:39:11 And people are always teaching each other something and people are always more than
01:39:14 happy to be clear that, so I feel I’m an expert on certain kinds of things, but I’m very much
01:39:20 not expert on lots of other things and a lot of them are relevant and a lot of them are,
01:39:23 I should know, but should in some society, you know, you don’t.
01:39:26 So I’m always willing to reveal my ignorance to people around me so they can teach me things.
01:39:32 And I think a lot of us feel that way about our field.
01:39:34 So it’s very cooperative.
01:39:35 I might add it’s also very international because it’s so cooperative.
01:39:39 We see no barriers.
01:39:40 And so that the nationalism that you see, especially in the current era and everything
01:39:44 is just at odds with the way that most of us think about what we’re doing here, where
01:39:48 this is a human endeavor and we cooperate and are very much trying to do it together
01:39:53 for the, you know, the benefit of everybody.
01:39:56 So last question, where and how and why did you learn French and which language is more
01:40:02 beautiful English or French?
01:40:05 Great question.
01:40:06 So first of all, I think Italian is actually more beautiful than French and English.
01:40:10 And I also speak that.
01:40:11 So I’m married to an Italian and I have kids and we speak Italian.
01:40:15 Anyway, all kidding aside, every language allows you to express things a bit differently.
01:40:23 And it is one of the great fun things to do in life is to explore those things.
01:40:26 So in fact, when I kids or teens or college students ask me what they study, I say, well,
01:40:34 do what your heart, where your heart is, certainly do a lot of math.
01:40:36 Math is good for everybody, but do some poetry and do some history and do some language too.
01:40:42 You know, throughout your life, you’ll want to be a thinking person.
01:40:44 You’ll want to have done that.
01:40:47 For me, French I learned when I was, I’d say a late teen, I was living in the middle of
01:40:54 the country in Kansas and not much was going on in Kansas with all due respect to Kansas.
01:41:01 And so my parents happened to have some French books on the shelf and just in my boredom,
01:41:04 I pulled them down and I found this is fun.
01:41:07 And I kind of learned the language by reading.
01:41:09 And when I first heard it spoken, I had no idea what was being spoken, but I realized
01:41:13 I had somehow knew it from some previous life and so I made the connection.
01:41:18 But then I traveled and just I love to go beyond my own barriers and my own comfort
01:41:23 or whatever.
01:41:24 And I found myself on trains in France next to say older people who had lived a whole
01:41:29 life of their own.
01:41:30 And the ability to communicate with them was special and the ability to also see myself
01:41:37 in other people’s shoes and have empathy and kind of work on that language as part of that.
01:41:43 So after that kind of experience and also embedding myself in French culture, which
01:41:49 is quite amazing, languages are rich, not just because there’s something inherently
01:41:53 beautiful about it, but it’s all the creativity that went into it.
01:41:55 So I learned a lot of songs, read poems, read books.
01:41:59 And then I was here actually at MIT where we’re doing the podcast today and a young
01:42:05 professor not yet married and not having a lot of friends in the area.
01:42:11 So I just didn’t have, I was kind of a bored person.
01:42:13 I said, I heard a lot of Italians around.
01:42:16 There’s happened to be a lot of Italians at MIT, an Italian professor for some reason.
01:42:20 And so I was kind of vaguely understanding what they were talking about.
01:42:22 I said, well, I should learn this language too.
01:42:23 So I did.
01:42:25 And then later met my spouse and Italian became a part of my life.
01:42:30 But I go to China a lot these days.
01:42:32 I go to Asia, I go to Europe and every time I go, I kind of am amazed by the richness
01:42:38 of human experience and the people don’t have any idea if you haven’t traveled, kind of
01:42:42 how amazingly rich and I love the diversity.
01:42:46 It’s not just a buzzword to me.
01:42:48 It really means something.
01:42:49 I love to embed myself with other people’s experiences.
01:42:53 And so yeah, learning language is a big part of that.
01:42:56 I think I’ve said in some interview at some point that if I had millions of dollars and
01:43:00 infinite time or whatever, what would you really work on if you really wanted to do
01:43:03 AI?
01:43:04 And for me, that is natural language and really done right.
01:43:07 Deep understanding of language.
01:43:09 That’s to me, an amazingly interesting scientific challenge.
01:43:13 One we’re very far away on.
01:43:15 One we’re very far away, but good natural language.
01:43:17 People are kind of really invested then.
01:43:19 I think a lot of them see that’s where the core of AI is that if you understand that
01:43:22 you really help human communication, you understand something about the human mind, the semantics
01:43:26 that come out of the human mind and I agree, I think that will be such a long time.
01:43:30 So I didn’t do that in my career just cause I kind of, I was behind in the early days.
01:43:34 I didn’t kind of know enough of that stuff.
01:43:36 I was at MIT, I didn’t learn much language and it was too late at some point to kind
01:43:41 of spend a whole career doing that, but I admire that field and so in my little way
01:43:47 by learning language, you know, kind of that part of my brain has been trained up.
01:43:53 Jan was right.
01:43:55 You truly are the Miles Davis of machine learning.
01:43:57 I don’t think there’s a better place than it.
01:43:59 Mike it was a huge honor talking to you today.
01:44:01 Merci beaucoup.
01:44:02 All right.
01:44:03 It’s been my pleasure.
01:44:04 Thanks for listening to this conversation with Michael I. Jordan and thank you to our
01:44:09 presenting sponsor, Cash App.
01:44:11 Download it, use code LEXPodcast, you’ll get $10 and $10 will go to FIRST, an organization
01:44:18 that inspires and educates young minds to become science and technology innovators of
01:44:22 tomorrow.
01:44:23 If you enjoy this podcast, subscribe on YouTube, give it five stars on Apple Podcast, support
01:44:28 on Patreon, or simply connect with me on Twitter at Lex Friedman.
01:44:34 And now let me leave you with some words of wisdom from Michael I. Jordan from his blog
01:44:39 post titled Artificial Intelligence, the revolution hasn’t happened yet, calling for broadening
01:44:45 the scope of the AI field.
01:44:48 We should embrace the fact that what we are witnessing is the creation of a new branch
01:44:52 of engineering.
01:44:54 The term engineering is often invoked in a narrow sense in academia and beyond with overtones
01:45:00 of cold, effectless machinery and negative connotations of loss of control by humans.
01:45:07 But an engineering discipline can be what we want it to be.
01:45:11 In the current era, we have a real opportunity to conceive of something historically new,
01:45:16 a human centric engineering discipline.
01:45:19 I will resist giving this emerging discipline a name, but if the acronym AI continues to
01:45:24 be used, let’s be aware of the very real limitations of this placeholder.
01:45:29 Let’s broaden our scope, tone down the hype, and recognize the serious challenges ahead.
01:45:37 Thank you for listening and hope to see you next time.