back to indexMichael I. Jordan: Machine Learning, Recommender Systems, and Future of AI | Lex Fridman Podcast #74
link |
The following is a conversation with Michael I. Jordan, a professor at Berkeley and one
link |
of the most influential people in the history of machine learning, statistics, and artificial
link |
He has been cited over 170,000 times and has mentored many of the world class researchers
link |
defining the field of AI today, including Andrew Eng, Zubin Garamani, Bantaskar, and
link |
All this, to me, is as impressive as the over 32,000 points in the six NBA championships
link |
of the Michael J. Jordan of basketball fame.
link |
There's a nonzero probability that I'd talk to the other Michael Jordan given my connection
link |
to and love of the Chicago Bulls of the 90s, but if I had to pick one, I'm going with the
link |
Michael Jordan of statistics and computer science, or as Jan Lacoon calls him, the Miles
link |
Davis of machine learning.
link |
In his blog post titled Artificial Intelligence, The Revolution Hasn't Happened Yet, Michael
link |
argues for broadening the scope or the artificial intelligence field.
link |
In many ways, the underlying spirit of this podcast is the same, to see artificial intelligence
link |
as a deeply human endeavor, to not only engineer algorithms and robots, but to understand and
link |
empower human beings at all levels of abstraction, from the individual to our civilization as
link |
This is the Artificial Intelligence Podcast.
link |
If you enjoy it, subscribe on YouTube, give us five stars at Apple Podcast, support it
link |
on Patreon, or simply connect with me on Twitter, at Lex Friedman, spelled F R I D M A N.
link |
As usual, I'll do one or two minutes of ads now and never any ads in the middle that can
link |
break the flow of the conversation.
link |
I hope that works for you and doesn't hurt the listening experience.
link |
This show is presented by Cash App, the number one finance app in the App Store.
link |
When you get it, use code LEX Podcast.
link |
Cash App lets you send money to friends, buy Bitcoin, and invest in the stock market with
link |
Since Cash App does fractional share trading, let me mention that the order execution algorithm
link |
that works behind the scenes to create the abstraction of the fractional orders is to
link |
me an algorithmic marvel.
link |
So big props for the Cash App engineers for solving a hard problem that in the end provides
link |
an easy interface that takes a step up to the next layer of abstraction over the stock
link |
market, making trading more accessible for new investors and diversification much easier.
link |
So once again, if you get Cash App from the App Store, Google Play, and use the code LEX
link |
Podcast, you'll get $10 and Cash App will also donate $10 to first, one of my favorite
link |
organizations that is helping to advance robotics and STEM education for young people around
link |
And now, here's my conversation with Michael I. Jordan.
link |
Given that you're one of the greats in the field of AI, machine learning, computer science,
link |
and so on, you're trivially called the Michael Jordan of machine learning, although as you
link |
know, you were born first, so technically MJ is the Michael I. Jordan of basketball,
link |
but anyway, my favorite is Yanlacoon calling you the Miles Davis of machine learning, because
link |
as he says, you reinvent yourself periodically and sometimes leave fans scratching their
link |
heads after you change direction.
link |
So can you put at first your historian hat on and give a history of computer science and
link |
AI as you saw it, as you experienced it, including the four generations of AI successes that
link |
I've seen you talk about?
link |
First of all, I much prefer Yan's metaphor.
link |
Miles Davis was a real explorer in jazz, and he had a coherent story.
link |
So I think I have one, but it's not just the one you lived.
link |
It's the one you think about later.
link |
What a good historian does is they look back and they revisit.
link |
I think what's happening right now is not AI.
link |
That was an intellectual aspiration that's still alive today as an aspiration.
link |
But I think this is akin to the development of chemical engineering from chemistry or
link |
electrical engineering from electromagnetism.
link |
So if you go back to the 30s or 40s, there wasn't yet chemical engineering.
link |
There was chemistry.
link |
There was fluid flow.
link |
There was mechanics and so on.
link |
But people pretty clearly viewed interesting goals to try to build factories that make
link |
chemicals products and do it viably, safely, make good ones, do it at scale.
link |
So people started to try to do that, of course, and some factories worked, some didn't.
link |
Some were not viable, some exploded.
link |
But in parallel, developed a whole field called chemical engineering.
link |
Chemical engineering is a field.
link |
It's no bones about it.
link |
It has theoretical aspects to it.
link |
It has practical aspects.
link |
It's not just engineering, quote, unquote.
link |
It's the real thing, real concepts are needed.
link |
Same thing with electrical engineering.
link |
There was Maxwell's equations, which in some sense were everything you know about electromagnetism.
link |
But you needed to figure out how to build circuits, how to build modules, how to put
link |
them together, how to bring electricity from one point to another safely and so on and
link |
So a whole field that developed called electrical engineering.
link |
I think that's what's happening right now is that we have a proto field, which is statistics,
link |
computer, more of the theoretical side of the algorithmic side of computer science.
link |
That was enough to start to build things.
link |
Systems that bring value to human beings and use human data and mix in human decisions.
link |
The engineering side of that is all ad hoc.
link |
That's what's emerging.
link |
In fact, if you want to call machine learning a field, I think that's what it is.
link |
That's a proto form of engineering based on statistical and computational ideas in previous
link |
But do you think there's something deeper about AI in his dreams and aspirations as
link |
compared to chemical engineering and electrical engineering?
link |
Well, the dreams and aspirations may be, but those are 500 years from now.
link |
I think that that's like the Greek sitting there and saying, it would be neat to get
link |
to the moon someday.
link |
I think we have no clue how the brain does computation.
link |
We're just a clue.
link |
We're even worse than the Greeks on most anything interesting scientifically of our era.
link |
Can you linger on that just for a moment because you stand not completely unique, but a little
link |
bit unique in the clarity of that.
link |
Can you elaborate your intuition of where we stand in our understanding of the human
link |
A lot of people say, and your scientists say, we're not very far in understanding human
link |
brain, but you're saying we're in the dark here.
link |
Well, I know I'm not unique.
link |
I don't even think in the clarity, but if you talk to real neuroscientists that really
link |
study real synapses or real neurons, they agree.
link |
It's 100 years, hundreds of year task, and they're building it up slowly, surely.
link |
What the signal is there is not clear.
link |
We have all of our metaphors.
link |
We think it's electrical, maybe it's chemical, it's a whole soup.
link |
It's ions and proteins, and it's a cell, and that's even around like a single synapse.
link |
If you look at an electron micrograph of a single synapse, it's a city of its own.
link |
That's one little thing on a dendritic tree, which is extremely complicated, electrochemical
link |
thing, and it's doing these spikes and voltages are flying around, and then proteins are taking
link |
that and taking it down into the DNA, and who knows what.
link |
It is the problem of the next few centuries.
link |
It is fantastic, but we have our metaphors about it.
link |
Is it an economic device?
link |
Is it like the immune system, or is it like a layered set of arithmetic computations?
link |
We have all these metaphors, and they're fun, but that's not real science per se.
link |
There is neuroscience.
link |
That's not neuroscience.
link |
That's like the Greek speculating about how to get to the moon, fun.
link |
I think that I like to say this fairly strongly, because I think a lot of young people think
link |
that we're on the verge, because a lot of people who don't talk about it clearly, let
link |
it be understood that, yes, we kind of, this is brain inspired, we're kind of close, breakthroughs
link |
are on the horizon, and unscrupulous people sometimes who need money for their labs.
link |
As I'm saying, unscrupulous, but people will oversell.
link |
I need money from a lab.
link |
I'm studying computational neuroscience.
link |
I'm going to oversell it, and so there's been too much of that.
link |
So I'll step into the gray area between metaphor and engineering with, I'm not sure if you're
link |
familiar with brain computer interfaces.
link |
So a company like Elon Musk has Neuralink that's working on putting electrodes into
link |
the brain and trying to be able to read both read and send electrical signals, just as
link |
you said, even the basic mechanism of communication in the brain is not something we understand.
link |
But do you hope, without understanding the fundamental principles of how the brain works,
link |
we'll be able to do something interesting at that gray area of metaphor?
link |
So I hope in the sense like anybody else hopes for some interesting things to happen from
link |
research, I would expect more something like Alzheimer's will get figured out from modern
link |
There's a lot of humans offering based on brain disease, and we throw things like lithium
link |
That's not quite true, but mostly we don't know.
link |
And that's even just about the biochemistry of the brain and how it leads to mood swings
link |
How thought emerges from that.
link |
We're really, really completely dim.
link |
So that you might want to hook up electrodes and try to do some signal processing on that
link |
and try to find patterns, fine, by all means go for it.
link |
It's just not scientific at this point.
link |
So it's like kind of sitting in a satellite and watching the emissions from a city and
link |
trying to affirm things about the microeconomy, even though you don't have microeconomic concepts.
link |
I mean, it's really that kind of thing.
link |
And so yes, can you find some signals that do something interesting or useful?
link |
Can you control a cursor or a mouse with your brain?
link |
And then I can imagine business models based on that and even medical applications of
link |
But from there to understanding algorithms that allow us to really tie in deeply from
link |
the brain to computer, I just, no, I don't agree with Elon Musk.
link |
I don't think that's even, that's not for our generation, it's not even for the century.
link |
So just in the hopes of getting you to dream, you've mentioned Komogorov and touring might
link |
Do you think that there might be breakthroughs that will get you to sit back in five, 10 years
link |
Oh, I'm sure there will be, but I don't think that there'll be demos that impress me.
link |
I don't think that having a computer call a restaurant and pretend to be a human is
link |
breakthrough and people, you know, some people presented as such.
link |
It's imitating human intelligence.
link |
It's even putting coughs in the thing to make a bit of a PR stunt.
link |
And so fine that the world runs on those things too.
link |
And I don't want to diminish all the hard work and engineering that goes behind things
link |
like that and the ultimate value to the human race.
link |
But that's not scientific understanding.
link |
And I know the people that work on these things, they are after scientific understanding, you
link |
know, in the meantime, they've got to kind of, you know, the trains got to run and they
link |
got mouths to feed and they got things to do.
link |
And there's nothing wrong with all that.
link |
I would call that though, just engineering.
link |
And I want to distinguish that between an engineering field like electrical engineering
link |
that originally that originally emerged that had real principles and you really know what
link |
you're doing and you had a little scientific understanding, maybe not even complete.
link |
So it became more predictable and it was really gave value to human life because it was understood.
link |
And so we don't want to muddle too much these waters of what we're able to do versus what
link |
we really can do in a way that's going to impress the next.
link |
So I don't need to be wowed, but I think that someone comes along in 20 years, a younger
link |
person who's absorbed all the technology and for them to be wowed, I think they have to
link |
be more deeply impressed.
link |
A young Kolmogorov would not be wowed by some of the stunts that you see right now coming
link |
from the big companies.
link |
The demos, but do you think the breakthroughs from Kolmogorov would be and give this question
link |
Do you think they'll be in the scientific fundamental principles arena?
link |
Or do you think it's possible to have fundamental breakthroughs in engineering?
link |
Meaning you know, I would say some of the things that Elon Musk is working with SpaceX
link |
and then others sort of trying to revolutionize the fundamentals of engineering of manufacturing
link |
of saying, here's a problem, we know how to do a demo of and actually taking it to scale.
link |
So there's going to be all kinds of breakthroughs.
link |
I just don't like that terminology.
link |
I'm a scientist and I work on things day in and day out and things move along and eventually
link |
say, wow, something happened, but I don't like that language very much.
link |
Also I don't like to prize theoretical breakthroughs over practical ones.
link |
I tend to be more of a theoretician and I think there's lots to do in that arena right now.
link |
And so I wouldn't point to the Kolmogorovs, I might point to the Edison's of the era
link |
and maybe Musk is a bit more like that.
link |
But you know, Musk, God bless him also, we'll say things about AI that he knows very little
link |
about and he doesn't know what he, he is, you know, it leads people astray when he talks
link |
about things he doesn't know anything about.
link |
Trying to program a computer to understand natural language, to be involved in a dialogue
link |
we're having right now, that can happen in our lifetime.
link |
You could fake it, you can mimic, sort of take old sentences that humans use and retread
link |
them with the deep understanding of language, no, it's not going to happen.
link |
And so from that, you know, I hope you can perceive that deeper, yet deeper kind of aspects
link |
and intelligence are not going to happen.
link |
Now will there be breakthroughs?
link |
You know, I think that Google was a breakthrough, I think Amazon is a breakthrough.
link |
You know, I think Uber is a breakthrough, you know, that bring value to human beings
link |
at scale in new brand new ways based on data flows and so on.
link |
A lot of these things are slightly broken because there's not a kind of a engineering
link |
field that takes economic value in context of data and, you know, planetary scale and
link |
worries about all the externalities, the privacy.
link |
You know, we don't have that field, so we don't think these things through very well.
link |
But I see that as emerging and that will be constant, that will, you know, looking back
link |
from a hundred years, that will be constantly a breakthrough in this era, just like electrical
link |
engineering was a breakthrough in the early part of the last century and chemical engineering
link |
was a breakthrough.
link |
So the scale, the markets that you talk about and we'll get to will be seen as sort of breakthrough
link |
and we're in very early days of really doing interesting stuff there.
link |
And we'll get to that, but it's just taking a quick step back.
link |
Can you give, kind of throw off the historian hat, I mean, you briefly said that in the
link |
history of AI kind of mimics the history of chemical engineering, but I keep saying machine
link |
learning, you keep wanting to say AI, just to let you know, I don't, you know, I resist
link |
I don't think this is about AI really was John McCarthy as almost a philosopher.
link |
Saying, wouldn't it be cool if we could put thought in a computer?
link |
If we could mimic the human capability to think or put intelligence in in some sense
link |
That's an interesting philosophical question.
link |
And he wanted to make it more than philosophy.
link |
He wanted to actually write down logical formula and algorithms that would do that.
link |
And that is a perfectly valid reasonable thing to do.
link |
That's not what's happening in this era.
link |
So the reason I keep saying AI actually, and I'd love to hear what you think about it,
link |
machine learning has a has a very particular set of methods and tools.
link |
Maybe your version of it is that mine doesn't know it doesn't very, very open.
link |
It does optimization.
link |
So systems that learn is what machine learning is systems that learn and make decisions and
link |
So what is pattern recognition and, you know, finding patterns is all about making decisions
link |
in real worlds and having close feedback loops.
link |
So something like symbolic AI expert systems, reasoning systems, knowledge based representation
link |
and all of those kinds of things search.
link |
Does that neighbor fit into what you think of as machine learning?
link |
So I don't even like the word machine learning.
link |
I think that with the field you're talking about is all about making large collections
link |
of decisions under uncertainty by large collections of entities.
link |
And there are principles for that at that scale.
link |
You don't have to say the principles are for a single entity that's making decisions,
link |
single agent or single human.
link |
It really immediately goes to the network of decisions.
link |
Is a good word for that?
link |
No, there's no good words for any of this.
link |
That's kind of part of the problem.
link |
So we can continue the conversation, use AI for all that.
link |
I just want to kind of raise our flag here that this is not about, we don't know what
link |
intelligence is and real intelligence.
link |
We don't know much about abstraction and reasoning at the level of humans.
link |
We don't have a clue.
link |
We're not trying to build that because we don't have a clue.
link |
Eventually it may emerge.
link |
They'll make, I don't know if they'll be breakthroughs, but eventually we'll start to get glimmers
link |
It's not what's happening right now.
link |
We're taking data.
link |
We're trying to make good decisions based on that.
link |
We're trying to do a scale.
link |
We're trying to do it economically, viably.
link |
We're trying to build markets.
link |
We're trying to keep value at that scale.
link |
And aspects of this will look intelligent.
link |
They will look, computers were so dumb before, they will see more intelligent.
link |
We will use that buzzword of intelligence.
link |
So we can use it in that sense, but you know, so machine learning, you can scope it narrowly
link |
is just learning from data and pattern recognition.
link |
But whatever I, when I talk about these topics, I maybe data science is another word you could
link |
It really is important that the decisions are, as part of it, it's consequential decisions
link |
in the real world.
link |
Am I going to have a medical operation?
link |
Am I going to drive down the street, you know, things that were, they're scarcity, things
link |
that impact other human beings or other, you know, the environment and so on.
link |
How do I do that based on data?
link |
I definitely how do I use computers to help those kinds of things go forward, whatever
link |
you want to call that.
link |
So let's call it AI.
link |
Let's agree to call it AI, but it's, let's, let's not say that what the goal of that
link |
is, is intelligence, the goal of that is really good working systems at planetary scale that
link |
we've never seen before.
link |
So reclaim the word AI from the Dartmouth conference from many decades ago of the dream
link |
I don't want to reclaim it.
link |
I want a new word.
link |
I think it was a bad choice.
link |
I mean, I, you know, I, if you read one of my little things, the history was basically
link |
that McCarthy needed a new name because cybernetics already existed.
link |
And he didn't like, you know, no one really liked Norbert Wiener.
link |
Norbert Wiener was kind of an island to himself.
link |
And he felt that he had encompassed all this and in some sense he did.
link |
If you look at the language of cybernetics, it was everything we're talking about.
link |
It was control theory and single processing and some notions of intelligence and close
link |
feedback loops and data.
link |
It's just not a word that lived on partly because of the, maybe the personalities.
link |
But McCarthy needed a new word to say, I'm different from you.
link |
I'm not part of your, your show.
link |
I got my own invented this word.
link |
And again, as a kind of a thinking forward about the movies that would be made about
link |
it, it was a great choice, but thinking forward about creating a sober academic and
link |
real world discipline.
link |
It was a terrible choice because it led to promises that are not true, that we understand.
link |
We understand artificial perhaps, but we don't understand intelligence.
link |
It's a small tangent because you're one of the great personalities of machine learning,
link |
whatever the heck you call the field.
link |
The, do you think science progresses by personalities or by the fundamental principles and theories
link |
and research that's outside of personality?
link |
And I wouldn't say there should be one kind of personality.
link |
I have mine and I have my preferences and I have a kind of network around me that feeds
link |
me and some of them agree with me and some disagree, but you know, all kinds of personalities
link |
Right now, I think the personality that it's a little too exuberant, a little bit too ready
link |
to promise the moon is, is a little bit too much in ascendance.
link |
And I do, I do think that that's, there's some good to that.
link |
It certainly attracts lots of young people to our field, but a lot of those people come
link |
in with strong misconceptions and they have to then unlearn those and then find something
link |
in, you know, to do.
link |
And so I think there's just got to be some, you know, multiple voices and there's, I didn't,
link |
I wasn't hearing enough of the more sober voice.
link |
So as a continuation of a fun tangent and speaking of vibrant personalities, what would
link |
you say is the most interesting disagreement you have with Jan Lacoon?
link |
So Jan's an old friend and I just say that I don't think we disagree about very much
link |
He and I both kind of have a let's build that kind of mentality and does it work and kind
link |
of mentality and kind of concrete.
link |
We both speak French and we speak French more together and we have, we have a lot, a lot
link |
And so, you know, if one wanted to highlight a disagreement, it's not really a fundamental.
link |
When I think it's just kind of where we're emphasizing, Jan has emphasized pattern recognition
link |
and has emphasized prediction.
link |
So, you know, and it's interesting to try to take that as far as you can.
link |
If you could do perfect prediction, what would that give you kind of as a thought experiment?
link |
And I think that's way too limited.
link |
We cannot do perfect prediction.
link |
We will never have the data sets that allow me to figure out what you're about ready to
link |
do, what question you're going to ask next.
link |
I will never know such things.
link |
Moreover, most of us find ourselves during the day in all kinds of situations we had
link |
no anticipation of that are kind of various, their novel in various ways.
link |
And in that moment, we want to think through what we want.
link |
And also, there's going to be market forces acting on us.
link |
I'd like to go down that street, but now it's full because there's a crane in the street.
link |
I got to think about that.
link |
I got to think about what I might really want here.
link |
And I got to sort of think about how much it costs me to do this action versus this
link |
I got to think about the risks involved.
link |
You know, a lot of our current pattern recognition and prediction systems don't do any risk evaluations.
link |
They have no error bars, right?
link |
I got to think about other people's decisions around me.
link |
I got to think about a collection of my decisions, even just thinking about like a medical treatment.
link |
You know, I'm not going to take the prediction of a neural net about my health, about something
link |
I'm not about ready to have a heart attack because some number is over.7.
link |
Even if you had all the data in the world, they've ever been collected about heart attacks
link |
better than any doctor ever had.
link |
I'm not going to trust the output of that neural net to predict my heart attack.
link |
I'm going to want to ask what if questions around that.
link |
I'm going to want to look at some other possible data I didn't have.
link |
I'm going to want to have a dialogue with a doctor about things we didn't think about
link |
when we gathered the data.
link |
You know, I could go on and on.
link |
I hope you can see.
link |
And I don't, I think that if you say predictions, everything that, that, that you're missing
link |
all of this stuff.
link |
And so prediction plus decision making is everything, but both of them are equally important.
link |
And so the field has emphasized prediction.
link |
Yon rightly so has seen how powerful that is.
link |
But at the cost of people not being aware of the decision making is where the rubber
link |
really hits the road, where human lives are at stake, where risks are being taken, where
link |
you got to gather more data.
link |
You got to think about the air bars.
link |
You got to think about the consequences of your decisions on others.
link |
You got about the economy around your decisions, blah, blah, blah, blah.
link |
I'm not the only one working on those, but we're a smaller tribe.
link |
And right now we're not the one that people talk about the most.
link |
But you know, if you go out in the real world and industry, you know, at Amazon, I'd say
link |
half the people there are working on decision making and the other half are doing, you know,
link |
the pattern recognition.
link |
And the words of pattern recognition and prediction, I think the distinction there, not to linger
link |
on words, but the distinction there is more a constrained sort of in the lab data set
link |
versus decision making is talking about consequential decisions in the real world under the messiness
link |
and the uncertainty of the real world.
link |
And just the whole of it, the whole mess of it that actually touches human beings and
link |
scale, like you said, market forces, that's the distinction.
link |
It helps add those, that perspective, that broader perspective.
link |
On the other hand, if you're a real prediction person, of course you want it to be in the
link |
You want to predict real world events.
link |
I'm just saying that's not possible with just data sets, that it has to be in the context
link |
of, you know, strategic things that someone's doing, data they might gather, things they
link |
could have gathered, the reasoning process around data.
link |
It's not just taking data and making predictions based on the data.
link |
So one of the things that you're working on, I'm sure there's others working on it, but
link |
I don't hear often it talked about, especially in the clarity that you talk about it.
link |
And I think it's both the most exciting and the most concerning area of AI in terms of
link |
So you've talked about AI systems that help make decisions that scale in a distributed
link |
way, millions, billions of decisions, sort of markets of decisions.
link |
Can you, as a starting point, sort of give an example of a system that you think about
link |
when you're thinking about these kinds of systems?
link |
So first of all, you're absolutely getting into some territory, which I will be beyond
link |
And there are lots of things that are going to be very not obvious to think about.
link |
Just like, again, I like to think about history a little bit, but think about, put yourself
link |
There was kind of a banking system that wasn't computerized really.
link |
There was database theory emerging.
link |
And database people had to think about, how do I actually not just move data around, but
link |
actual money, and have it be valid, and have transactions at ATMs happen that are actually
link |
all valid, and so on and so forth.
link |
So that's the kind of issues you get into when you start to get serious about things
link |
I like to think about as kind of almost a thought experiment to help me think something
link |
simpler, which is the music market, because to first order, there is no music market in
link |
the world right now in our country, for sure.
link |
There are things called record companies, and they make money, and they prop up a few
link |
really good musicians, and make them superstars, and they all make huge amounts of money.
link |
But there's a long tale of huge numbers of people that make lots and lots of really
link |
good music that is actually listened to by more people than the famous people.
link |
They are not in a market.
link |
They cannot have a career.
link |
They do not make money.
link |
The so called influencers or whatever.
link |
The managers who they are, right?
link |
So there are people who make extremely good music, especially in the hip hop or Latin
link |
They do it on their laptop.
link |
That's what they do on the weekend, and they have another job during the week, and they
link |
put it up on SoundCloud or other sites.
link |
Eventually, it gets streamed.
link |
It down gets turned into bits.
link |
It's not economically valuable.
link |
The information is lost.
link |
There are people stream it.
link |
You walk around in a big city.
link |
You see people with headphones all, you know, especially young kids listening to music all
link |
Especially the data, none of them, very little of the music they listen to is the famous
link |
And none of it's old music.
link |
It's all the latest stuff.
link |
But the people who made that latest stuff are like some 16 year old somewhere who will
link |
never make a career out of this, who will never make money.
link |
Of course, there will be a few counter examples.
link |
The record companies incentivize to pick out a few and highlight them.
link |
Long story short, there's a missing market there.
link |
There is not a consumer producer relationship at the level of the actual creative acts.
link |
The pipelines and spotifies of the world that take this stuff and stream it along, they
link |
make money off of subscriptions or advertising and those things.
link |
They're making the money, right?
link |
And then they will offer bits and pieces of it to a few people again to highlight that,
link |
you know, they're the simulator market.
link |
Anyway, a real market would be if you're a creator of music that you actually are somebody
link |
who's good enough that people want to listen to you.
link |
You should have the data available to you.
link |
There should be a dashboard showing a map of the United States.
link |
So in last week, here's all the places your songs were listened to.
link |
It should be transparent, vettable so that if someone down in Providence sees that you're
link |
being listened to 10,000 times in Providence, that they know that's real data.
link |
You know it's real data.
link |
They will have you come give a show down there.
link |
They will broadcast to the people who've been listening to you that you're coming.
link |
If you do this right, you could, you could, you know, go down there and make $20,000.
link |
You do that three times a year, you start to have a career.
link |
So in this sense, AI creates jobs.
link |
It's not about taking away human jobs.
link |
It's creating new jobs because it creates a new market.
link |
Once you've created a market, you've now connected up producers and consumers.
link |
You know, the person who's making the music can say to someone who comes to their shows
link |
a lot, hey, I'll play your daughter's wedding for $10,000.
link |
You'll say $8,000.
link |
They'll say $9,000.
link |
Then you, again, you can now get an income up to $100,000.
link |
You're not going to be a millionaire, all right.
link |
And now even think about really the value of music is in these personal connections, even
link |
so much so that a young kid wants to wear a T shirt with their favorite musician's signature
link |
So if they listen to the music on the internet, the internet should be able to provide them
link |
with a button that they push and the merchandise arrives the next day.
link |
We can do that, right?
link |
And now why should we do that?
link |
Well, because the kid who bought the shirt will be happy, but more the person who made
link |
the music will get the money.
link |
There's no advertising needed, right?
link |
So you can create markets between producers and consumers, take 5% cut.
link |
Your company will be perfectly sound, it'll go forward into the future, and it will create
link |
new markets and that raises human happiness.
link |
Now this seems like it was easy, just create this dashboard, kind of create some connections
link |
But if you think about Uber or whatever, you think about the challenges in the real world
link |
of it doing things like this, and there are actually new principles going to be needed.
link |
You're trying to create a new kind of two way market at a different scale that's ever
link |
There's going to be unwanted aspects of the market, there'll be bad people, there'll
link |
be the data will get used in the wrong ways, it'll fail in some ways, it won't deliver
link |
about, you have to think that through.
link |
Just like anyone who ran a big auction or ran a big matching service in economics will
link |
think these things through.
link |
And so that maybe didn't get at all the huge issues that can arise when you start to create
link |
markets, but it starts for me solidify my thoughts and allow me to move forward in my
link |
Yeah, so I talked to, had a research at Spotify actually, I think their long term goal they've
link |
said is to have at least one million creators make a comfortable living putting on Spotify.
link |
So I think you articulate a really nice vision of the world and the digital and the cyberspace
link |
What do you think companies like Spotify or YouTube or Netflix can do to create such
link |
Is it an AI problem?
link |
Is it an interface problems or interface design?
link |
Is it some other kind of, is it an economics problem?
link |
Who should they hire to solve these problems?
link |
Well, part of it's not just top down.
link |
So the Silicon Valley has this attitude that they know how to do it.
link |
They will create the system just like Google did with the search box that will be so good
link |
that they'll just everyone will adopt that.
link |
It's not, it's everything you said, but really I think missing the kind of culture.
link |
So it's literally that 16 year old who's, who's able to create the songs.
link |
You don't create that as a Silicon Valley entity.
link |
You don't hire them per se, right?
link |
You have to create an ecosystem in which they are wanted and that they're belong, right?
link |
And so you have to have some culture credibility to do things like this, you know, Netflix
link |
to their credit wanted some of that credibility and they created shows, you know, content.
link |
They call it content.
link |
It's such a terrible word, but it's culture, right?
link |
And so with movies, you can kind of go give a large sum of money to somebody graduate
link |
from the USC film school.
link |
It's a whole thing of its own, but it's kind of like rich white people's thing to do,
link |
you know, and you know, American culture has not been so much about rich white people.
link |
It's been about all the immigrants, all the, you know, the Africans who came and brought
link |
that culture and those, those rhythms and that to this world and created this whole
link |
new thing, you know, American culture.
link |
And so companies can't artificially create that.
link |
They can't just say, Hey, we're here.
link |
We're going to buy it up.
link |
You got a partner, right?
link |
And so, but anyway, you know, not to denigrate these companies are all trying and they should
link |
and they, they are, I'm sure they're asking these questions and some of them are even
link |
making an effort, but it is partly a respect the culture as you were a, as a technology
link |
person, you got to blend your technology with cultural, with cultural, you know, meaning.
link |
How much of a role do you think the algorithm machine learning has in connecting the consumer
link |
to the creator sort of the recommender system aspect of this?
link |
It's a great question.
link |
I think pretty high recommend, you know, there's no magic in the algorithms, but a good
link |
recommender system is way better than a bad recommender system and recommender systems
link |
is a billion dollar industry back even, you know, 10, 20 years ago.
link |
And it continues to be extremely important going forward.
link |
What's your favorite recommender system just so we can put something well, just historically
link |
I was one of the, you know, when I first went to Amazon, you know, I first didn't like
link |
Amazon because they put the book people are out of business or the library, you know,
link |
the local booksellers went out of business.
link |
I've come to accept that there, you know, there probably are more books being sold now
link |
and poor people reading them than ever before.
link |
And then local book stores are coming back.
link |
So you know, that's how economics sometimes work.
link |
You go up and you go down.
link |
But anyway, when I finally started going there and I bought a few books, I was really pleased
link |
to see another few books being recommended to me that I never would have thought of.
link |
And I bought a bunch of them, so they obviously had a good business model, but I learned things
link |
and I still to this day kind of browse using that service.
link |
And I think lots of people get a lot, you know, that is a good aspect of a recommendation
link |
I'm learning from my peers in an indirect way.
link |
And their algorithms are not meant to have them impose what we learn.
link |
It really is trying to find out what's in the data.
link |
It doesn't work so well for other kind of entities, but that's just the complexity of human life
link |
I'm not going to get recommendations on shirts, but that's interesting.
link |
If you try to recommend restaurants, it's hard.
link |
It's hard to do it at scale.
link |
But a blend of recommendation systems with other economic ideas, matchings and so on
link |
is really, really still very open, research wise, and there's new companies that are going
link |
to emerge that do that well.
link |
What do you think was going to the messy, difficult land of, say, politics and things
link |
like that that YouTube and Twitter have to deal with in terms of recommendation systems?
link |
Being able to suggest, I think Facebook just launched Facebook News, so having recommend
link |
the kind of news that are most likely for you to be interesting.
link |
Do you think this is AI solvable, again, whatever term you want to use, do you think it's a
link |
solvable problem for machines or is it a deeply human problem that's unsolvable?
link |
I don't even think about it at that level.
link |
I think that what's broken with some of these companies, it's all monetization by advertising.
link |
They're not at least Facebook.
link |
I want to critique them.
link |
They didn't really try to connect a producer and a consumer in an economic way.
link |
No one wants to pay for anything.
link |
They all started with Google and Facebook.
link |
They went back to the playbook of the television companies back in the day.
link |
Everyone wanted to pay for this signal, they will pay for the TV box, but not for the signal,
link |
at least back in the day.
link |
Advertising kind of filled that gap and advertising was new and interesting and it somehow didn't
link |
take over our lives quite.
link |
Fast forward, Google provides a service that people don't want to pay for.
link |
Somewhat surprisingly in the 90s, they ended up making huge amounts to the corner of the
link |
advertising market.
link |
It didn't seem like that was going to happen, at least to me.
link |
These little things on the right hand side of the screen just did not seem all that economically
link |
interesting, but companies had maybe no other choice.
link |
The TV market was going away and billboards and so on.
link |
I think that, sadly, that Google was doing so well with that and making it so much more.
link |
They didn't think much more about how, wait a minute, is there a producer or consumer
link |
relationship to be set up here, not just between us and the advertisers market to be created?
link |
Is there an actual market between the producer and consumer?
link |
There, the producer is the person who created that video clip.
link |
The person that made that website, the person who couldn't make more such things, the person
link |
who could adjust it as a function of demand, the person on the other side who's asking
link |
for different kinds of things.
link |
You see glimmers of that now, there's influencers and there's a little glimmering of a market,
link |
but it should have been done 20 years ago, it should have been thought about.
link |
It should have been created in parallel with the advertising ecosystem.
link |
Then Facebook inherited that and I think they also didn't think very much about that.
link |
Fast forward and now they are making huge amounts of money off of advertising and the
link |
news thing and all these clicks is just feeding the advertising.
link |
It's all connected up to the advertising.
link |
You want more people to click on certain things because that money flows to you, Facebook.
link |
You're very much incentivized to do that and when you start to find it's breaking, people
link |
are telling you, well, we're getting into some troubles, you try to adjust it with your
link |
smart AI algorithms and figure out what are bad clicks though, maybe shouldn't be clicked
link |
through the radar.
link |
I find that pretty much hopeless, it does get into all the complexity of human life
link |
and you can try to fix it, you should, but you could also fix the whole business model
link |
and the business model is that really, what are, are there some human producers and consumers
link |
Is there some economic value to be liberated by connecting them directly?
link |
Is it such that it's so valuable that people will be going to pay for it?
link |
Like micro payment, like small payment.
link |
Micro, but even after you micro, so I like the example, suppose I'm going, next week
link |
I'm going to India, never been to India before, right?
link |
I have a couple of days in, in Mumbai, I have no idea what to do there, right?
link |
And I could go on the web right now and search, it's going to be kind of hopeless.
link |
I'm not going to find, you know, I'll have lots of advertisers in my face, right?
link |
What I really want to do is broadcast to the world that I am going to Mumbai and have someone
link |
on the other side of a market look at me and, and there's a recommendation system there.
link |
So they're not looking at all possible people coming to Mumbai, they're looking at the people
link |
who are relevant to them.
link |
So someone in my age group, someone who kind of knows me in some level, I give up a little
link |
But I'm happy because what I'm going to get back is this person can make a little video
link |
for me or they're going to write a little two page paper on, here's the cool things
link |
that you want to do and move by this week, especially, right?
link |
I'm going to look at that.
link |
I'm not going to pay a micro payment.
link |
I'm going to pay, you know, a hundred dollars or whatever for that.
link |
It's like journalism.
link |
And as an odd subscription, it's that I'm going to pay that person in that moment.
link |
I mean, it's going to take 5% of that and that person has now got it.
link |
It's a gig economy, if you will, but, you know, done for it, you know, thinking about
link |
a little bit behind YouTube, there was actually people who could make more of those things.
link |
If they were connected to a market, they would make more of those things independently.
link |
You don't have to tell them what to do.
link |
You don't have to incentivize them in any other way.
link |
And so yeah, these companies, I don't think have thought long, long and heard about that.
link |
So I do distinguish on, you know, Facebook on the one side who just not thought about
link |
these things at all.
link |
They were thinking that AI will fix everything and Amazon thinks about them all the time
link |
because they were already out in the real world.
link |
They were delivering packages to people's doors.
link |
They were worried about a market.
link |
They were worried about sellers and, you know, they worry and some things they do are great.
link |
Some things maybe not so great, but, you know, they're in that business model.
link |
And then I'd say Google sort of hovers somewhere in between.
link |
I don't think for a long, long time they got it.
link |
I think they probably see that YouTube is more pregnant with possibility than they might
link |
have thought and that they're probably heading that direction.
link |
But you know, Silicon Valley has been dominated by the Google, Facebook kind of mentality
link |
and the subscription and advertising and that is, that's the core problem, right?
link |
The fake news actually rides on top of that because it means that you're monetizing with
link |
clip through rate and that is the core problem.
link |
You got to remove that.
link |
So advertisement, if we're going to linger on that, I mean, that's an interesting thesis.
link |
I don't know if everyone really deeply thinks about that.
link |
The thought is the advertisement model is the only thing we have, the only thing we'll
link |
So we have to fix, we have to build algorithms that despite that business model, you know,
link |
find the better angels of our nature and do good by society and by the individual.
link |
But you think we can slowly, you think, first of all, there's a difference between should
link |
So you're saying we should slowly move away from the advertising model and have a direct
link |
connection between the consumer and the creator.
link |
The question I also have is, can we, because the advertising model is so successful now
link |
in terms of just making a huge amount of money and therefore being able to build a big company
link |
that provides, has really smart people working that create a good service.
link |
Do you think it's possible?
link |
And just to clarify, you think we should move away?
link |
Well, I think we should.
link |
But we is the, you know, me.
link |
Well, the companies.
link |
I mean, so first of all, full disclosure.
link |
I'm doing a day a week at Amazon because I kind of want to learn more about how they
link |
So, you know, I'm not speaking for Amazon in any way, but, you know, I did go there
link |
because I actually believe they get a little bit of this or trying to create these markets.
link |
And they don't really use, advertising is not a crucial part of Amazon.
link |
That's a good question.
link |
So it has become not crucial, but it's become more and more present if you go to Amazon
link |
And, you know, without revealing too many deep secrets about Amazon, I can tell you
link |
that, you know, a lot of people come to question this and there's a huge questioning going
link |
You do not want a world where there's zero advertising.
link |
That actually is a bad world.
link |
So here's the way to think about it.
link |
You're a company that like Amazon is trying to bring products to customers, right?
link |
And the customer and you get more, you want to buy a vacuum cleaner, say, you want to
link |
know what's available for me.
link |
And, you know, it's not going to be that obvious.
link |
You have to do a little bit of work at it.
link |
The recommendation system will sort of help, right?
link |
But now suppose this other person over here has just made the world, you know, they spend
link |
a huge amount of energy.
link |
They had a great idea.
link |
They made a great vacuum cleaner.
link |
They know they really did it.
link |
They, you know, whiz kid that made a great new vacuum cleaner, right?
link |
It's not going to be in the recommendation system.
link |
No one will know about it.
link |
The algorithms will not find it and AI will not fix that.
link |
How do you allow that vacuum cleaner to start to get in front of people?
link |
Well, advertising and here what advertising is, it's a signal that you're, you believe in
link |
your product enough that you're willing to pay some real money for it.
link |
And to me as a consumer, I look at that signal, I say, well, first of all, I know these are
link |
not just cheap little ads because we have now right now that I know that, you know, these
link |
are super cheap, you know, pennies.
link |
If I see an ad where it's actually, I know the company is only doing a few of these and
link |
they're making real money is kind of flowing and I see an ad, I may pay more attention
link |
And I actually might want that because I see, Hey, that guy spent money on his vacuum cleaner.
link |
Maybe there's something good there.
link |
So I will look at it.
link |
And so that's part of the overall information flow in a good market.
link |
So advertising has a role.
link |
But the problem is, of course, that that signal is now completely gone because it just, you
link |
know, dominated by these tiny little things that add up to big money for the company.
link |
You know, so I think it will just, I think it will change because societies just don't,
link |
you know, stick with things that annoy a lot of people and advertising currently annoys
link |
people more than it provides information.
link |
And I think that a Google probably is smart enough to figure out that this is a dead, this
link |
is a bad model, even though it's a hard huge amount of money and they'll have to figure
link |
out how to pull it away from it slowly, and I'm sure the CEO there will figure it out.
link |
But they need to do it and they needed it.
link |
So if you reduce advertising, not to zero, but you reduce it at the same time you bring
link |
up producer, consumer, actual real value being delivered.
link |
So real money is being paid and they take a 5% cut.
link |
That 5% could start to get big enough to cancel out the lost revenue from the kind of the
link |
poor kind of advertising.
link |
And I think that a good company will do that, will realize that.
link |
And they're, you know, Facebook, you know, again, God bless them.
link |
They bring, you know, grandmothers, you know, they bring children's pictures into grandmothers
link |
But they need to think of a new business model and that's the core problem there.
link |
Until they start to connect producer, consumer, I think they will just continue to make money
link |
and then buy the next social network company and then buy the next one.
link |
And the innovation level will not be high and the health issues will not go away.
link |
So I apologize that we kind of returned to words.
link |
I don't think the exact terms matter, but in sort of defensive advertisement, don't
link |
you think the kind of direct connection between consumer and creator, producer is the best,
link |
like the, is what advertisement strives to do, right?
link |
So it is best advertisement is literally now the Facebook is listening to our conversation
link |
and heard that you're going to India and we'll be able to actually start automatically for
link |
you making these connections and start giving this offer.
link |
So like, I apologize if it's just a matter of terms, but just to draw a distinction,
link |
is it possible to make advertisements just better and better and better algorithmically
link |
to where it actually becomes a connection almost a direct connection?
link |
That's a good question.
link |
So let's put it on that first of all, what we just talked about, I was defending advertising.
link |
So I was defending it as a way to get signals into a market that don't come any other way,
link |
especially algorithmically.
link |
It's a sign that someone spent money on it is a sign they think it's valuable.
link |
And if I think that if other things, someone else thinks it's valuable, then if I trust
link |
other people, I might be willing to listen.
link |
I don't trust that Facebook though is who's an intermediary between this.
link |
I don't think they care about me.
link |
I don't think they do.
link |
And I find it creepy that they know I'm going to India next week because of our conversation.
link |
Why do you think that?
link |
Can we just put your PR hat on?
link |
Why do you think you find Facebook creepy and not trust them as do majority of the population?
link |
So they're out of the Silicon Valley companies.
link |
I saw like, not approval rate, but there's ranking of how much people trust companies
link |
and Facebook is in the gutter.
link |
In the gutter, including people inside of Facebook.
link |
So what do you attribute that to?
link |
You don't find it creepy that right now we're talking that I might walk out on the street
link |
right now that some unknown person who I don't know kind of comes up to me and says,
link |
I hear you going to India.
link |
I mean, that's not even Facebook.
link |
I want transparency in human society.
link |
If you know something about me, there's actually some reason you know something about me.
link |
It's something that if I look at it later and audit it kind of, I approve.
link |
You know something about me because you care in some way.
link |
There's a caring relationship even or an economic one or something.
link |
Not just that you're someone who could exploit it in ways I don't know about or care about
link |
or I'm troubled by or whatever.
link |
And we're in a world right now where that happened way too much and that Facebook knows
link |
things about a lot of people and could exploit it and does exploit it at times.
link |
I think most people do find that creepy.
link |
It's not for them.
link |
Facebook does not do it because they care about them in any real sense.
link |
And they shouldn't.
link |
They should not be a big brother caring about us.
link |
That is not the role of a company like that.
link |
Wait, not the big brother part but the caring, the trust thing.
link |
I mean, don't those companies...
link |
Just to linger on it because a lot of companies have a lot of information about us.
link |
I would argue that there's companies like Microsoft that has more information about us
link |
than Facebook does and yet we trust Microsoft more.
link |
But Microsoft is pivoting.
link |
Microsoft has decided this is really important.
link |
We don't want to do creepy things.
link |
Really want people to trust us to actually only use information in ways that they really
link |
would approve of, that we don't decide.
link |
And I'm just kind of adding that the health of a market is that when I connect to someone
link |
who produced or consumers, not just a random producer or consumer, it's people who see
link |
They don't like each other but they sense that if they transact, some happiness will
link |
go up on both sides.
link |
If a company helps me to do that and moments that I choose of my choosing, then fine.
link |
So and also think about the difference between browsing versus buying, right?
link |
There are moments in my life, I just want to buy a gadget or something.
link |
I need something for that moment.
link |
I need some ammonia for my house or something because I got a problem in this bill.
link |
I want to just go in.
link |
I don't want to be advertised at that moment.
link |
I don't want to be led down very straight.
link |
Let's just go and have it extremely easy to do what I want.
link |
Other moments I might say, no, it's like today I'm going to the shopping mall.
link |
I want to walk around and see things and see people and be exposed to stuff.
link |
So I want control over that though.
link |
I don't want the company's algorithms to decide for me, right?
link |
And I think that's the thing.
link |
It's a total loss of control if Facebook thinks they should take the control from us of deciding
link |
when we want to have certain kinds of information, when we don't, what information that is, how
link |
much it relates to what they know about us that we didn't really want them to know about
link |
I don't want them to be helping me in that way.
link |
I don't want them to be helping them by they decide they have control over what I want
link |
So Facebook, by the way, I have this optimistic thing where I think Facebook has the kind
link |
of personal information about us that could create a beautiful thing.
link |
So I'm really optimistic of what Facebook could do.
link |
It's not what it's doing, but what it could do.
link |
So I don't see that, I think that optimism is misplaced because you have to have a business
link |
model behind these things.
link |
Create a beautiful thing is really, let's be clear.
link |
It's about something that people would value.
link |
And I don't think they have that business model.
link |
And I don't think they will suddenly discover it by what, you know, a long hot shower.
link |
I disagree in terms of, you can discover a lot of amazing things in a shower.
link |
So I didn't say that.
link |
I said they won't come.
link |
But in the shower, I think a lot of other people will discover it.
link |
I think that this guy, so I should also full disclosure, there's a company called United
link |
Masters, which I'm on their board and they've created this music market and they have 100,000
link |
artists now signed on and they've done things like gone to the NBA and the NBA, the music
link |
you find behind NBA Eclipse right now is their music, right?
link |
That's a company that had the right business model in mind from the get go, right, executed
link |
And from day one, there was value brought to, so here you have a kid who made some songs
link |
who suddenly their songs are on the NBA website, right?
link |
That's really economic value to people.
link |
So you and I differ on the optimism of being able to sort of change the direction of the
link |
I'm older than you.
link |
So I think the Titanic's crash.
link |
But just to elaborate, because I totally agree with you and I just want to know how difficult
link |
you think this problem is of, so for example, I want to read some news and I would, there's
link |
a lot of times in the day where something makes me either smile or think in a way where
link |
I like consciously think this really gave me value.
link |
Like I sometimes listen to the daily podcast in the New York Times, way better than the
link |
New York Times themselves, by the way, for people listening.
link |
That's like real journalism is happening for some reason in the podcast space.
link |
It doesn't make sense to me.
link |
But often I listen to it 20 minutes and I would be willing to pay for that, like $5,
link |
$10 for that experience and how difficult.
link |
That's kind of what you're getting at is that little transaction.
link |
How difficult is it to create a frictionless system like Uber has, for example, for other
link |
What's your intuition there?
link |
So first of all, I pay little bits of money to, you know, to send there's something called
link |
courts that does financial things.
link |
I like medium as a site, I don't pay there, but I would.
link |
You had a great post on medium.
link |
I would have loved to pay you a dollar and not others.
link |
I wouldn't have wanted it per se because there should be also sites where that's not actually
link |
The goal is to actually have a broadcast channel that I monetize in some other way if I chose
link |
I mean, I could now.
link |
People know about it.
link |
I'm not doing it, but that's fine with me.
link |
Also the musicians who are making all this music, I don't think the right model is that
link |
you pay a little subscription fee to them, right?
link |
Because people can copy the bits too easily and it's just not that somewhere the value
link |
The value is that a connection was made with real between real human beings, then you can
link |
follow up on that, right?
link |
And create yet more value.
link |
So no, I think there's a lot of open questions here.
link |
Hot open questions, but also, yeah, I do want good recommendation systems that recommend
link |
But it's pretty hard, right?
link |
I don't like them to recommend stuff just based on my browsing history.
link |
I don't like that they're based on stuff they know about me, quote unquote.
link |
That's unknown about me is the most interesting.
link |
So this is the really interesting question.
link |
We may disagree, maybe not.
link |
I think that I love recommender systems and I want to give them everything about me in
link |
a way that I trust.
link |
Yeah, but you don't.
link |
Because so for example, this morning I clicked on, you know, I was pretty sleepy this morning.
link |
I clicked on a story about the Queen of England, right?
link |
I do not give a damn about the Queen of England.
link |
But it was clickbait.
link |
It kind of looked funny and I had to say, what the heck are they talking about there?
link |
I don't want to have my life, you know, heading that direction.
link |
Now that's in my browsing history.
link |
The system and any reasonable system will think that I care about Queen of England.
link |
But that's browsing history.
link |
Right, but you're saying all the trace, all the digital exhaust or whatever, that's been
link |
kind of the models.
link |
If you collect all this stuff, you're going to figure all of us out.
link |
Well, if you're trying to figure out like kind of one person, like Trump or something,
link |
maybe you could figure him out.
link |
But if you're trying to figure out, you know, 500 million people, you know, no way, no way.
link |
I think we are humans are just amazingly rich and complicated.
link |
Every one of us has our little quirks.
link |
Everyone else has our little things that could intrigue us, that we don't even know and will
link |
And there's no sign of it in our past.
link |
But by God, there it comes and, you know, you fall in love with it.
link |
And I don't want a company trying to figure that out for me and anticipate that.
link |
Okay, well, I want them to provide a forum, a market, a place that I kind of go and by
link |
hook or by crook, this happens.
link |
You know, I'm walking down the street and I hear some Chilean music being played and
link |
I never knew I liked Chilean music.
link |
So there is that side and I want them to provide a limited but, you know, interesting place
link |
And so don't try to use your AI to kind of, you know, figure me out and then put me in
link |
a world where you figured me out, you know, no, create huge spaces for human beings where
link |
our creativity and our style will be enriched and come forward and it'll be a lot of more
link |
I won't have people randomly, anonymously putting comments up and especially based on
link |
stuff they know about me, facts that, you know, we are so broken right now.
link |
If you're, you know, especially if you're a celebrity, but, you know, it's about anybody
link |
that anonymous people are hurting lots and lots of people right now.
link |
And that's part of this thing that Silicon Valley is thinking that, you know, just collect
link |
all this information and use it in a great way.
link |
So, you know, I'm not a pessimist, I'm very much an optimist, my nature, but I think that's
link |
just been the wrong path for the whole technology to take.
link |
Be more limited, create, let humans rise up.
link |
Don't try to replace them.
link |
That's the AI mantra.
link |
Don't try to anticipate them.
link |
Don't try to predict them because you're not going to, you're not going to be able to do
link |
You're going to make things worse.
link |
So, right now, just give this a chance.
link |
Right now, the recommender systems are the creepy people in the shadow watching your
link |
So they're looking at traces of you.
link |
They're not directly interacting with you, sort of your close friends and family, the
link |
way they know you is by having conversation, by actually having interactions back and forth.
link |
Do you think there's a place for recommender systems, sort of to step, because you just
link |
emphasize the value of human to human connection.
link |
But yeah, let's give it a chance, AI human connection.
link |
Is there a role for an AI system to have conversations with you in terms of, to try
link |
to figure out what kind of music you like, not by just watching what you listen to, but
link |
actually having a conversation, natural language or otherwise.
link |
So I'm not against this, I just wanted to push back against the, maybe you were saying,
link |
you have options for Facebook.
link |
So there I think it's misplaced.
link |
But I think that distributing...
link |
I'm the one that's depending on Facebook.
link |
That's a hard spot to be.
link |
Human interaction, like on our daily, the context around me in my own home is something that
link |
I don't want some big company to know about at all.
link |
But I would be more than happy to have technology help me with it.
link |
Which kind of technology?
link |
Well, you know, just...
link |
Well, Alexa's done right.
link |
I think Alexa's a research platform right now more than anything else.
link |
But Alexa done right, you know, could do things like I leave the water running in my garden
link |
and I say, hey, Alexa, the water's running in my garden.
link |
And even have Alexa figure out that that means when my wife comes home that she should be
link |
That's a little bit of a reasoning.
link |
I would call that AI and by any kind of stretch, it's a little bit of reasoning.
link |
And it actually kind of would make my life a little easier and better.
link |
And you know, I wouldn't call this a wow moment, but I kind of think that overall rice is human
link |
happiness up to have that kind of thing.
link |
And not when you're lonely, Alexa knowing loneliness.
link |
No, no, I don't want Alexa to feel intrusive.
link |
And I don't want just the designer of the system to kind of work all this out.
link |
I really want to have a lot of control and I want transparency and control.
link |
And if the company can stand up and give me that in the context of technology, I think
link |
they're going to first of all be way more successful than our current generation.
link |
And like I said, I was measuring Microsoft, you know, I really think they're pivoting
link |
to kind of be the trusted old uncle.
link |
You know, I think that they get that this is a way to go, that if you let people find
link |
technology empowers them to have more control and have control, not just over privacy, but
link |
over this rich set of interactions, that that people are going to like that a lot more.
link |
And that's, that's the right business model going forward.
link |
What does control over privacy look like?
link |
Do you think you should be able to just view all the data that?
link |
No, it's much more than that.
link |
I mean, first of all, it should be an individual decision.
link |
Some people don't want privacy.
link |
They want their whole life out there.
link |
Other people's want it.
link |
Privacy is not a zero one.
link |
It's not a legal thing.
link |
It's not just about which date is available, which is not.
link |
I like to recall to people that, you know, a couple of hundred years ago, everyone, there
link |
was not really big cities.
link |
Everyone lived in on the countryside and villages and in villages, everybody knew everything
link |
Very, you didn't have any privacy.
link |
Are we better off now?
link |
Well, you know, arguably no, because what did you get for that loss of at least certain
link |
Well, people helped each other because they know everything about you.
link |
They know something bad's happening.
link |
They will help you with that.
link |
And now you live in a big city, no one knows the amount.
link |
So it kind of depends the answer.
link |
I want certain people who I trust and there should be relationships.
link |
I should kind of manage all those, but who knows what about me?
link |
I should have some agency there.
link |
It shouldn't be a drift in a city of technology where I have no agency.
link |
I don't want to go reading things and checking boxes.
link |
So I don't know how to do that.
link |
And I'm not a privacy researcher per se.
link |
I recognize the vast complexity of this.
link |
It's not just technology.
link |
It's not just legal scholars meeting technologists.
link |
There's got to be kind of a whole layers around it.
link |
And so when I alluded to this emerging engineering field, this is a big part of it.
link |
Well, like an electrical engineering come game, I'm not running around in the time,
link |
but you just didn't plug electricity into walls and all kind of work.
link |
You know, I had to have like underwriters laboratory that reassured you that that plug
link |
is not going to burn up your house and that that machine will do this and that and everything.
link |
There'll be whole people who can install things.
link |
There'll be people who can watch the installers.
link |
There'll be a whole layers, you know, an onion of these kind of things.
link |
And for things as deep and interesting as privacy, which is as least as interested as electricity,
link |
that's going to take decades to kind of work out, but it's going to require a lot of new
link |
structures that we don't have right now.
link |
So it's kind of hard to talk about it.
link |
And you're saying there's a lot of money to be made if you get it right.
link |
So I should look at a lot of money to be made.
link |
And all these things that provide human services and people recognize them as useful parts
link |
So yeah, the dialogue sometimes goes from the exuberant technologists to the no technology
link |
And that's, you know, in a public discourse, you know, in newsrooms, you see too much of
link |
this kind of thing.
link |
And the sober discussions in the middle, which are the challenging ones to have or where
link |
we need to be having our conversations.
link |
And you know, it's just not, actually, there's not many forum for those.
link |
You know, there's, that's, that's kind of what I would look for.
link |
Maybe I could go and I could read a comment section of something and it would actually
link |
be this kind of dialogue going back and forth.
link |
You don't see much of this, right?
link |
Which is why actually there's a resurgence of podcasts out of all, because people are
link |
really hungry for conversation, but there's technology is not helping much.
link |
So comment sections of anything, including YouTube is not hurting and not helping.
link |
And you think technically speaking is possible to help?
link |
I don't know the answers, but it's a less anonymity, a little more locality, you know,
link |
worlds that you kind of enter in and you trust the people there in those worlds so that when
link |
you start having a discussion, you know, not only is that people are not going to hurt
link |
you, but it's not going to be a total waste of your time because there's a lot of wasting
link |
of time that, you know, a lot of us, I pulled out of Facebook early on because it was clearly
link |
going to waste a lot of my time, even though there was some value.
link |
And so yeah, worlds that are somehow you enter in, you know what you're getting, and it's
link |
kind of appeals to you, new things might happen, but you kind of have some trust in that world.
link |
And there's some deep, interesting, complex psychological aspects around anonymity, how
link |
that changes human behavior that's quite dark.
link |
I think a lot of us are, especially those of us who really loved the advent of technology,
link |
I loved social networks when they came out, I didn't see any negatives there at all.
link |
But then I started seeing comment sections, I think it was maybe, you know, the CNN or
link |
And I started to go, wow, this, this darkness I just did not know about, and our technology
link |
now amplifying it.
link |
So sorry for the big philosophical question, but on that topic, do you think human beings,
link |
because you've also, out of all things, had a foot in psychology too, do you think human
link |
beings are fundamentally good?
link |
Like all of us have good intent that could be mined, or is it, depending on context and
link |
environment, everybody could be evil?
link |
So my answer is fundamentally good, but fundamentally limited.
link |
All of us have very, you know, blinkers on.
link |
We don't see the other person's pain that easily.
link |
We don't see the other person's point of view that easily.
link |
We're very much in our own head, in our own world.
link |
And on my good days, I think that technology could open us up to more perspectives and
link |
more, less blinkered and more understanding, you know, a lot of wars in human history happened
link |
because of just ignorance.
link |
They didn't, they thought the other person was doing this, well, that person wasn't
link |
doing this, and we have a huge amounts of that.
link |
But in my lifetime, I've not seen technology really help in that way yet.
link |
And I do, I do, I do believe in that, but, you know, no, I think fundamentally humans
link |
People have grieves, and so you have grudges, and those things cause them to do things they
link |
probably wouldn't want.
link |
They regret it often.
link |
So no, I, I, I think it's a, you know, part of the progress that technology is to indeed
link |
allow it to be a little easier to be the real good person you actually are.
link |
Well, but do you think individual human life or society can be modeled as an optimization
link |
Um, not the way I think typically, I mean, that's, you're talking about one of the most
link |
complex phenomena in the whole, you know, in all, which individual human life or society
link |
I mean, individual human life is, is amazingly complex and, um, so, uh, you know, optimization
link |
is kind of just one branch of mathematics that talks about certain kind of things.
link |
And, uh, it just feels way too limited for the complexity of, uh, such things.
link |
What properties of optimization problems do you think, so do you think most interesting
link |
problems that could be solved through optimization, uh, what kind of properties does that surface
link |
have nonconvexity, convexity, linearity, all those kinds of things, saddle points.
link |
Well, so optimization is just one piece of mathematics.
link |
You know, there's like, you just, even in our era, we're aware that say sampling, um,
link |
it's coming up examples of something, um, come in with a distribution.
link |
What's optimization?
link |
Well, you think you can, if you're a kind of a certain kind of math, but you can try
link |
to blend them and make them seem to be sort of the same thing.
link |
But optimization is roughly speaking, trying to, uh, find a point that, um, a single point
link |
that is the optimum of a criterion function of some kind, um, and sampling is trying to,
link |
from that same surface, treat that as a distribution or density and find prop points that have
link |
So, um, I want the entire distribution and the sampling paradigm and I want the, um,
link |
you know, the single point, that's the best point in the, in the sample, in the, uh, optimization
link |
Now, if you were optimizing in the space of probability measures, the output of that could
link |
be a whole probability distribution.
link |
So you can start to make these things the same, but in mathematics, if you go too high
link |
up that kind of abstraction hierarchy, you start to lose the, uh, you know, the ability
link |
to do the interesting theorems.
link |
So you kind of don't try that.
link |
You don't try to overly over abstract.
link |
So as a small tangent, what kind of world do you, do you find more appealing?
link |
One that is deterministic or stochastic?
link |
Uh, well, that's easy.
link |
I mean, I'm a statistician, you know, the world is highly stochastic.
link |
Wait, I don't know what's going to happen in the next five minutes.
link |
Cause what you're going to ask, what we're going to do, what I'll say, massive uncertainty,
link |
you know, massive uncertainty.
link |
And so the best I can do is have come rough sense or probability distribution on things
link |
and somehow use that in my reasoning about what to do now.
link |
So how does the distributed at scale, when you have multi agent systems, uh, look like,
link |
so optimization can optimize sort of, it makes a lot more sense sort of, uh, at least mine
link |
from a robotics perspective, for a single robot, for a single agent, trying to optimize
link |
some objective function, uh, what, when you start to enter the real world, this game theory
link |
ready concepts starts popping up and that's how do you see optimization in this?
link |
Cause you've talked about markets and a scale.
link |
What does that look like?
link |
Do you see it as optimization?
link |
Do you see it as a sampling?
link |
Do you see like how, how should you, yeah, these all blend together.
link |
Um, and a system designer thinking about how to build an incentivized system will have
link |
a blend of all these things.
link |
So you know, a particle in a potential well is optimizing a functional called a Lagrangian.
link |
The particle doesn't know that.
link |
There's no algorithm running that does that.
link |
It's, it's, so it's a description mathematically of something that helps us understand as analysts,
link |
And so the same thing will happen when we talk about, you know, mixtures of humans and
link |
computers and markets and so on and so forth, there'll be certain principles that allow
link |
us to understand what's happening and whether or not the actual algorithms are being used
link |
by any sense is not clear.
link |
Um, now at, at some point I may have set up a multi agent or market kind of system and
link |
I'm now thinking about an individual agent in that system and they're asked to do some
link |
task and they're incentivized and somewhere they get certain signals and they, they have
link |
Maybe what they will do at that point is they just won't know the answer.
link |
They may have to optimize to find an answer.
link |
So an autos could be embedded inside of an overall market, uh, you know, and game theory
link |
is, is very, very broad.
link |
Um, it is often studied very narrowly for certain kinds of problems.
link |
Um, but it's roughly speaking, this is just the, I don't know what you're going to do.
link |
I kind of anticipate that a little bit and you anticipate what I'm anticipating and we
link |
kind of go back and forth in our own minds.
link |
We run kind of thought experiments.
link |
You've talked about this interesting point in terms of game theory, you know, most optimization
link |
problems really hate saddle points.
link |
Maybe you can describe what saddle points are, but I had heard you kind of mentioned
link |
that there's a, there's a branch of optimization that you could try to explicitly look for
link |
saddle points as a good thing.
link |
Oh, not optimization.
link |
That's just game theory.
link |
That's, so, uh, there's all kinds of different equilibrium game theory and some of them are
link |
highly explanatory behavior.
link |
They're not attempting to be algorithmic.
link |
They're just trying to say, if you happen to be at this equilibrium, you would see certain
link |
kind of behavior and we see that in real life.
link |
That's what an economist wants to do, especially behavioral economists, um, uh, in, in continuous,
link |
uh, differential game theory, you're in continuous spaces, a, um, some of the simplest
link |
equilibria are saddle points and Nash equilibrium as a saddle point, it's a special kind of
link |
So, uh, classically in game theory, you were trying to find Nash equilibria and algorithmic
link |
game theory, you're trying to find algorithms that would find them, uh, and so you're trying
link |
to find saddle points.
link |
I mean, so that's, that's literally what you're trying to do.
link |
Um, but you know, any economist knows that Nash equilibria, uh, have their limitations.
link |
They are definitely not that explanatory in many situations.
link |
You're not what you really want.
link |
Um, there's other kind of equilibria and there's names associated with these cause they came
link |
from history with certain people working on them, but there'll be new ones emerging.
link |
So you know, one example is a Stackelberg equilibrium.
link |
So you know, Nash, you and I are both playing this game against each other or for each other,
link |
maybe it's cooperative and we're both going to think it through and then we're going to
link |
decide and we're going to do our thing simultaneously.
link |
You know, in a Stackelberg, no, I'm going to be the first mover.
link |
I'm going to make a move.
link |
You're going to look at my move and then you're going to make yours.
link |
Now, since I know you're going to look at my move, I anticipate what you're going to
link |
And so I don't do something stupid, but, and, but then I know that you were also anticipating
link |
So we're kind of going back and so forth.
link |
Why, but there is then a first mover thing.
link |
And so there's a different equilibria, all right.
link |
And, uh, so just mathematically, yeah, these things have certain topologies and certain
link |
shapes that are like salivates and then algorithmically or dynamically, how do you move towards them?
link |
How do you move away from things?
link |
Um, you know, so some of these questions have answers.
link |
They've been studied.
link |
Others do not, and especially if it becomes stochastic, especially if there's large numbers
link |
of decentralized things, there's just, uh, you know, young people get in this field who
link |
kind of think it's all done because we have, you know, TensorFlow.
link |
Well, no, these are all open problems and they're really important and interesting.
link |
And it's about strategic settings.
link |
How do I collect data?
link |
I suppose I don't know what you're going to do because I don't know you very well, right?
link |
Well, I got to collect data about you.
link |
So maybe I want to push you in a part of the space where I don't know much about you.
link |
So I can get data.
link |
And then later I'll realize that you'll never, you'll never go there because of the way the
link |
But, you know, that's part of the overall, you know, data analysis context.
link |
Even the game of poker is fascinating space.
link |
Whenever there's any uncertainty or lack of information, it's, it's a super exciting
link |
Uh, just a lingering optimization for a second.
link |
So when we look at deep learning, it's essentially a minimization of a complicated loss function.
link |
So is there something insightful or hopeful that you see in the kinds of function surface
link |
that lost functions that deep learning in, in the real world is trying to optimize over?
link |
Is there something interesting as it's just the usual kind of problems of optimization?
link |
I think from an optimization point of view, that surface first of all, it's pretty smooth.
link |
And secondly, if there's over, if it's over parameterized, there's kind of lots of paths
link |
down to reasonable optima.
link |
And so kind of the getting downhill to the, to an optimum is, is viewed as not as hard
link |
as you might have expected in high dimensions.
link |
The fact that some optima tend to be really good ones and others not so good and you tend
link |
to, it's not, sometimes you find the good ones is, is sort of still needs explanation.
link |
But, but the particular surface is coming from the particular generation of neural nets.
link |
I kind of suspect those will, those will change in 10 years.
link |
It will not be exactly those surfaces.
link |
There'll be some others that are, an optimization theory will help contribute to why other surfaces
link |
or why other algorithms.
link |
Layers of arithmetic operations with a little bit of nonlinearity, that's not, that didn't
link |
come from neuroscience per se.
link |
I mean, maybe in the minds of some of the people working on it, they were thinking about brains,
link |
but they were arithmetic circuits in all kinds of fields, you know, computer science control
link |
And that layers of these could transform things in certain ways and that if it's smooth, maybe
link |
you could, you know, find parameter values, you know, it's a big, is a, is a, is a, is
link |
a sort of big discovery that it's, it's working.
link |
It's able to work at this scale, but I don't think that we're stuck with that and we're,
link |
we're certainly not stuck with that because we're understanding the brain.
link |
So in terms of, on the algorithm side, sort of gradient descent, do you think we're stuck
link |
with gradient descent as variants of it?
link |
What variants do you find interesting or do you think there'll be something else invented
link |
that is able to walk all over these optimization spaces in more interesting ways?
link |
So there's a co design of the surface and they are the architecture and the algorithm.
link |
So if you just ask if we stay with the kind of architectures that we have now, not just
link |
neural nets, but, you know, phase retrieval architectures or maybe completion architectures
link |
You know, I think we've kind of come to a place where a stochastic gradient algorithms
link |
are dominant and there are versions that, you know, they're a little better than others.
link |
They, you know, have more guarantees, they're more robust and so on and there's ongoing
link |
research to kind of figure out which is the best time for which situation.
link |
But I think that that'll start to co evolve, that that'll put pressure on the actual architecture
link |
and so we shouldn't do it in this particular way, we should do it in a different way because
link |
this other algorithm is now available if you do it in a different way.
link |
So that I can't really anticipate that co evolution process, but you know, gradients
link |
are amazing mathematical objects.
link |
They have a lot of people who sort of study them more deeply mathematically or kind of
link |
shocked about what they are and what they can do.
link |
I mean, to think about this way, if I suppose that I tell you, if you move along the x axis,
link |
you get, you know, you go uphill in some objective by, you know, three units, whereas if you move
link |
along the y axis, you go uphill by seven units, right now, I'm going to only allow you to
link |
move a certain, you know, unit distance, right?
link |
What are you going to do?
link |
Well, the most not people will say, I'm going to go along the y axis, I'm getting the biggest
link |
bang for my buck, you know, and my buck is only one unit.
link |
So I'm going to put all of it in the y axis, right?
link |
And why should I even take any of my strength, my step size and put any of it in the x axis
link |
because I'm getting less bang for my buck.
link |
That seems like a completely, you know, clear argument and it's wrong because the gradient
link |
direction is not to go along the y axis, it's to take a little bit of the x axis.
link |
And that, to understand that, you have to, you have to know some math.
link |
And so even a, you know, trivial, so called operator like gradient is not trivial and
link |
so, you know, exploiting its properties is still very, very important.
link |
Now we know that just creating descent has got all kinds of problems that get stuck in
link |
many ways and it had never, you know, good dimension dependence and so on.
link |
So my own line of work recently has been about what kinds of stochasticity, how can we get
link |
dimension dependence?
link |
How can we do the theory of that?
link |
And we've come up with pretty favorable results with certain kinds of stochasticity.
link |
We have sufficient conditions generally.
link |
We know if you, if you do this, we will give you a good guarantee.
link |
We don't have necessary conditions that it must be done a certain way in general.
link |
So stochasticity, how much randomness to inject into the, into the walking along the gradient.
link |
And what kind of randomness?
link |
Why is randomness good in this process?
link |
Why is stochasticity good?
link |
So I can give you simple answers, but in some sense, again, it's kind of amazing.
link |
Stochasticity just, you know, particular features of a surface that could have hurt
link |
you if you were doing one thing, you know, deterministically won't hurt you because,
link |
you know, by chance, you know, there's very little chance that you would get hurt.
link |
And, you know, so here stochasticity, you know, it's just kind of saves you from some
link |
of the particular features of surfaces that, you know, in fact, if you think about, you
link |
know, surfaces that are discontinuous in a first derivative, like, you know, absolute
link |
value function, you will go down and hit that point where there's non differentiability.
link |
And if you're running a deterministic argument at that point, you can really do something
link |
Where stochasticity just means it's pretty unlikely that's going to happen.
link |
You're going to get, you're going to hit that point.
link |
So, you know, it's, again, not trivial to analyze, but, especially in higher dimensions,
link |
also stochasticity, our intuition isn't very good about it, but it has properties that
link |
kind of are very appealing in high dimensions for kind of law of large number of reasons.
link |
So it's all part of the mathematics to kind of, that's what's fun to work in the field
link |
is that you get to try to understand this mathematics.
link |
But long story short, you know, partly empirically, it was discovered stochastic gradient is very
link |
effective and theory kind of followed, I'd say, that, but I don't see that we're getting
link |
it clearly out of that.
link |
What's the most beautiful, mysterious, or profound idea to you in optimization?
link |
I don't know the most, but let me just say that, you know, Nesterov's work on Nesterov
link |
acceleration to me is pretty surprising and pretty deep.
link |
Can you elaborate?
link |
Well, Nesterov acceleration is just that, I suppose that we are going to use gradients
link |
to move around into space for the reasons I've alluded to, there are nice directions
link |
And suppose that I tell you that you're only allowed to use gradients, you're not going
link |
to be allowed to use this local person, it can only sense kind of the change in the surface.
link |
But I'm going to give you kind of a computer that's able to store all your previous gradients.
link |
And so you start to learn something about the surface.
link |
And I'm going to restrict you to maybe move in the direction of like a linear span of
link |
all the gradients.
link |
So you can't kind of just move in some arbitrary direction, right?
link |
So now we have a well defined mathematical complexity model.
link |
There's a certain classes of algorithms that can do that and others that can't.
link |
And we can ask for certain kinds of surfaces, how fast can you get down to the optimum?
link |
So there's an answers to these.
link |
So for a, you know, a smooth convex function, there's an answer, which is one over the number
link |
of steps squared is that you will be within a ball of that size after K steps.
link |
Gradient descent in particular has a slower rate.
link |
It's one over K. Okay.
link |
So you could ask, is gradient descent actually, even though we know it's a good algorithm,
link |
is it the best algorithm in the sense of the answer is no, but well, well, not clear yet
link |
because one over K score is a lower bound.
link |
That's probably the best you can do.
link |
Gradient is one over K, but these are something better.
link |
And so I think as a surprise to most, the Nesterov discovered a new algorithm that has got two
link |
It uses two gradients and puts those together in a certain kind of obscure way.
link |
And the thing doesn't even move downhill all the time.
link |
It sometimes goes back uphill.
link |
And if you're a physicist, that kind of makes some sense.
link |
You're building up some momentum and that is kind of the right intuition, but that that
link |
intuition is not enough to understand kind of how to do it and why it works.
link |
It achieves one over K squared and it has a mathematical structure and it's still kind
link |
of to this day, a lot of us are writing papers and trying to explore that and understand it.
link |
So there are lots of cool ideas and optimization, but just kind of using gradients, I think
link |
is number one that goes back, you know, 150 years.
link |
And then Nesterov, I think has made a major contribution with this idea.
link |
So like you said, gradients themselves are in some sense mysterious.
link |
Yeah, they're not, they're not as trivial as not as much as coordinate descent is more
link |
You just pick one of the coordinates.
link |
That's how we think that's how our human minds think and gradients are not that easy
link |
for our human mind to grapple with.
link |
An absurd question, but what is statistics?
link |
So here it's a little bit, it's somewhere between math and science and technology.
link |
It's somewhere in that convex hole.
link |
So it's a set of principles that allow you to make inferences that have got some reason
link |
to be believed and also principles that allow you to make decisions where you can have some
link |
reason to believe you're not going to make errors.
link |
So all of that requires some assumptions about what do you mean by an error?
link |
What do you mean by, you know, the probabilities and, but, you know, after you start making
link |
some assumptions, you're led to conclusions that, yes, I can guarantee that, you know,
link |
if you do this in this way, your probability of making an error will be small.
link |
Your probability of continuing to not make errors over time will be small.
link |
And probability you found something that's real will be small, will be high.
link |
So decision making is a big part of that?
link |
So decision making is a big part, yeah.
link |
So the original, so statistics, you know, short history was that, you know, it's kind
link |
of goes back sort of as a formal discipline, you know, 250 years or so.
link |
It was called inverse probability because around that era, probability was developed
link |
sort of especially to explain gambling situations.
link |
So you would say, well, given the state of nature is this, there's a certain roulette
link |
board that has a certain mechanism in it.
link |
What kind of outcomes do I expect to see?
link |
And especially if I do things longer, longer amounts of time, what outcomes will I see
link |
and the physicists start to pay attention to this?
link |
And then people say, well, given, let's turn the problem around.
link |
What if I saw certain outcomes, could I infer what the underlying mechanism was?
link |
That's an inverse problem.
link |
And in fact, for quite a while, statistics was called inverse probability.
link |
That was the name of the field.
link |
And I believe that it was Laplace, who was working in Napoleon's government, who was
link |
trying, who needed to do a census of France, learn about the people there.
link |
So he went and gathered data and he analyzed that data to determine policy and said, let's
link |
call this field that does this kind of thing, statistics, because the word state is in there.
link |
And in French, that's état, but it's the study of data for the state.
link |
So anyway, that caught on and it's been called statistics ever since.
link |
But by the time it got formalized, it was sort of in the 30s.
link |
And around that time, there was game theory and decision theory developed nearby.
link |
People in that era didn't think of themselves as either computer science or statistics or
link |
They were all, they were all the above.
link |
And so, you know, von Neumann is developing game theory, but also thinking of that as
link |
decision theory, Wald is an econometrician, developing decision theory, and then, you
link |
know, turning that into statistics.
link |
And so it's all about, here's not just data and you analyze it, here's a loss function,
link |
here's what you care about, here's the question you're trying to ask.
link |
Here is a probability model and here's the risk you will face if you make certain decisions.
link |
And to this day, in most advanced statistical curricula, you teach decision theory as the
link |
starting point, and then it branches out into the two branches of Bayesian and Frequentist.
link |
But that's, it's all about decisions.
link |
In statistics, what is the most beautiful, mysterious, maybe surprising idea that you've
link |
Yeah, good question.
link |
I mean, there's a bunch of surprising ones, there's something that's way too technical
link |
for this thing, but something called James Stein estimation, which is kind of surprising
link |
and really takes time to wrap your head around.
link |
Can you try to maybe?
link |
I think I don't want to even want to try.
link |
Let me just say a colleague at Steven Stigler at University of Chicago wrote a really beautiful
link |
paper on James Stein estimation, which is helps to, its views of paradox, it kind of
link |
defeats the mind's attempts to understand it, but you can, and Steve has a nice perspective
link |
So one of the troubles with statistics is that it's like in physics, that are in quantum
link |
physics, you have multiple interpretations.
link |
There's a wave and particle duality in physics and you get used to that over time, but it's
link |
still kind of haunts you that you don't really, you know, quite understand the relationship.
link |
The electrons of wave and electrons of particle, well, well, same thing happens here.
link |
There's Bayesian ways of thinking and Frequentist and they are different.
link |
They, they often, they sometimes become sort of the same in practice, but they are physically
link |
And then in some practice, they are not the same at all.
link |
They give you rather different answers.
link |
And so it is very much like wave and particle duality and that is something that you have
link |
to kind of get used to in the field.
link |
Can you define Bayesian and Frequentist?
link |
In decision theory, you can make, I have a, like I have a video that people could see
link |
it's called, are you a Bayesian or a Frequentist and kind of help try to, to, to make it really
link |
It comes from decision theory.
link |
So, you know, decision theory, you're talking about loss functions, which are a function
link |
of data X and parameter theta.
link |
So there are a function of two arguments, okay?
link |
Neither one of those arguments is known.
link |
You don't know the data a priori.
link |
It's random and the parameters unknown, all right?
link |
So you have this function of two things you don't know and you're trying to say, I want
link |
that function to be small.
link |
I want small loss, right?
link |
Well, what are you going to do?
link |
So you sort of say, well, I'm going to average over these quantities or maximize over them
link |
or something so that, you know, I turn that uncertainty into something certain.
link |
So you could look at the first argument and average over it or you could look at the second
link |
argument average over it.
link |
That's Bayesian Frequentist.
link |
The Frequentist says, I'm going to look at the X, the data, and I'm going to take that
link |
as random and I'm going to average over the distribution.
link |
So I take the expectation loss under X, theta is held fixed, all right?
link |
That's called the risk.
link |
And so it's looking at all the data sets you could get, all right?
link |
And say how well will a certain procedure do under all those data sets?
link |
That's called a Frequentist guarantee, all right?
link |
So I think it is very appropriate when you're building a piece of software and you're shipping
link |
it out there and people are using all kinds of data sets.
link |
You want to have a stamp, a guarantee on it that as people run it on many, many data sets
link |
that you never even thought about that 95% of the time it will do the right thing.
link |
Perfectly reasonable.
link |
The Bayesian Perspective says, well, no, I'm going to look at the other argument of the
link |
loss function, the theta part, okay?
link |
That's unknown and I'm uncertain about it.
link |
So I could have my own personal probability for what it is, you know, how many tall people
link |
are there out there?
link |
I'm trying to infer the average height of the population.
link |
Well, I have an idea roughly what the height is.
link |
So I'm going to average over the theta.
link |
So now that loss function has only now, again, one argument's gone.
link |
Now it's a function of X.
link |
And that's what a Bayesian does is they say, well, let's just focus on the particular X
link |
we got, the data set we got, we condition on that.
link |
Condition on the X, I say something about my loss.
link |
That's a Bayesian approach to things.
link |
And the Bayesian will argue that it's not relevant to look at all the other data sets
link |
you could have gotten and average over them, the frequentist approach.
link |
It's really only the data sets you got, all right?
link |
And I do agree with that, especially in situations where you're working with a scientist, you
link |
can learn a lot about the domain and you're really only focused on certain kinds of data
link |
and you've gathered your data and you make inferences.
link |
I don't agree with it though that, you know, in the sense that there are needs for frequentist
link |
In the software people are using out there, you want to say something.
link |
So these two things have to fight each other a little bit, but they have to blend.
link |
So long story short, there's a set of ideas that are right in the middle.
link |
They're called empirical bays.
link |
And empirical bays sort of starts with the Bayesian framework.
link |
It's kind of arguably philosophically more, you know, reasonable and kosher, right down
link |
a bunch of the math that kind of flows from that and then realize there's a bunch of things
link |
you don't know because it's the real world and you don't know everything.
link |
So you're uncertain about certain quantities.
link |
At that point, ask, is there a reasonable way to plug in an estimate for those things?
link |
And in some cases, there's quite a reasonable thing to do to plug in.
link |
There's a natural thing you can observe in the world that you can plug in and then do
link |
a little bit more mathematics and assure yourself it's really good.
link |
So based on math or based on human expertise, what are the good things?
link |
They're both going in.
link |
The Bayesian framework allows you to put a lot of human expertise in.
link |
But the math kind of guides you along that path and then kind of reassures at the end,
link |
you could put that stamp of approval under certain assumptions, this thing will work.
link |
So you asked question, was my favorite, you know, or was the most surprising, nice idea.
link |
So one that is more accessible is something called false discovery rate, which is, you
link |
know, you're making not just one hypothesis test or making one decision, you're making
link |
a whole bag of them.
link |
And in that bag of decisions, you look at the ones where you made a discovery, you announced
link |
that something interesting had happened.
link |
That's going to be some subset of your big bag.
link |
In the ones you made a discovery, which subset of those are bad?
link |
There are false, false discoveries.
link |
You like the fraction of your false discoveries among your discoveries to be small.
link |
That's a different criterion than accuracy or precision or recall or sensitivity and
link |
It's a different quantity.
link |
Those latter ones are almost all of them have more of a frequentist flavor.
link |
They say, given the truth is that the null hypothesis is true, here's what accuracy I
link |
would get or given that the alternative is true.
link |
Here's what I would get.
link |
So it's kind of going forward from the state of nature to the data.
link |
The Bayesian goes the other direction from the data back to the state of nature.
link |
And that's actually what false discovery rate is.
link |
It says, given you made a discovery, okay, that's conditioned on your data.
link |
What's the probability of the hypothesis is going the other direction.
link |
And so the classical frequents look at that and say, well, I can't know that there's
link |
some priors needed in that.
link |
And the empirical Bayesian goes ahead and plows forward and starts writing down to these
link |
formulas and realizes at some point, some of those things can actually be estimated
link |
in a reasonable way.
link |
And so it's a beautiful set of ideas.
link |
So this kind of line of argument has come out, it's not certainly mine, but it sort
link |
of came out from Robbins around 1960.
link |
Brad Efron has written beautifully about this in various papers and books.
link |
And the FDR is, you know, Ben Yamini in Israel, John Story did this Bayesian interpretation
link |
So I've just absorbed these things over the years and finally did a very healthy way to
link |
think about statistics.
link |
Let me ask you about intelligence to jump slightly back out into philosophy, perhaps.
link |
You said that maybe you can elaborate, but you said that defining just even the question
link |
of what is intelligence is a very difficult question.
link |
Is it a useful question?
link |
Do you think we'll one day understand the fundamentals of human intelligence and what
link |
it means, you know, have good benchmarks for general intelligence that we put before
link |
So I don't work on these topics so much.
link |
You're really asking the question of for a psychologist, really, and I studied some,
link |
but I don't consider myself at least an expert at this point.
link |
You know, a psychologist aims to understand human intelligence, right?
link |
And I think maybe the psychologists, I know are fairly humble about this.
link |
They might try to understand how a baby understands, you know, whether something's a solid or liquid
link |
or whether something's hidden or not.
link |
And maybe how a child starts to learn the meaning of certain words, what's a verb, what's
link |
a noun, and also, you know, slowly but surely trying to figure out things.
link |
But humans ability to take a really complicated environment, reason about it, abstract about
link |
it, find the right abstractions, communicate about it, interact and so on is just, you
link |
know, really staggeringly rich and complicated.
link |
And so, you know, I think in all humidly, we don't think we're kind of aiming for that
link |
in the near future, a certain like psychologist doing experiments with babies in the lab or
link |
with people talking has a much more limited aspiration.
link |
And you know, Kahneman Dversky would look at our reasoning patterns and they're not deeply
link |
understanding all the how we do our reasoning, but they're sort of saying, here's some oddities
link |
about the reasoning and some things you need to think about it.
link |
But also, as I emphasize in some things I've been writing about, you know, AI, the revolution
link |
hasn't happened yet.
link |
I've been emphasizing that, you know, if you step back and look at intelligence systems
link |
of any kind and whatever you mean by intelligence, it's not just the humans or the animals or
link |
you know, the plants or whatever, you know, so a market that brings goods into a city,
link |
you know, food to restaurants or something every day is a system.
link |
It's a decentralized set of decisions looking at it from far enough away, just like a collection
link |
Everyone, every neuron is making its own little decisions, presumably in some way.
link |
And if you step back enough, every little part of an economic system is making it solid
link |
And just like with a brain, who knows what any of the neuron doesn't know what the overall
link |
goal is, right, but something happens at some aggregate level.
link |
Same thing with the economy.
link |
People eat in a city and it's robust.
link |
It works at all scales, small villages to big cities.
link |
It's been working for thousands of years, it works rain or shine, so it's adaptive.
link |
So all the kind of, you know, those are adjectives, one tends to apply to intelligent systems,
link |
robust, adaptive, you know, you don't need to keep adjusting it, it's self, self healing,
link |
whatever, plus not perfect, you know, intelligences are never perfect and markets are not perfect.
link |
But I do not believe in this ear that you cannot, that you can say, well, our computers
link |
are humans are smart, but you know, no markets are not more markets are.
link |
So they are intelligent.
link |
Now we humans didn't evolve to be markets.
link |
We've been participating in them, right, but we are not ourselves a market per se.
link |
The neurons could be viewed as the market.
link |
There's economic, you know, neuroscience kind of perspective, that's interesting to pursue
link |
The point though is, is that if you were to study humans and really be a world's best
link |
psychologist, studied for thousands of years and come up with the theory of human intelligence,
link |
you might have never discovered principles of markets, you know, spy demand curves and
link |
you know, matching and auctions and all that.
link |
Those are real principles and they lead to an form of intelligence that's not maybe human
link |
It's arguably another kind of intelligence.
link |
There probably are third kinds of intelligence or fourth that none of us are really thinking
link |
too much about right now.
link |
So if you really, and then all those are relevant to computer systems in the future, certainly
link |
the market one is relevant right now, whereas understanding human intelligence is not so
link |
clear that it's relevant right now, probably not.
link |
So if you want general intelligence, whatever one means by that or, you know, understanding
link |
intelligence in a deep sense and all that, it is definitely has to be not just human
link |
It's got to be this broader thing.
link |
And that's not a mystery.
link |
Markets are intelligent.
link |
So you know, it's definitely not just a philosophical sense to say, we got to move beyond intelligence,
link |
human intelligence.
link |
That sounds ridiculous.
link |
And in a block, we'll see to find different kinds of like intelligent infrastructure,
link |
II, which I really like.
link |
Some of the concept you just been describing, do you see ourselves, we see earth, human
link |
civilization as a single organism.
link |
Do you think the intelligence of that organism, when you think from the perspective of markets
link |
and intelligence infrastructure is increasing?
link |
Is it increasing linearly?
link |
Is it increasing exponentially?
link |
What do you think the future of that intelligence?
link |
Yeah, I don't know.
link |
I don't tend to think, I don't tend to answer questions like that because, you know, that's
link |
I was hoping to catch your off guard.
link |
Well, again, because you said it's so far in the future, it's fun to ask and you'll
link |
probably, you know, like you said, predicting the future is really nearly impossible.
link |
But say as an axiom, one day we create a human level, a super human level intelligent, not
link |
the scale of markets, but the scale of an individual.
link |
What do you think is, what do you think it would take to do that?
link |
Or maybe to ask another question is how would that system be different than the biological
link |
human beings that we see around us today?
link |
Is it possible to say anything interesting to that question or is it just a stupid question?
link |
It's not a stupid question, but it's science fiction.
link |
And so I'm totally happy to read science fiction and think about it from time to time
link |
I loved that there was this like brain in a vat kind of, you know, little thing that
link |
people were talking about when I was a student.
link |
I remember, you know, imagine that, you know, between your brain and your body, there's,
link |
you know, there's a bunch of wires, right?
link |
And suppose that every one of them was replaced with a literal wire and then suppose that
link |
wire was turned into actually a little wireless, you know, there's a receiver and sender.
link |
So the brain has got all the senders and receiver, you know, on all of its exiting, you know,
link |
axons and all the dendrites down in the body are replaced with senders and receivers.
link |
Now you could move the body off somewhere and put the brain in a vat, right?
link |
And then you could do things like start killing off those senders receivers one by one.
link |
And after you've killed off all of them, where is that person?
link |
You know, they thought they were out in the body walking around the world and they moved
link |
So those are science fiction things.
link |
Those are fun to think about.
link |
It's just intriguing about where's, what is thought, where is it and all that.
link |
And I think every 18 year old, it's to take philosophy classes and think about these things.
link |
And I think that everyone should think about what could happen in society that's kind of
link |
But I really don't think that's the right thing for most of us that are my age group
link |
to be doing and thinking about.
link |
I really think that we have so many more present, you know, first challenges and dangers and
link |
real things to build and all that, such that, you know, spending too much time on science
link |
fiction, at least in public fora like this, I think is, is not what we should be doing.
link |
Maybe over beers in private.
link |
I'm well, well, I'm not going to broadcast where I have beers because this is going to
link |
And I know a lot of people showing up there, but yeah, I'll, I love Facebook.
link |
Twitter, Amazon, YouTube, I have, I'm optimistic and hopeful, but maybe, maybe I don't have
link |
grounds for such optimism and hope.
link |
Let me ask you, you've mentored some of the brightest sort of some of the seminal figures
link |
Can you give advice to people who are undergraduates today?
link |
What does it take to take, you know, advice on their journey if they're interested in
link |
machine learning and AI and, and the ideas of markets from economics and psychology and
link |
all the kinds of things that you're exploring?
link |
What, what, what steps they take on that journey?
link |
Well, yeah, first of all, the door is open and second, it's a journey.
link |
I like your language there.
link |
It is not that you're so brilliant and you have great brilliant ideas and therefore that's,
link |
that's just, you know, that's how you have success or that's how you enter into the field.
link |
It's that you apprentice yourself, you, you spend a lot of time, you work on hard things,
link |
you try and pull back and you be as broad as you can.
link |
You talk to lots of people and it's like entering in any kind of a creative community.
link |
There's years that are needed and human connections are critical to it.
link |
So, you know, I think about, you know, being a musician or being an artist or something,
link |
you don't just, you know, immediately from day one, you know, you're a genius and therefore
link |
So, you know, practice really, really hard on basics and you be humble about where you
link |
are and then you realize you'll never be an expert on everything.
link |
So, you kind of pick and then there's a lot of randomness and a lot of kind of luck.
link |
But luck just kind of picks out which branch of the tree you go down, but you'll go down
link |
So yeah, it's a community.
link |
So the graduate school is a, I still think is one of the wonderful phenomena that we
link |
have in our, in our world.
link |
It's very much about apprenticeship with an advisor.
link |
It's very much about a group of people you belong to.
link |
It's a four or five year process.
link |
So it's plenty of time to start from kind of nothing to come up to something, you know,
link |
more expertise and then start to have your own creativity start to flower or even surprise
link |
into your own self.
link |
And it's a very cooperative endeavor.
link |
It's I think a lot of people think of science as highly competitive and I think in some
link |
other fields it might be more so.
link |
Here it's way more cooperative than you might imagine.
link |
And people are always teaching each other something and people are always more than
link |
happy to be clear that.
link |
So I feel I'm an expert on certain kinds of things, but I'm very much not expert on
link |
lots of other things and a lot of them are relevant and a lot of them are, I should know,
link |
but it should in some sense, you know, you don't.
link |
So I'm always willing to reveal my ignorance to people around me so they can teach me things.
link |
And I think a lot of us feel that way about our field.
link |
So it's very cooperative.
link |
I might add it's also very international because it's so cooperative.
link |
We see no barriers and so that the nationalism that you see, especially in the current era
link |
and everything is just at odds with the way that most of us think about what we're doing
link |
here where this is a human endeavor and we cooperate and are very much trying to do it
link |
together for the, you know, the benefit of everybody.
link |
So last question, where and how and why did you learn French and which language is more
link |
beautiful English or French?
link |
So first of all, I think Italian is actually more beautiful than French and English.
link |
And I also speak that.
link |
So I'm, I'm, I'm married to an Italian and I have kids and we speak Italian.
link |
Anyway, no, all kidding aside, every language allows you to express things a bit differently.
link |
And it is one of the great fun things to do in life is to explore those things.
link |
So in fact, when I kids or, you know, teens or college kids ask me, what is your study?
link |
I say, well, you know, do what your heart, where your heart is, certainly do a lot of
link |
math, math is good for everybody, but do some poetry and do some history and do some language
link |
You know, throughout your life, you'll want to be a thinking person, you'll want to have
link |
For me, yeah, French, I learned when I was, I'd say a late teen, I was living in the middle
link |
of the country in Kansas and not much was going on in Kansas with all due respect to Kansas.
link |
But, and so my parents happened to have some French books on the shelf and just in my boredom,
link |
I pulled them down and I found this is fun.
link |
And I kind of learned the language by reading.
link |
And when I first heard it spoken, I had no idea what was being spoken, but I realized
link |
I had somehow knew it from some previous life.
link |
And so I made the connection.
link |
But then, you know, I traveled and just I love to go beyond my own barriers and my own
link |
comfort or whatever, and I found myself in, you know, on trains in France next to say
link |
older people who had, you know, lived a whole life of their own and the ability to communicate
link |
with them was, was, was, you know, special and ability to also see myself in other people's
link |
shoes and have empathy and kind of work on that language as part of that.
link |
So, so after that kind of experience and also embedding myself in French culture, which
link |
is, you know, quite, quite amazing, you know, languages are rich, not just because there
link |
is something inherently beautiful about it, but it's all the creativity that went into
link |
So I learned a lot of songs, read poems, read books.
link |
And then I was here actually at MIT where we're doing the podcast today and a young
link |
professor, you know, not yet married and, you know, not having a lot of friends in the
link |
So I just didn't have, I was kind of a bored person.
link |
I said, I heard a lot of Italians around.
link |
There's happened to be a lot of Italians at MIT, a lot of Italian professors for some
link |
And so I was kind of vaguely understanding what they were talking about.
link |
I said, well, I should learn this language too.
link |
And then later met my spouse and, you know, Italian became a more important part of my
link |
But, but I go to China a lot these days.
link |
I go to Asia, I go to Europe and every time I go, I kind of am amazed by the richness
link |
of human experience and the people don't have any idea if you haven't traveled kind
link |
of how amazingly rich and I love the diversity.
link |
It's not just a buzzword to me.
link |
It really means something.
link |
I love the, you know, the, you know, embed myself with other people's experiences.
link |
And so, yeah, learning language is a big part of that.
link |
I think I've said in some interview at some point that if I had, you know, millions of
link |
dollars and infinite time whatever, what would you really work on if you really wanted to
link |
And for me, that is natural language and, and really done right, you know, deep understanding
link |
Um, that's to me an amazingly interesting scientific challenge and one we're very far
link |
away on one we're very far away, but good natural language people are kind of really
link |
I think a lot of them see that's where the core of AI is that if you understand that
link |
you really help human communication, you understand something about the human mind,
link |
the semantics that come out of the human mind and I agree, uh, I think that will be such
link |
So I didn't do that in my career just because I kind of, I was behind in the early days.
link |
I didn't kind of know enough of that stuff.
link |
I didn't learn much language, uh, and it was too late at some point to kind of spend a
link |
whole career doing that, but I admire that field and, um, and so my little way by learning
link |
language, um, you know, kind of, uh, that part of my brain is, um, it's been trained
link |
You're truly are the Miles Davis and machine learning.
link |
I don't think there's a better place than it was.
link |
Mike, it was a huge honor talking to you today.
link |
It's been my pleasure.
link |
Thanks for listening to this conversation with Michael, I, Jordan, and thank you to
link |
our presenting sponsor, cash app, download it, use code lex podcast.
link |
You get $10 and $10 will go to first an organization that inspires and educates young minds to
link |
become science and technology innovators of tomorrow.
link |
If you enjoy this podcast, subscribe on YouTube, give it five stars on Apple podcast, support
link |
on Patreon, or simply connect with me on Twitter at Lex Friedman.
link |
And now let me leave you with some words of wisdom from Michael, I, Jordan from his blog
link |
post titled artificial intelligence, the revolution hasn't happened yet calling for broadening
link |
the scope of the AI field.
link |
We should embrace the fact that what we are witnessing is the creation of a new branch
link |
The term engineering is often invoked in a narrow sense in academia and beyond with
link |
overtones of cold, affect less machinery and negative connotations of loss of control
link |
by humans, but an engineering discipline can be what we want it to be in the current
link |
era with a real opportunity to conceive of something historically new, a human centric
link |
engineering discipline, I will resist giving this emerging discipline a name, but if the
link |
acronym AI continues to be used, let's be aware of the very real limitations of this
link |
placeholder, let's broaden our scope, tone down the hype and recognize the serious challenges
link |
Thank you for listening and hope to see you next time.