Lex Fridman PodcastGary Marcus: Toward a Hybrid of Deep Learning and Symbolic AI | Lex Fridman Podcast #43
EVERY SPOKEN WORD
150 min read · 30,339 words- 0:00 – 15:00
The following is a…
- LFLex Fridman
The following is a conversation with Gary Marcus. He's a professor emeritus at NYU, founder of Robust.AI and Geometric Intelligence. The latter is a machine learning company that was acquired by Uber in 2016. He's the author of several books on natural and artificial intelligence, including his new book, Rebooting AI: Building Machines We Can Trust. Gary has been a critical voice highlighting the limits of deep learning and AI in general, and discussing the challenges before our AI community that must be solved in order to achieve artificial general intelligence. As I'm having these conversations, I try to find paths toward insight, towards new ideas. I try to have no ego in the process, it gets in the way. I'll often continuously try on several hats, several roles. One, for example is the role of a three-year-old who understands very little about anything and asks big what and why questions. The other might be a role of a devil's advocate who presents counterideas with a goal of arriving at greater understanding through debate. Hopefully, both are useful, interesting, and even entertaining at times. I ask for your patience as I learn to have better conversations. This is the Artificial Intelligence podcast. If you enjoy it, subscribe on YouTube, give it five stars on iTunes, support it on Patreon, or simply connect with me on Twitter @lexfridman, spelled F-R-I-D-M-A-N. And now here's my conversation with Gary Marcus. Do you think human civilization will one day have to face an AI-driven technological singularity that will, uh, in a societal way modify our place in the food chain of intelligent living beings on this planet?
- GMGary Marcus
I think our place in the food chain has already changed. So there are lots of things people used to do by hand that they do with machine. If you think of a singularity as like one single moment, which is I guess what it suggests, I don't know if it'll be like that. But I think that there's a lot of gradual change and AI is getting better and better. I mean, I'm here to tell you why I think it's not nearly as good as people think, but, you know, the overall trend is clear. Maybe, you know, maybe Ray Kurzweil thinks it's an exponential and I think it's linear. In some cases, it's close to zero right now, but it's all gonna happen. I mean, we are gonna get to human-level intelligence or whatever you want, w- what you will, um, artificial general intelligence at some point. And that's certainly gonna change our place in the food chain, 'cause a lot of the tedious things that we do now, we're gonna have machines do. And a lot of the dangerous things that we do now, we're gonna have machines do. And I think our whole lives are gonna change from, uh, people finding their meaning through their work through people finding their meaning through creative expression.
- LFLex Fridman
So the- the singularity will be a very gradual, in fact removing the meaning of the word singularity, it'll be a very gradual transformation in your view?
- GMGary Marcus
I- I- I think that it'll be somewhere in between, and I guess it depends what you mean by gradual and sudden. I don't think it's gonna be one day. I think it's important to realize that m- intelligence is a multi-dimensional variable, so you know, people sort of write this stuff as if like IQ was one number and, you know, the day that you hit 262 or whatever, you displace the human beings. And really, there's lots of facets to intelligence. So there's verbal intelligence and there's motor intelligence and the- there's mathematical intelligence and so forth. Um, machines in their mathematical intelligence far exceed most people already. In their ability to play games, they far exceed most people already. In their ability to understand language, they lag behind my five-year-old, far behind my five-year-old. So there are some facets of intelligence that machines have grasped and some that they haven't, and you know, we have a lot of work left to do to get them to say, understand natural language or to understand how to flexibly approach some, you know, kind of novel MacGyver pr- problem-solving kind of situation. And I- I don't know that all of these things will come at once. I think there are certain vital prerequisites that we're missing now. So for example, machines don't really have common sense now, so they don't understand that bottles contain water and that people drink water to quench their thirst and that they don't want to dehydrate. They- they don't know these basic facts about human beings, and I think that that's a rate-limiting step for many things. It's a rate-limiting step for reading, for example, because stories depend on things like oh my god, that per- person's running out of water, that's why they did this thing. Or you know, the- if they only h- they had water, they could put out the fire, whatever. So you know, you watch a movie and your knowledge about how things work matter, um, and so a computer can't understand that movie if it doesn't have that background knowledge. Same thing if you read a book. And so there are lots of places where if we had a good machine-interpretable set of common sense, many things would accelerate relatively quickly, but I don't think even that is like a single point. There's- there's many different aspects of knowledge and we might, for example, find that we make a lot of progress on physical reasoning, getting machines to understand, for example, how keys fit into locks or you know, that kind of stuff. Or- or I mean, h- how this gadget here works and- and uh, and so forth and so on. Machines might do that long before they do really good psychological reasoning, 'cause it's easier to get kind of labeled data or to do direct experimentation on, um, a microphone stand than it is to do direct experimentation on human beings to understand the levers that- that guide them.
- LFLex Fridman
That's- that's a really interesting point actually, uh, whether it's easier to- to gain common sense knowledge or psychological knowledge.
- GMGary Marcus
I would say that common sense knowledge includes both physical knowledge and psychological knowledge. The- the ph- and the argument I was making-
- LFLex Fridman
Oh, you said the physical versus psychological.
- GMGary Marcus
Yeah, physical versus psychological.
- LFLex Fridman
Got it.
- GMGary Marcus
The argument I was making is physical knowledge might be more accessible because you could have a robot, for example, lift a bottle, try putting a bottle cap on it, see that it, you know, falls off, if it does this, and see that it could turn it upside down. And so the robot could do some experimentation. Um, we do some of our psychological r- reasoning by looking at our own minds. So, I can sort of guess how you might react to something based on how I think I would react to it. And robots don't have that intuition, and they also can't do experiments on people in the same way, or we'll probably shut them down. So, you know, if we wanted to have robots figure out how I respond to pain by pinching me in different ways, like that's probably, you know, it's not gonna make it past the human subjects board and, you know, companies are gonna get sued or whatever. So, like there's certain kinds of practical experience that are limited or off-limits t- to robots. And-
- LFLex Fridman
That- that's a really interesting point. What is more, uh, difficult to gain a grounding in? Because to play devil's advocate, I would say that, uh, you know, human behavior is easier expressed in data, in digital form. And so when you look at Facebook algorithms, they- they get to observe human behavior.
- GMGary Marcus
Mm-hmm.
- LFLex Fridman
So you get to study and manipulate even human behavior in a way that you perhaps cannot study or manipulate the physical world. So, it's true why you said pain is, like physical pain, but that's, again, the physical world. Emotional pain might be much easier to experiment with, perhaps unethical, but nevertheless, some would argue it's already going on.
- GMGary Marcus
I- I think that you're right, for example, that Facebook does a lot of experimentation in- in psychological reasoning. In fact, Zuckerberg talked about AI at a talk that he gave at NIPS. I wasn't there, but the conference has been renamed NeurIPS, but it used to be called NIPS when he gave the talk. And he talked about Facebook basically having a gigantic theory of mind. So, I think it is certainly possible. I mean, Facebook does some of that. I think they have a really good idea of how to addict people to things. They understand what draws people back to things, and I think they exploit it in ways that I'm not very comfortable with. But even so, I think that there- there are only some slices of human experience that they can access through the kind of interface they have. And of course, they're doing all kinds of VR stuff, and maybe that'll change and they'll, um, expand their data and- and, you know, I'm- I'm sure that that's part of their goal. Um, so it is an interesting question.
- LFLex Fridman
I think love, fear, insecurity, all of the things that I would say some of the deepest things about human nature and the human mind could be explored through digital form. It's that... it's- you're actually the first person just now that brought up, I wonder what is more difficult. Because I think folks who are the slow... and we'll talk a lot about deep learning, but the people who are thinking beyond deep learning are thinking about the physical world. You're starting to think about robotics, in-the-home robotics. How do we make robots manipulate objects? Which requires an understanding of the physical world and it requires common sense reasoning. And that has felt to be like the next step for common sense reasoning. But you've now brought up the idea that there's also the emotional part, and it's interesting whether that's hard or easy.
- GMGary Marcus
I think some parts of it are and some aren't. So my company that I- I recently founded with Rod Brooks, um, you know, from MIT for many years and Roomba and so forth, um, we're interested in both. We're interested in physical reasoning and psychological reasoning, among many other things. And, uh, you know, there are pieces of each of these that are accessible. So if you want a robot to figure out whether it can fit under a table, that's a relatively accessible piece of physical reasoning. You know, if you know the height of the table and you know the height of the robot, it's not that hard. If you wanted to do physical reasoning about Jenga, it gets a little bit more complicated and you have to have, you know, higher resolution data in order to do it. Um, with psychological reasoning, it's not that hard to know, for example, that people have goals and they like to act on those goals, but it's really hard to know exactly what those goals are.
- LFLex Fridman
What about ideas of frustration? I mean, you could argue it's extremely difficult to understand the sources of human frustration as they're playing Jenga with you, or, or not.
- GMGary Marcus
Yeah, I mean-
- LFLex Fridman
You could argue that it is very accessible.
- GMGary Marcus
There- there's some things that are gonna be obvious and some not. So like, I don't think anybody really can do this well yet, but I think it's not inconceivable to imagine machines in the not so distant future being able to understand that if people lose in a game, that they don't like that.
- LFLex Fridman
Right.
- GMGary Marcus
Right? You know, that's not such a hard thing to program and it's pretty consistent across people. Most people don't enjoy losing, and so, you know, that makes it, you know, relatively easy to code. On the other hand, if you wanted to capture everything about frustration, well, people can get frustrated for a lot of different reasons. They might get sexually frustrated. They might get frustrated they didn't get their promotion at work. The- all kinds of different things. Um, and the more you expand the scope, the harder it is for anything like the existing techniques to really do that.
- LFLex Fridman
So I'm talking to Garry Kasparov next week, and he seemed pretty frustrated with his game against Deep Blue, so.
- GMGary Marcus
Yeah. Well, I'm frustrated with my game against him last year 'cause I- I played him. I had two excuses. I'll give you my excuses up front-
- LFLex Fridman
Okay.
- GMGary Marcus
... but it- it won't mitigate the outcome.
- LFLex Fridman
Okay.
- GMGary Marcus
Um, I was jet-lagged and I hadn't played in 25 or 30 years. Um, but the outcome is he completely destroyed me and it wasn't even close.
- LFLex Fridman
Have you ever been beaten in any board game by a machine?
- GMGary Marcus
I have. I- I actually beat, uh, or played the predecessor to Deep Blue, um, Deep Thought I believe it was called. Um, and that too crushed me. (laughs)
- 15:00 – 30:00
The- there's something to…
- LFLex Fridman
is still very constrained. When you say the possibility of the number of sentences that could come, it is huge, but it nevertheless is much more constrained, it feels, maybe I'm wrong, than the physi- the possibilities that the physical world brings us.
- GMGary Marcus
The- there's something to what you say in, in some ways in which I disagree. So, one interesting thing about language is that it abstracts away. This bottle, I don't know if the, it'll be in the field of view, is on this table. And I use the word on here, and I can use the word on here. Maybe not here, but there's all... That one word encompasses, you know, in analog space, a sort of infinite number of, of possibilities. So, there is a way in which language filters down the variation of the world. Um, and there's other ways. So, you know, we have a grammar and more or less, you have to f- follow the rules of that grammar. You can break them a little bit, but by and large, we follow the rules of grammar. And so that's a constraint on language. So, there are ways in which language is a constrained system. On the other hand, there are many arguments that say there's an infinite number of possible sentences, and you can establish that by just, you know, stacking them up. So, I think there's water on the table. You think that I think that there's water on the table. Your mother thinks that you think that I think the wa- that water is on the table. Your brother thinks that maybe your mom is wrong to think that you think that I think. Right? So we can, you know, we can make it in sentences of infinite length or we can stack up adjectives. This is a very silly example, a very, very silly example, a very, very, very, very, very, very silly example and so forth. So, so I mean, there are good arguments that there's an infinite range of sentences. In any case, it's, it, it's vast-
- LFLex Fridman
Mm-hmm.
- GMGary Marcus
... by any reasonable measure. And for example, almost anything in the physical world we can talk about in the language world. And interestingly, many of the sentences that we understand, we can only understand if we have a very rich model of the physical world. So, I don't ultimately want to adjudicate the debate that I think you just set up, but I, I, I find it interesting. Um, you know, maybe the physical world is even more complicated than language. I th- I think that's fair. But, um-
- LFLex Fridman
But you think that language is-
- GMGary Marcus
Language is really, really complicated.
- LFLex Fridman
... is hard.
- GMGary Marcus
It's really, really hard. Well, it's really, really hard for machines, for linguists, you know, people trying to understand it. It's not that hard for children, and that's part of what's driven my whole career, right? I was a student of Steven Pinker's and we were trying to figure out why kids could learn language when machines couldn't.
- LFLex Fridman
I think we're gonna get into language, we're gonna get into communication, intelligence and, and neural networks and so on, but let me return to the high level, uh, the futuristic for, for a brief moment. So, you've, uh, written in your book, in your new book, "It would be arrogant to suppose that we could forecast where AI will be or the impact it will have in 1,000 years or even 500 years." So, let me ask you to be arrogant.
- GMGary Marcus
(laughs)
- LFLex Fridman
... uh, what do AI systems with or without physical bodies look like 100 years from now? If you were to just, uh, you can't predict, but if you were to-
- GMGary Marcus
It's so hard.
- LFLex Fridman
... philosophize and imagine, do?
- GMGary Marcus
Can I, can I first justify the arrogance before you try to push me beyond it?
- LFLex Fridman
Sure.
- GMGary Marcus
I mean, there are examples, like, you know, people figured out how electricity worked. They had no idea that that was gonna lead to cell phones, right? I mean, things can move awfully fast once new technologies are perfected. Even when they made transistors, they weren't really thinking that cell phones would lead to social networking.
- LFLex Fridman
There are, nevertheless, predictions of the future which are statistically unlikely to come to, to be, but nevertheless, is the-
- GMGary Marcus
(laughs) You're asking me to be wrong.
- LFLex Fridman
(laughs) Asking you to be statistic-
- GMGary Marcus
In which way would I like to be wrong?
- LFLex Fridman
Pick the least unlikely to be wrong thing, (laughs) even though it's most very likely to be wrong.
- GMGary Marcus
I mean, here are some things that we can safely predict, I suppose.
- LFLex Fridman
Sure.
- GMGary Marcus
We can predict that AI will be faster than it is now. It will be cheaper than, than it is now. It will be better in the sense of being more general and applicable in, in more places. It will be pervasive. You know, I mean, these are e- easy predictions that I, I'm sort of modeling them in my head on, um, Jeff Bezos's famous predictions. He says, "I can't predict the future, not in every way," I'm paraphrasing, but, "I can predict that people will never wanna pay more money for their stuff, they're never gonna want it to take longer to get there." And you know, so like, you can't predict everything, but you can predict some things. Sure, of course it's gonna be faster and better and w- what we can't really predict is the full scope of, of where AI will be in a certain period. I mean, I think it's safe to say that although I'm very skeptical about current AI, that it's possible to do much better. You know, there's no in-principled argument that says AI is an insolvable problem, that there's magic inside our brains that will never be captured. I mean, I've heard people make l- those kind of arguments. I don't think they're very good. So AI is gonna come, and probably 500 years is plenty to get there. And then once it's here, it really will change everything.
- LFLex Fridman
So when you say AI is gonna come, are, are you talking about human level intelligence? So maybe-
- GMGary Marcus
I, I, I like the term general intelligence. So I don't think that the ultimate AI, if there is such a thing, is gonna look just like humans. I think it's gonna do some things that humans do better than current machines, like reason flexibly, um, and understand language and so forth. But that doesn't mean they have to be identical to humans. So for example, humans have terrible memory and they suffer from what some people call motivated reasoning. So, they like arguments that seem to support them and they, they dismiss arguments that they don't like. There's no reason that a machine should ever do that.
- LFLex Fridman
Uh, so you see the, those, the limitations of memory as a bug, not a feature?
- GMGary Marcus
Absolutely. I'll say two things about that. One is, I was on a panel with Danny Kahneman, the Nobel Prize winner, last night, and we were talking about this stuff. And I think, you know, what we converged on is that the humans are a low bar to exceed. They may be outside of our skill right now, but as- as, you know, AI programmers, but eventually AI will exceed it. So, we're not talking about human level AI, we're talking about general intelligence that can do all kinds of different things and do it without some of the flaws that human beings have. The other thing I'll say is I wrote a whole book actually about the flaws of humans. It's, it's actually a nice bookend to the, um, or counterpoint to the current book. So I wrote a book called Kluge which was about the limits of the human mind. The current book is kind of about those few things that humans do a lot better than machines.
- LFLex Fridman
Do you think it's possible that the flaws of the human mind, the limits in memory, our mortality, our bias is a strength not a weakness, that that is the thing that enables from which motivation springs and meaning springs, or no?
- GMGary Marcus
I've heard a lot of arguments like this. I've never found them that convincing. I think that there's a lot of making lemonade out of lemons.
- 30:00 – 45:00
So the, the idea…
- GMGary Marcus
fall into at least-... a little bit better place. There's a- right now, you're, like, learning correlations between pixels when you play a video game or something like that, and it doesn't work very well. It works when the video game is just the way that you studied it, and then you alter the video game in small ways, like you move the paddle in Breakout a few pixels, and the system falls apart 'cause it doesn't understand. It doesn't have a representation of a paddle, a ball, a wall, a set of bricks, and so forth. And so it's reasoning at the wr- wrong level.
- LFLex Fridman
So the, the idea of common sense, it's full of mystery. You've worked on it, but it's nevertheless full of mystery, uh, full of promise. What is, what does common sense mean? What does knowledge mean? So the way you've been discussing it now is very intuitive. It makes a lot of sense that that is something we should have and that's something deep learning systems don't have. But the argument could be that we're oversimplifying it because, um, we're oversimplifying the notion of common sense because that's how we f- we, it feels like we as humans at the cognitive level approach problems. So maybe-
- GMGary Marcus
So a lot of people aren't actually gonna read my book, but if they did read the book, one of the things that might come as a surprise to them is that we actually say common sense is really hard and really complicated. So they would prob- you know, m- my critics know that I like common sense, but, you know, that chapter actually starts by us beating up not on deep learning, but kind of on our own home team as it will. So Ernie and I are first and foremost people that believe in at least some of what good old-fashioned AI tried to do. So we believe in symbols and logic and programming. Um, things like that are important. And we go through why even those tools that we hold fairly dear aren't really enough. So we talk about why common sense is actually many things, and some of them fit really well with those classical sets of tools. So things like taxonomy. So I know that a bottle is an object or it's a vessel, let's say, and I know a vessel is an object and objects are material things in the physical world. So, like, I can make some inferences. If I know that vessels need to, you know, h- not have holes in them, then I can infer that in order, you know, to carry their contents, then I can infer that a bottle shouldn't have a hole in it in order to carry its contents. So you can do hierarchical inference and so forth. And we say that's great, but it's only a tiny piece of what you need for common sense, and we give lots of examples that don't fit into that. So, um, another one that we talk about is a cheese grater. You got holes in a cheese grater. You got a handle on top. You can build a model in the game engine sense of a model so that you could have a little cartoon character flying around through the holes of the grater, but we don't have a system yet, taxonomy doesn't help us that much, that really understands why the handle is on top and what you do with the handle, or why all those circles are sharp, or how you'd hold the cheese with respect to the grater in order to make it actually work.
- LFLex Fridman
Do you think these ideas are just abstractions that could emerge on a system like a very large deep neural network?
- GMGary Marcus
I'm a skeptic that that kind of emergence per se can work. So I think that deep learning might play a role in the systems that do what I want systems to do, but it won't do it by itself. I- I've never seen a deep learning system really extract an abstract concept. What they do, principled reasons for that m- stemming from how back propagation works, how the architectures are set up. One example is deep learning people actually all build in something like conv- build in something called convolution, which Yann LeCun is famous for, which is an abstraction. They don't have their systems learn this. So the abstraction is an object looks the same if it appears in different places. And what LeCun figured out and why, you know, essentially why he was a co-winner of the Turing Award was that if you programmed this in innately, then your system would be a whole lot more efficient. In principle, this should be learnable, but people don't have systems that kind of reify things that make them more abstract. And so you'd, what you'd really wind up with if you don't program that in advance is a system that kind of realizes that this is the same thing as this, but then I take your little clock there and I move it over and it doesn't realize that the same thing applies to the clock.
- LFLex Fridman
So the really nice thing, you're right, that convolution is just one of the things that's, like, it's an innate feature that's programmed by the human expert. But-
- GMGary Marcus
We need more of those, not less.
- LFLex Fridman
Yes. Yes. So the, but the nice feature is it feels like that requires coming up with that brilliant idea, can get you a Turing Award, but it requires less effort than encoding something we'll talk about, the expert system, so encoding a lot of knowledge by hand. So it feels like one, there's a huge amount of limitations which you clearly outline with deep learning, but the nice feature of deep learning, whatever it is able to accomplish, it does it, it does a lot of stuff automatically without human intervention, which is-
- GMGary Marcus
Well, and that's part of why people love it, right?
- LFLex Fridman
Right.
- GMGary Marcus
But I always think of this quote from, uh, Bertrand Russell, which is, "It ha- has all the adv- advantages of theft over honest toil." It's really hard to program into a machine a notion of causality or, you know, even how a bottle works or what containers are. Ernie Davis and I wrote a, I don't know, 45-page academic paper trying just to understand what a container is, which I don't think anybody ever read the paper. But it's a very detailed analysis of all the things, well, not even all, some of the things you need to do in order to understand a container. It would be a whole lot nice, and, you know, I'm- I'm a co-author on the paper, I made it a little bit better. Ernie did the hard work for that particular paper. And it took him, like, three months to get the logical statements correct. And maybe that's not the right way to do it, um, it's a way to do it, but on that way of doing it, i- it's really hard work to do something as simple as understanding containers. And n- nobody wants to do that hard work. Even Ernie didn't want to do that hard work.Everybody would rather just, like, feed their system in with a bunch of videos with a bunch of containers and have the systems infer how contain- containers work. It would be, like, so much less effort. Let the machine do the work. And so I understand the impulse. I understand why people want to do that. I just don't think that it works. I've never seen anybody build a system that, in a robust way, can actually m- watch videos and predict exactly, you know, which containers would leak and which ones wouldn't or something l- like. Um, and I, I know someone's gonna go out and do that since I said it, and, uh, you know, I look forward to seeing it.
- LFLex Fridman
(laughs)
- GMGary Marcus
Um, but getting these things to work robustly is really, really hard. So, Yann LeCun, who was my colleague at NYU for m- many years, thinks that the hard work should go into d- finding an unsupervised learning algorithm-
- LFLex Fridman
Yes.
- GMGary Marcus
... that will watch videos, use the next frame, basically, in order to tell it what's going on, and he thinks that's the royal road and he's willing to put in the work in devising that algorithm. Then he wants the machine to do the rest. And I, again, I understand the impulse. My intuition, based on years of watching this stuff and making predictions 20 years ago that still hold even though there's a lot more computation and so forth, is that we actually have to do a different kind of hard work, which is more like building a design specification for what we want the system to do, doing hard engineering work to figure out how we do things like what Yann did, um, for convolution in order to figure out how to encode complex knowledge into the systems. The current systems don't have that much knowledge other than convolution, um, which is, again, this, you know, objects being s- in different places, and having the same per- perception, I guess I'll say, same appearance. People don't wanna do that work. They don't see how to naturally fit one with the other.
- LFLex Fridman
I think that's ... Yes, absolutely. But also on the expert system side, there's a temptation to go too far the other way. So it was just having an expert sort of sit down and encode the description, the framework for what a container is, and then having the system reason the rest. From my view, like, o- one really exciting possibility is of active learning where it's continuous interaction between a human and machine as the machine, there's kind of deep learning type extraction with information from data, patterns, and so on, but humans also guiding the, the learning procedures, guiding the, the, both the, the process and the framework of how the m- machine learns whatever the task is.
- GMGary Marcus
I, I, I was with you with almost everything you said except the phrase deep learning.
- LFLex Fridman
(laughs)
- GMGary Marcus
Um, what I think you really want there is a new form of machine learning.
- LFLex Fridman
Mm-hmm.
- GMGary Marcus
So let's remember, deep learning is a particular way of doing machine learning. Most often it's done with supervised data for perceptual categories. There are other things you can do with deep learning, and some of them quite technical, but the standard use of deep learning is I have a lot of examples and I have labels for them. So here are pictures. This one's the Eiffel Tower. This one's the Sears tower. This one's the Empire State Building. This one's a cat. This one's a pig and so forth. And you just get, you know, millions of examples and millions of labels, and deep learning is extremely good at that. Um, it's better than any other solution that anybody's, has devised. But it is not good at representing abstract knowledge. It's not good at representing things like bottles contain liquid and, you know, have tops to them and so forth. I- it's not very good at learning or representing that kind of knowledge. It is an example of having a machine learn something-
- LFLex Fridman
Right.
- GMGary Marcus
... but it's a machine that learns a particular kind of thing, which is object classification. It's not a particularly good algorithm for learning about the abstractions that govern our world. There may be such a thing. Part of what we counsel in the book is maybe people should be working on devising such things.
- LFLex Fridman
So o- one possibility, just, uh, I wonder what you think about it, is so d- deep neural networks do form abstractions, but they're not a- accessible to us humans in terms of we can't-
- GMGary Marcus
There's some truth in that.
- LFLex Fridman
So is it possible that either current or future neural networks form very high level abstractions which are as powerful as, as our human abstractions of common sense, we just can't get ahold of them and so the problem is essentially, well, we need to make them explainable
- GMGary Marcus
This is an astute question, but I think the answer is at least partly no. One of the kinds of classical neural network architectures is what we call an autoassociator. It just tries to take an input, um, goes through a set of hidden layers, and comes out with an output. And it's supposed to learn essentially the identity function, that your input is the same as your output. So you think of it as binary numbers. You've got, like, the one, the two, the four, the eight, the 16, and so forth. Um, and so if you wanna input 24, you turn on the 16, you turn on the eight, and it's like binary one one and a bunch of zeros. Um, so I did some experiments in 1998 with, with the proto- uh, sorry, the precursors of, um, contemporary deep learning, and what I showed was you could train these networks on all the even numbers and they would never generalize to the odd number. A lot of people thought that I was, I don't know, an idiot or faking the experiment or it wasn't true or whatever, but it is true that with this class of networks th- that we had i- in that day, that they would never ever make this generalization. And it's not that the networks were stupid, it's that they see the world in a different way than we do. They were basically concerned, what is the probability that the right most output node is going to be a one? And as far as they were concerned, in everything they'd ever been trained on, it was a zero. Um, th- ne- that node had never been turned on, and so they figured, "Why turn it on now?" Whereas a person would look at the same problem and say, "Well, it's obvious. We're just doing the thing that corresponds." The Latin for it is mutatis mutandis. We'll change what needs to be changed. And we do this ... This is what algebra is.
- LFLex Fridman
Right.
- GMGary Marcus
So I can do F of X equals Y plus two, and I can do it for a couple of values. I can tell you if Y is three then X is five, and if Y is four, X is six. And now I can do it with some totally different number, like a million, and then you can say, "Well, obviously it's a million and two," because you have an algebraic operation that you're applying to a variable. And deep learning systems kind of emulate that, but they don't actually do it.... the particular example, you could, mm, uh, fudge a solution to that particular problem, but the general form of that problem remains, that what they learn is really correlations between different input and output nodes, and they're complex correlations where... with multiple nodes i- i- involved and so forth. But they're... ultimately, they're correlative. They're not structured over these operations over variables. Now someday, people may do a new form of deep learning that incorporates that stuff, and I think it will help a lot. And there's some tentative work on things like differentiable programming right now that fall into that category. Um, but the sort of classic stuff, like people use for ImageNet, doe- doesn't have it, and you have people like Hinton going around saying, "Symbol manipulation," like what Marcus, what I, um, advocate, "is like the gasoline engine. It's obsolete. We should just use this cool electric s- power that we've got with the deep learning." And that's really destructive 'cause we really do need to have the gasoline engine stuff that represents, um... I mean, I, I don't think it's a good analogy, but-
- LFLex Fridman
(laughs) .
- 45:00 – 1:00:00
Yes. …
- GMGary Marcus
contrasting them, and there is a contrast, but that's not the thing that I'm endorsing.
- LFLex Fridman
Yes.
- GMGary Marcus
So, expert systems try to capture things like medical knowledge with a large set of rules. So, if the patient has this symptom and this other symptom, then it is likely that they have this disease. Um, so there are logical rules, and they were symbol manipulating rules of just the sort that I'm talking about. Um, and the pro-
- LFLex Fridman
They encode a set of knowledge that the experts then put in, and then-
- GMGary Marcus
And very explicitly so. So, you, you'd have somebody interview an expert and then try to turn that stuff into rules.
- LFLex Fridman
Yeah.
- GMGary Marcus
And at some level, I'm arguing for rules, but the difference is, w- those guys did in the '80s was almost entirely rules, almost entirely handwritten with no machine learning. What a lot of people are doing now is almost entirely one species of machine learning with no rules. And what I'm counseling is actually a hybrid. I'm saying that both of these things have their advantage. So, if you're talking about perceptual classification, "How do I recognize a bottle?" deep learning is the best tool we've got right now. If you're talking about making inferences about what a bottle does, something closer to the expert systems is probably still, um, the best available alternative, and probably we want something that is better able to handle quantitative and statistical information than those classical systems typically were. So, we need new technologies that are gonna draw some of the strengths of both the expert systems and the deep learning, but are gonna find new ways to synthesize them.
- LFLex Fridman
How, how hard do you think it is to add knowledge at the low level? So, mine human intellects to add extra information to, uh, s- symbol manipulating systems?
- GMGary Marcus
In some domains, it's not that hard, but it's often really hard, partly because a lot of the things that are important, people wouldn't bother to tell you. So, if you pay someone on Amazon Mechanical Turk to tell you stuff about bottles, they probably won't even bother to tell you some of the basic level stuff that's just so obvious to a human being and yet so hard to capture in machines. Th- you know, they, they're gonna tell you more exotic things. And, like, they're all well and good, but they're not getting to the, the, the root of the problem. So, untutored humans aren't very good at knowing, and why should they be, what kind of knowledge the computer m- system developers actually need. I, I don't think that that's an irremediable problem. I think it's historically been a problem. People have had crowdsourcing efforts, and, and they don't work that well. There's one at MIT, we're recording this at MIT, called Virtual Home, where, and we talk about this in the book, find the exact example there, but, um, people were asked to do things like describe an exercise routine. And the things that the people describe are at a very low level and don't really capture what's going on. So they're like, "Go to the room with the television and the, and the weights. Turn on the television. Press th- or press the remote to turn on the television. Lift weight, put weight down," or whatever. It's, like, very micro-level, and it's not telling you what an exercise routine is really about, which is, like, "I wanna fit a certain number of exercises in a certain time period. I wanna emphasize these muscles." I mean-You, you want some kind of abstract description. The fact that you happened to press the remote control in this room when you, you know, watched this television isn't really the essence of, um, the exercise routine, but if you just ask people like what did they do, then they, they give you this fine grain. And so it takes a level of expertise, um, about how the AI works in order to craft the right kind of knowledge.
- LFLex Fridman
So there's this ocean of knowledge that we all operate on. Some of it may not even be conscious, or at least we're not able to communicate it effectively.
- GMGary Marcus
Yeah, most of it we would recognize if somebody said it, if it was true or not-
- LFLex Fridman
Right.
- GMGary Marcus
... but we wouldn't think to say that it's true or not.
- LFLex Fridman
It's a really interesting, uh, mathematical property. This ocean has the property that every piece of knowledge in it, we would recognize it as true if it w- we're told, but we're unlikely to retrieve it, uh, in the reverse. So that-
- GMGary Marcus
Exactly.
- LFLex Fridman
... that interesting property, I would say there's a huge ocean of that knowledge.
- GMGary Marcus
Yeah.
- LFLex Fridman
What's your intuition? Is it accessible to AI systems somehow? Can we ... so you said-
- GMGary Marcus
Not yet. I mean mo- most of it is not ... well, I'll give you an asterisk on this in a second, but most of it has not ever been encoded in machine interpretable form. And so, I mean if, if you say accessible, there's two meanings of that. One is like could you build it into a machine? Yes. The other is like is there some da- database that we could go-
- LFLex Fridman
Right.
- GMGary Marcus
... you know, um, download and stick into our machine? No.
- LFLex Fridman
But the first thing, could, could we? Is, what's your intuition?
- GMGary Marcus
I think we could. I, I, I think it hasn't been done right. You know, the closest, and this is the asterisk, is the C-Y-C, CYC system tried to do this. A lot of logicians worked for Doug Lenat for 30 years on this project. I think they stuck too closely to logic, didn't represent enough about probabilities, tried to hand code it. There are various issues, and it, it hasn't been that successful. That is the closest existing system to trying to encode this.
- LFLex Fridman
Why do you think there's not more excitement/money behind this idea currently?
- GMGary Marcus
There was. People view that project as a failure. I think that they confused the failure of a specific instance that was conceived 30 years ago for the failure of an approach, which they don't do for deep learning.
- LFLex Fridman
Yeah.
- GMGary Marcus
So, you know, uh, in 2010, people had the same attitude towards deep learning. They were like, "This stuff doesn't really work and, you know, all these other algorithms work better," and so forth, and then certain key technical advances were made. But mo- mostly, it was the advent of graphics processing units that, that changed that. It wasn't even anything foundational in, in the techniques, though there were some new tricks. But, um, mostly it was just more compute and w- and more data, things like ImageNet that didn't exist before, um, that allowed deep learning. And it could be, to work, it could be that, you know, CYC just needs a few more things or something like CYC, but the widespread view is that that just doesn't work and people are, are reasoning from a single example. They, they don't do that with deep learning. They don't say, "Nothing that existed in 2010," and there were many, many efforts in, in deep learning, "was really worth anything." Right? I mean, really, there's no model from 2010 in deep learning, the pred- predecessors to deep learning, that has any commercial value whatsoever at this point, right? They're, they're a- all failures. Um, but that doesn't mean that there wasn't anything there. I have a friend, um, wh- I was getting to know him and he sa- uh, he said, "I h- I, I had a company too." I was talking about I had a new company. And he said, "I had a company too and it, and it failed." And I s- I said, "Well, what did you do?" And he said, "Deep learning," and the problem was he did it in 1986 or something like that and we didn't have the tools then, or 1990. We didn't have the tools then, not the algorithms. You know, his algorithms weren't that different from modern algorithms, but he didn't have the GPUs to run it fast enough, he didn't have the data, and so it failed. Um, it could be that, you know, symbol manipulation per se, with modern amounts of data and c- compute and maybe some advance in compute for that kind of compute, um, might be great. My, my perspective on it is not that we want to resuscitate that stuff per se, but we want to borrow lessons from it and bring it together with other things that we've learned.
- LFLex Fridman
And it might have an ImageNet moment where it would spark the world's imagination and there would be an explosion of s- symbol manipulation, uh, efforts.
- GMGary Marcus
Yeah, I think the people at AI2, the Paul Allen's, uh, AI Institute, are trying to do that. They're trying to build datasets that ... well, they're not doing it for quite the reason that you said, but th- they're trying to build datasets that at least spark interest in common sense reasoning.
- LFLex Fridman
To create benchmarks that-
- 1:00:00 – 1:15:00
Are they disjoint in…
- GMGary Marcus
be the same. Um, but I would say that, you know, it- it's a pretty interesting set of things that we are equipped with that allows us to do a lot of interesting things. So, I would argue or guess, based on my reading of the developmental psychology literature, which I've also participated in, that children are born with a notion of space, time, other agents, places.... and also this kind of mental algebra that I was describing before. No certain of causation, if I didn't just say that. So at least those kinds of things. They're, they're like frameworks for learning the other things, so-
- LFLex Fridman
Are they disjoint in your view or is it just somehow all connected? You've talked a lot about language, is it, is it all k- kind of connected in some mesh that's language-like of understanding concepts altogether or-
- GMGary Marcus
I don't think we know for people how they're represented, and machines just don't really do this yet. Um, so I think it's an interesting open question both for science and for engineering. Um, some of it has to be at least interrelated in the way that like the interfaces of a software package have to be able to talk to one another. So you know-
- LFLex Fridman
Right.
- GMGary Marcus
... the, the, the systems that represent space and time can't be totally disjoint because a lot of the things that we reason about are relations between space and time and cause. So you know, I put this on and I have expectations about what's gonna happen with the bottle cap on, on top of the bottle, um, and those span space and time, you know. If, if, if the cap is over here, I get a different outcome. If, um, the timing is different, if I put this here after I move that, then you know, I get a different outcome, um, that relates to causality. So obviously these mechanisms, whatever they are, um, can certainly communicate with each other.
- LFLex Fridman
S- so I think evolution had a significant role to play in the development of this whole kludge, right? How efficient do you think is evolution?
- GMGary Marcus
Oh, it's terribly inefficient, except that-
- LFLex Fridman
No, g- well, can we do better? (laughs)
- GMGary Marcus
(laughs) Well, let's go, I'll come to that in a second.
- LFLex Fridman
Sure.
- GMGary Marcus
It's inefficient except that once it gets a good idea, it runs with it.
- LFLex Fridman
Huh.
- GMGary Marcus
So it took, m- um, I guess a billion years if I've been... roughly a billion years to evolve to a vertebrate brain plan. Once that vertebrate plane- plan evolved, it spread everywhere. So fish have it and dogs have it and we have it. We have adaptations of it and specializations of it but, um... and the same thing with the primate bra- brain plan. So monkeys have it and apes have it and we have it so, you know, there are additional innovations like color vision and those spread really rapidly. So takes evolution a long time to get a good idea but, um, and, you know, being anthropomorphic and not literal here.
- LFLex Fridman
Right.
- GMGary Marcus
But once it has that idea, so to speak, which cashes out into once a set of genes are in the genome, those genes spread very rapidly and they're like subroutines or libraries, I guess is the word people might use nowadays or be more familiar with. They're libraries that can get used over and over again.
- LFLex Fridman
Yeah.
- GMGary Marcus
So once you have the library for building something with multiple digits, you can use it for a hand but you can also use it for a foot and you just kind of reuse the library with slightly different parameters. Evolution does a lot of that which means that the speed over time picks up, so evolution can happen faster because you have bigger and bigger libraries. And what I think has happened in attempts at evolutionary computation is that people start with libraries that are very, very minimal, like almost nothing and then, you know, progress is slow and it's hard for someone to get a good PhD thesis out of it and they give up. Um, if we had richer libraries to begin with, if you were evolving from systems that hadn't a rich innate structure to begin with, then things might speed up.
- LFLex Fridman
Or more PhD students. If the evolution process is indeed in a meta way, runs away with good ideas, you need to have a lot of ideas, pool of ideas in order for it to discover one that you can run away with and PhD students representing individual ideas as well.
- GMGary Marcus
Yeah, I mean, you could throw a billion PhD students at it.
- LFLex Fridman
Yeah. The monkeys are typewriters with Shakespeare, yep.
- GMGary Marcus
Well, but, b- I mean, those aren't cumulative, right? That's just random and part of the point that I'm making is that evolution is cumulative. So if, if, if, if you have-
- LFLex Fridman
(laughs)
- GMGary Marcus
... a billion monkeys independently-
- LFLex Fridman
Yeah.
- GMGary Marcus
... you don't really get anywhere but if you have a billion monkeys, and I think Dawkins made this point originally or probably other people, but Dawkins made it very nice in either Selfish Gene or Blind Watchmaker, um, if- if there is some sort of fitness function that can drive you towards something, um, I guess that's Dawkins' point and my point which is a variation on that is that if the evolution is cumulative, um, they're related points, um, then you can start going faster.
- LFLex Fridman
Do you think something like the process of evolution is required to build intelligent systems? So if we-
- GMGary Marcus
Not logically. So all the stuff that evolution did, a good engineer might be able to do. So for example, evolution made quadrupeds which distribute the load across, um, a horizontal surface. A good engineer could come up with that idea. Um, I mean, sometimes good engineers come up with ideas by looking at biology. There's lots of ways to, to get your ideas. Um, part of what I'm suggesting is we should look at biology a lot more. We should look at the biology of thought and understanding and, you know, the biology by which creatures intuitively reason about physics or other agents or like how do dogs reason about people? Like they're actually pretty good at it. If we could understand, um, we... at my college we joked, "dognition," if we could understand dognition well and how it was implemented, that might help us with our AI.
- LFLex Fridman
(inhales) So do you think, do you think it's possible that the kind of time scale that evolution took is the kind of time scale that will be needed to build intelligent systems or can we significantly accelerate that process in- inside a computer?
- GMGary Marcus
I mean, I think the way that we accelerate that process is we borrow from biology. Not slavishly but I think we look at biolo- how biology has solved problems and we say, "Does that inspire any engineering solutions here?" Um-
- LFLex Fridman
Try to mimic biological systems and then therefore have a shortcut.
- 1:15:00 – 1:15:39
Section 6
- GMGary Marcus
hold. I- in the world of computer science, that's amazing, right? Because, you know, there's a thousand or a million times more memory and, you know, computations a million times, m- m- you do a million times more operations per second, you know, spread across a cluster. And there's been advances in, you know, repl- replacing sigmoids with, um, uh, with, um, other functions and- and so forth. There's all kinds of advances, but the fundamental architecture hasn't changed and the fundamental limit hasn't changed. And what I said then is- is kind of still true. Then here's a second example. I r- recently had a piece in Wired that's adapted from the book, and the- the book didn't, w- was, went to press before GPT-2 came out.
Episode duration: 1:25:00
Install uListen for AI-powered chat & search across the full episode — Get Full Transcript
Transcript of episode vNOTDn3D_RI
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome