Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4

1. 0:00 – 1:22

   Biological vs artificial neural nets: the mystery of long-term credit assignment

   1. LFLex Fridman0:00

      What difference between biological neural networks and artificial neural networks is most mysterious, captivating, and profound for you?

   2. YBYoshua Bengio0:11

      First of all, there is so much we don't know about biological neural networks.

   3. LFLex Fridman0:15

      Right.

   4. YBYoshua Bengio0:15

      And that's very mysterious and captivating, because maybe it holds the key to improving artificial neural networks. One of the things I studied recently, uh, something that we don't know how biological neural networks do, but would be really useful for artificial ones, is the ability to do credit assignment through very long time spans. There are things that we can, in principle, do with artificial neural nets, but it's not very convenient, and it's not biologically plausible. And this mismatch, I think, this kind of mismatch may be an interesting thing to study to, A, understand better how brains might do these things, 'cause we don't have good corresponding theories with artificial neural nets, and B, maybe provide new ideas that we could explore about, um, things that brain do differently, and that we could incorporate in artificial neural nets.

2. 1:22 – 2:36

   What “credit assignment” means: episodic memory, causality, and revising past beliefs

   1. LFLex Fridman1:22

      So, let's break credit assignment up a little bit.

   2. YBYoshua Bengio1:24

      Yes.

   3. LFLex Fridman1:24

      So, what ... It's a beautifully technical term, but it can incorporate so many things. So, is it more on the RNN memory side, that, thinking like that? Or is it something about knowledge, building up common sense knowledge over time? Or is it, uh, more in the reinforcement learning sense, that you're picking up rewards over time for a particular, uh, to achieve a certain kind of goal?

   4. YBYoshua Bengio1:50

      So, I was thinking more about the first two meanings, whereby we store all kinds of memories, um, episodic memories in our brain, which we can access later in order to help us both infer causes of things that we are observing now, and assign credit to decisions or interpretations we came up with a while ago when, you know, those memories were stored. And then we can change the way we would have, uh, reacted or interpreted things in the past, and now that's credit assignment used for learning.

3. 2:36 – 4:03

   Why RNNs/LSTMs hit a wall: sequence length, forgetting, and consciousness-like selection

   1. LFLex Fridman2:36

      So, in which way do you think artificial neural networks, the current LSTM, the current architectures are not able to capture the ... Presumably, you're, you're, you're thinking of very long term.

   2. YBYoshua Bengio2:53

      Yes. So, current, recurrent nets are doing a fairly good jobs for sequences with dozens, or say, hundreds of time stamps, and then it gets sort of harder and harder, and depending on what you have to remember and so on, as you consider longer durations. Whereas humans seem to be able to do credit assignment through essentially arbitrary times. Like, I, I could remember something I did last year, and then now, because I see some new evidence, I'm gonna change my mind about, uh, the way I was thinking last year, and hopefully not do the same mistake again.

   3. LFLex Fridman3:33

      I think a big part of that is probably forgetting. You're only remembering the really important things, so it's very efficient forgetting. Um-

   4. YBYoshua Bengio3:42

      Uh, yes. So, there's a selection of what we remember, and I think, uh, there are really cool connection to higher level cognition here regarding consciousness, deciding and, and emotions. Like sort of the deciding what comes to consciousness and what gets stored in memory, wh- which, which are not trivial either.

4. 4:03 – 6:44

   What’s missing in deep nets: robust abstraction, causal explanation, and grounding language in the world

   1. LFLex Fridman4:03

      So, you've been at the forefront there all along, showing some of the amazing things that neural networks, deep neural networks can do in the field of artificial intelligence, and just broadly in all kinds of applications. But, uh, we can talk about that forever, but what in your view, because we're thinking towards the future, is the weakest aspect of the way deep neural networks represent the world? What is the, what is, in your view, is missing currently?

   2. YBYoshua Bengio4:33

      So, current, current state-of-the-art neural nets trained on large quantities of images or texts have some level of understanding of, you know, what explains those data sets, but it's very basic. It's, it's very low level, and it's not nearly as robust and abstract and general as our understanding. Okay, so that doesn't tell us how to fix things, but I think it encourages us to think about how we can, uh, maybe train our neural nets differently, so that they would focus, for example, on causal explanation. Something that we don't do currently with neural net training. Also, one thing I'll talk about in my talk this afternoon-

   3. LFLex Fridman5:32

      Mm-hmm.

   4. YBYoshua Bengio5:33

      ... is instead of learning separately from images and videos on one hand, and from text on the other hand, we need to do a better job of, um, jointly learning about language, and about the world to which it refers, so that, you know, both sides can help each other. We need to have good world models in, in our neural nets.... for them to really understand sentences which talk about what's going on in the world, and I think we need language input to help provide clues about what high-level concepts, like semantic concepts, should be represented at the top levels of these neural nets. In fact, there is evidence that the purely unsupervised learning of representations doesn't give rise to high-level representations that are as powerful as the ones we are getting from supervised learning.

   5. LFLex Fridman6:37

      Mm-hmm.

   6. YBYoshua Bengio6:38

      And so the- the- the clues we're getting just with the labels, not even sentences, is already very powerful.

5. 6:44 – 9:20

   Not architecture or dataset—training objectives and active agents that intervene

   1. LFLex Fridman6:44

      Do you think that's an architecture challenge or is it a dataset challenge?

   2. YBYoshua Bengio6:49

      Neither.

   3. LFLex Fridman6:50

      (laughs)

   4. YBYoshua Bengio6:50

      (laughs)

   5. LFLex Fridman6:54

      Uh, I'm tempted to just end it there. (laughs)

   6. YBYoshua Bengio6:57

      No, I guess-

   7. LFLex Fridman6:57

      But, uh, can you elaborate slightly? (laughs)

   8. YBYoshua Bengio6:59

      Yes, yes. (laughs) Ah, of course, datasets and architectures are- are something you wanna always play with, but- but I think the crucial thing is more the training objectives, the training frameworks. Um, for example, going from, um, passive observation of data to more active agents, which, um, uh, learn by intervening in the world, uh, the relationships between causes and effects, um, the sort of objective functions which could be important to allow the- the- the- the highest level, uh, explanations to- to- to rise from- from the learning, which I don't think we have now, the kinds of objective functions which could be used to, uh, reward exploration, the right kind of exploration. So these kinds of questions are neither in the dataset nor in the architecture, but more in how we learn, under, uh, what objectives and so on.

   9. LFLex Fridman8:01

      Yeah, that's, uh, I- I've heard you mention in several contexts, the idea sort of the way children learn, they interact with objects of the world, and it seems f- uh, fascinating because, uh, in some sense, except with some cases in reinforcement learning, that idea is not part of the learning process, uh, in, uh, artificial neural networks. It's almost like do you envision something like, uh, an objective function saying, "You know what? If you poke this object in this kinda way, it would be really helpful for me to-"

   10. YBYoshua Bengio8:36

       Yes. Yes.

   11. LFLex Fridman8:38

       "... further- further learn?" (laughs)

   12. YBYoshua Bengio8:39

       Right, right.

   13. LFLex Fridman8:39

       So- so almost guiding some aspect of the learning?

   14. YBYoshua Bengio8:43

       Right, right, right. So I was talking to Rebecca Saxe just, uh, an hour ago, and she was talking about lots and lots of evidence from, uh, infants seem to clearly pick what interests them i- in a directed way, and so they're not passive learners. They- they focus their attention on aspects of the world which are most interesting, surprising in- in a non-trivial way that makes them change their theories of the world.

6. 9:20 – 12:38

   Scaling limits: why “just bigger/deeper” isn’t enough, and the sample-efficiency gap

   1. LFLex Fridman9:20

      So, uh, that's a fascinating view of the future progress, but on a- on a more maybe boring, uh, question, do you think going deeper and lar- so do you think just increasing the size of the things that have been increasing a lot in the past few years will- will also make significant progress? So, some of the representational issues that you- you mentioned, that is, they're kind of shallow in some sense.

   2. YBYoshua Bengio9:52

      Oh, shallow, you mean-

   3. LFLex Fridman9:53

      Understanding prior...

   4. YBYoshua Bengio9:53

      ... i- in the sense of abstraction?

   5. LFLex Fridman9:55

      Abstr- in the sense of abstraction. They're not getting some, like you said-

   6. YBYoshua Bengio9:58

      I don't- I don't think that having-

   7. LFLex Fridman9:59

      More-

   8. YBYoshua Bengio10:00

      ... m- more depth in the network in the sense of instead of 100 layers we have 10,000 is going to solve our problem.

   9. LFLex Fridman10:06

      You don't think so?

   10. YBYoshua Bengio10:07

       No.

   11. LFLex Fridman10:09

       Is that obvious to you?

   12. YBYoshua Bengio10:10

       Yes. What is clear to me-

   13. LFLex Fridman10:12

       (laughs)

   14. YBYoshua Bengio10:12

       ... is that engineers and companies and labs, uh, grad students will continue to tune architectures and explore all kinds of tweaks to make the current state of the art slightly, ever slightly better.

   15. LFLex Fridman10:26

       Right.

   16. YBYoshua Bengio10:27

       But I don't think that's gonna be nearly enough. I think we need some fairly drastic changes in the way that we are considering learning to achieve the goal that these learners actually understand, in a deep way, the environment in which they are, uh, you know, observing and acting.

   17. LFLex Fridman10:46

       But I- I guess, uh, I was trying to ask a question that's more interesting than just more layers, uh, is basically once you figure out a way to learn through interacting, how many parameters does it take to store that information? So, I think our brain is quite bigger than most neural networks.

   18. YBYoshua Bengio11:07

       Right, right. Oh, I see what you mean.

   19. LFLex Fridman11:09

       Yeah.

   20. YBYoshua Bengio11:09

       Oh, I- I- I'm with you there. So I agree that in order to build neural nets with the kind of, uh, broad knowledge of the world that typical adult humans have, probably the kind of computing power we have now is gonna be insufficient. So, well, the good news is there are hardware companies building neural net chips, and so it's gonna get better. However, the good news, in a way, which is also bad news, is that even our state-of-the-art deep learning methods fail to learn models that understand even very simple environments, like some grid worlds that we have built. Even these fairly simple environments, I mean, of course if you train them with enough examples, eventually they get it.

   21. LFLex Fridman11:57

       Right.

   22. YBYoshua Bengio11:57

       But it's just, like, instead of what, instead of, uh, what humans might need just, uh, dozens of examples, these things will need...... millions, right, for very, very, very simple tasks. And so, I think there's an opportunity for academics who don't have the (laughs) kind of computing power that, say, Google has, um, to do really important and exciting research to advance the state-of-the-art in, uh, training frameworks, learning models, uh, agent learning, in even simple environments that are synthetic, that seem trivial, but yet current machine learning fails on.

7. 12:38 – 15:40

   Common sense, symbolic AI’s failure, and what neural nets still lack: compositionality

   1. LFLex Fridman12:38

      You talked about priors and common sense knowledge. It seems like, uh, we humans take a lot of s- knowledge for granted. Uh, so w- what, what's your view of these priors, of forming this broad view of the world, uh, this accumulation of information, and how we can teach neural networks or learning systems to pick that knowledge up? So knowledge... You know, for a while, the artificial intelligence, uh, was m- may- maybe in the 80... Like, th- there was a time where knowledge representation, knowledge acquisition, expert systems, I mean, uh, symbolic AI was, was a view, was an interesting problem set to solve, and it, it was kind of put on hold a little bit, it seems like.

   2. YBYoshua Bengio13:27

      Because it doesn't work.

   3. LFLex Fridman13:28

      It doesn't work, that's right. But... (laughs)

   4. YBYoshua Bengio13:31

      (laughs)

   5. LFLex Fridman13:31

      That's right. But the goals of that...

   6. YBYoshua Bengio13:34

      Remain important.

   7. LFLex Fridman13:35

      Yes, remain important.

   8. YBYoshua Bengio13:36

      Absolutely.

   9. LFLex Fridman13:37

      And, and how do you think those goals can be addressed?

   10. YBYoshua Bengio13:40

       Right. So first of all, I believe that, uh, one reason why the, the classical expert systems approach failed is because a lot of the knowledge we have, so you talked about common sense, intuition, um, there's a lot of knowledge like this which is not consciously accessible. There are lots of decisions we're taking that we can't really explain, even if sometimes we make up a story.

   11. LFLex Fridman14:08

       Mm-hmm.

   12. YBYoshua Bengio14:09

       Um, and that knowledge is also necessary for machines to, to, to take good decisions, and that knowledge is hard to codify in, uh, expert systems, rule-based systems, and, you know, classical AI formalism. And there are other issues, of course, with the, uh, old AI, like, um, uh, not really good ways of handling uncertainty. I would say something more subtle, which we understand better now, but I think still isn't enough in the minds of people. There's something really powerful that comes from distributive representations. The thing that really makes neural nets work so well. And it's hard to replicate that kind of power in a symbolic world. The knowledge in, in expert systems and so on is nicely decomposed into, like, a bunch of rules, whereas if you think about a neural net, it's the opposite. You have this big blob of parameters which work intensely together to represent everything the network knows, and it's not sufficiently factorized. And so, I think this is one of the weaknesses of current neural nets, that we have to take lessons from classical AI in order to bring in another kind of compositionality which is common in language, for example, and in these rules, but that isn't so native to neural nets.

8. 15:40 – 18:12

   Disentangled representations—then disentangling the ‘rules’ to avoid catastrophic forgetting

   1. LFLex Fridman15:40

      And on that line of thinking, disentangled representations.

   2. YBYoshua Bengio15:45

      Yes.

   3. LFLex Fridman15:46

      So...

   4. YBYoshua Bengio15:47

      So let me connect with disentangled representations, if you mind, if you don't mind.

   5. LFLex Fridman15:51

      Yes. Yeah, that's exact- yep, yep.

   6. YBYoshua Bengio15:53

      So for many years, I've thought, and I still believe, that it's really important that we come up with learning algorithms, either unsupervised or supervised, but... Or enforcement, whatever, that build representations in which the, uh, important factors, uh, hopefully causal factors, are nicely separated and easy to pick up from the representation. So that's the idea of disentangled representations. Uh, uh, it says transform the data into a space where everything becomes easy. We can maybe just learn with linear models about the things we care about. And, and I still think this is important, but I think this is missing out on a very important ingredient which classical AI systems can remind us of. So let's say we have these disentangled representations. You still need to learn about the, the relationships between the variables, those high-level semantic variables. They're not gonna be independent. I mean, this is like-

   7. LFLex Fridman16:48

      Mm-hmm.

   8. YBYoshua Bengio16:49

      ... too much of a, an assumption. They're gonna have some interesting relationships that allow to predict things in the future, to explain what happened in the past. The kind of knowledge about those relationships in a classical AI system is encoded in the rules, like a rule is just like a little piece of knowledge that says, "Oh, I have these two, three, four variables that are linked in this interesting way, then I, I can say something about one or two of them given a couple others," right? In addition to disentangling the, the elements of the representation, which are like the variables in a rule-based system, you also need to disentangle the, the mechanisms that relate those variables to each other, so, like, the rules. So if the rules are neatly separated, like each rule is, you know, living on its own, and when I l- I change a rule because I'm learning, uh, it doesn't need to break other rules, whereas current neural nets, for example, are very sensitive to what's called catastrophic forgetting, where, uh, after I've learned some things and then I learn new things, I- they can destroy the old things that I had learned, right? If the knowledge was better factorized and, and, uh, separated, disentangled, then you would avoid a lot of that. Now, you can't do this in the sensory domain, but...... by ID-

9. 18:12 – 19:23

   From pixel space to semantic space: disentanglement as a path to stronger generalization

   1. LFLex Fridman18:12

      What do you mean by sensory domain?

   2. YBYoshua Bengio18:13

      Yeah, like in pixel space.

   3. LFLex Fridman18:14

      Mm-hmm.

   4. YBYoshua Bengio18:14

      But, but my idea is that when you project the data in the right semantic space, it becomes possible to now represent this extra knowledge beyond the transformation from input to representations, which is how representations act on each other and predict the future, and so on, i- i- in a way that can be neatly, um, disentangled. So now it's the rules that are disentangled from each other, and not just the variables that are disentangled from each other.

   5. LFLex Fridman18:41

      And you draw a distinction between semantic space and pixel. Like is-

   6. YBYoshua Bengio18:44

      Yes, the, the-

   7. LFLex Fridman18:45

      ... is, is there, does there need to be an architectural difference or is it could-

   8. YBYoshua Bengio18:47

      Well, yeah. So, so there's the sensory space like pixels which, where everything is entangled, uh, and the, the, the, the, the information, like the variables are completely interdependent in very complicated ways. And also computation, like the, the, the, the... it's not just the variables, it's also how they are related to each other is, is all intertwined. But, but I- I'm hypothesizing that in the right high-level representation space, both the variables and how they relate to each other can be disentangled, and that will provide a lot of generalization power.

   9. LFLex Fridman19:21

      Generalization power?

   10. YBYoshua Bengio19:23

       Yes.

10. 19:23 – 20:46

    Out-of-distribution generalization: the sci‑fi novel analogy and shared underlying laws

    1. LFLex Fridman19:23

       Uh, distribution of the test set-

    2. YBYoshua Bengio19:25

       Yes.

    3. LFLex Fridman19:26

       ... is assumed to be the same as the distribution of the training set.

    4. YBYoshua Bengio19:30

       Right. This is where current machine learning is too weak. It, it, it doesn't tell us anything... is not able to tell us anything about how our neural nets, say, are gonna generalize to a new distribution. And, and you know, people may think, "Well, but there's nothing we can say if we don't know what the new distribution will be." The truth is, uh, humans are able to generalize to new distributions.

    5. LFLex Fridman19:53

       Yeah, how are we able to do that? So-

    6. YBYoshua Bengio19:55

       Yeah, because there is something... these new distributions, even though they could look very different from the training distributions, they have things in common. So let me give you a concrete example. You read a science fiction novel. The science fiction novel maybe, you know, brings you in some other planet where things look very different on the surface, but it's still the same laws of physics, right? And so you can read the book and you understand what's going on. So the distribution is very different, but because you can transport a lot of the knowledge you had from Earth about the underlying cause and effect relationships and physical mechanisms and all that, and maybe even social interactions, uh, you can now make sense of what is going on on this planet where like visually, for example, things are totally different.

11. 20:46 – 24:01

    AI risk and public discourse: moving beyond Terminator to real societal impacts

    1. LFLex Fridman20:46

       Taking that analogy further and distorting it, let's enter a sci- science fiction world of say, uh, Space Odyssey 2001 with HAL.

    2. YBYoshua Bengio20:54

       Yeah.

    3. LFLex Fridman20:55

       Or, or maybe, uh, which is probably one of my favorite AI movies and then, uh-

    4. YBYoshua Bengio21:01

       Me too.

    5. LFLex Fridman21:02

       (laughs) And then there's another one that a lot of people love that are maybe a little bit outside the AI community, uh, is Ex Machina.

    6. YBYoshua Bengio21:09

       Right.

    7. LFLex Fridman21:10

       I don't know if you've seen it.

    8. YBYoshua Bengio21:10

       Yes, yes.

    9. LFLex Fridman21:12

       Uh, by the way, what are your views on that movie? I... does it, does... are you able to enjoy it?

    10. YBYoshua Bengio21:16

        So there are things I, I like and things I hate. (laughs)

    11. LFLex Fridman21:20

        (laughs) So let me, uh... (laughs) you could talk about that in the context of a question I wanna ask, which is, uh, there's quite a large community of people from different backgrounds often outside of AI, who are concerned about existential threat of artificial intelligence.

    12. YBYoshua Bengio21:35

        Right.

    13. LFLex Fridman21:35

        You've seen, uh, this community develop over time-

    14. YBYoshua Bengio21:38

        Yeah.

    15. LFLex Fridman21:38

        ... you've seen, you have a perspective. So, uh, what, what do you think is the best way to talk about AI safety, to think about it, to have discourse about it within AI community and outside, and grounded in the fact that Ex Machina is one of the main sources of information for the general public about AI?

    16. YBYoshua Bengio21:56

        So I think, I think you're putting it right. There is a big difference between the sort of discussion we ought to have within the AI community and the sort of discussion that really matter in the general public. So I think the, the picture of Terminator and, you know, AI loose and killing people, and superintelligence that's gonna destroy us whatever we try, isn't really so useful for the public discussion. Because, uh, for the public discussion, the things I believe really matter are the short-term and medium-term very likely negative impacts of AI on society, whether it's from, uh, security like, you know, Big Brother scenarios with face recognition or killer robots, or the impact on the job market or, uh, concentration of power and discrimination. All kinds of social issues which could actually, uh, some of them could really threaten democracy, for example.

    17. LFLex Fridman22:59

        Just to clarify, when you said killer robots, you mean, uh, autonomous weapon-

    18. YBYoshua Bengio23:03

        Yes.

    19. LFLex Fridman23:03

        ... like the weapons systems?

    20. YBYoshua Bengio23:04

        Yes. I don't, I don't mean-

    21. LFLex Fridman23:05

        But we're not talking about Terminator.

    22. YBYoshua Bengio23:05

        No, that's right.

    23. LFLex Fridman23:06

        Uh... (laughs)

    24. YBYoshua Bengio23:07

        So, so I think these, these, uh, short and medium-term concerns should be important parts of the public debate. Now, existential risk for me is a very unlikely consideration, but still worth, uh, academic investigation in the same way that you could say, "Should we study what could happen if a meteorite, you know, came to Earth and destroyed it?" So I think it's very unlikely that this is gonna happen in, or happen in a reasonable future. It's, it's very, um, the, the sort of scenario of, uh, an AI getting loose goes against my understanding of at least current machine learning and current neural nets and so on. It, it's not plausible to me. But of course, I don't have a crystal ball and who knows what AI will be in 50 years from now? So I think it is worth that scientists study those problems. It's just not a pressing question as far as I'm concerned.

12. 24:01 – 28:01

    Ex Machina’s realism problem and how science actually progresses (and why secrecy is unlikely)

    1. LFLex Fridman24:01

       So before I continue down that line, I have a few questions there, but, uh, what, what, what, uh, what do you like and not like about Ex Machina as a movie? 'Cause I, I actually watched it for the second time and enjoyed it. Uh, I hated it the first time and I enjoyed it-... quite a bit more the second time when I sort of learned to accept, uh, certain pieces of it. Uh, m- s- see, see it as a concept movie. How, what was your experience? What, what were your thoughts?

    2. YBYoshua Bengio24:29

       So, the negative is the picture it paints of science is totally wrong. Science in general, and AI in particular. Science is not happening, uh, in some, uh, hidden place by some, you know, really smart guy.

    3. LFLex Fridman24:48

       One person.

    4. YBYoshua Bengio24:49

       One person. This is totally unrealistic. This is not how it happens. Even a team of people in some isolated place will not make it. Science moves by small steps thanks to the collaboration and, um, community of a large number of people interacting. And, um, all the scientists who are expert in their field kinda know what is going on, even in the industrial labs. It's... Information flows and, and leaks and so on. And, uh, and, um-

    5. LFLex Fridman25:23

       Can-

    6. YBYoshua Bengio25:23

       And the spirit of it is very different from, uh, the way science is painted in, in this movie.

    7. LFLex Fridman25:28

       Yeah. Let me, let me ask on that, on that point. It's been the case to this point th- that kind of even if the research happens inside Google or Facebook, inside companies, it still kinda comes out.

    8. YBYoshua Bengio25:39

       Ideas come out.

    9. LFLex Fridman25:40

       Absolutely.

    10. YBYoshua Bengio25:41

        Do you think that will always be the case with AI? Is it, is it possible to bottle ideas to the point where there's a set of breakthroughs that go completely undiscovered by the general research community? Do you think that's even possible? It's possible, but it's unlikely.

    11. LFLex Fridman25:56

        Unlikely.

    12. YBYoshua Bengio25:58

        It's not how it is done now. It's not how I can foresee it in, in the foreseeable future. But, of course, I don't have, um, a crystal ball, and so who knows? This is science fiction after all. But, but usually sci-

    13. LFLex Fridman26:16

        I think it's ominous that the lights went off during, during that discussion (laughs) .

    14. YBYoshua Bengio26:20

        Uh... So, the problem, again, there's, uh, you know, one thing is the movie, and you could imagine all kinds of science fiction. The problem with, for me, maybe similar to the question about existential risk, is that this kinda movie paints such a, a wrong picture of what is actual, you know, the actual science and how it's going on, that, that it, it can have unfortunate effects on people's understanding of current science. And, and so, that's kinda sad. There's a, an important principle in research, which is diversity. So, in other words, research is exploration. Research is exploration in, in the space of ideas. And different people will focus on different directions. And this is not just good, it's essential. So, I'm totally fine with people exploring directions that are contrary to mine or look orthogonal to mine. It's, I, I, I'm more than fine. I think it's important. I and my friends don't claim we have universal truth about what will... especially about what will happen in the future. Now, that being said, we have our intuitions and then we act accordingly, according to where we think we can be most useful and where society has the most to gain or to lose. We should have those debates and, um, and, and, and not end up in a society where there's only one voice and one way of thinking, and research money is spread out. So...

    15. LFLex Fridman27:55

        So, disagreement is, uh, is a sign of, uh, good research, good science. So...

    16. YBYoshua Bengio27:59

        Yes.

13. 28:01 – 31:29

    Bias, regulation, and long-term alignment: from debiasing methods to modeling emotions and morality

    1. LFLex Fridman28:01

       The idea of bias, in, in the human sense of bias.

    2. YBYoshua Bengio28:04

       Yeah.

    3. LFLex Fridman28:05

       Uh, how do you think about instilling in machine learning something that's aligned with human values in terms of bias? We in- intuitively as human beings have a concept of what bias means, of what a fundamental respect to hu- for o- other human beings means. But how do we instill that into machine learning systems, do you think?

    4. YBYoshua Bengio28:26

       So, I think there are short-term things that are already happening, and then there are long-term things that we need to do. In the short term, there are techniques that have been proposed and I think will continue to be improved and maybe alternatives will come up, to take data sets in which we know there is bias, we can measure it. Pretty much any data set where humans are, you know, being observed taking decisions will have some sort of bias, discrimination against particular groups and so on. And we can use machine learning techniques to try to build predictors, classifiers that are gonna be less biased. Uh, we can do it, for example, using adversarial methods to make our systems less sensitive to these variables we should not be sensitive to. So, these are clear, well-defined ways of trying to address the problem, maybe they have weaknesses and, you know, more research is needed and so on. But I think in fact they're sufficiently mature, that governments should start regulating companies where it matters, say like insurance companies, so that they use those techniques. Because those techniques will provably reduce the bias, but at a cost. For example, maybe their predictions will be less accurate, and so companies will not do it until you force them. All right. So, this is short term. Long term, I'm really interested in thinking o- how we can instill, uh, moral values into computers. Obviously, this is not a- something we'll achieve in the next five or ten years. How can we... You know, there's already work in detecting emotions, for example, in images, in sounds, in texts, and also studying how different agents interacting in different ways may correspond to...... patterns of, say, injustice, which could trigger anger. So, these are things we can do in, in the medium term, and eventually train computers to model-

    5. LFLex Fridman30:37

       Mm-hmm.

    6. YBYoshua Bengio30:37

       ... for example, how humans react emotionally, uh, I would say the simplest thing is unfair, uh, situations which trigger anger. This is one of the most basic emotions that we share with other animals.

    7. LFLex Fridman30:52

       Mm-hmm.

    8. YBYoshua Bengio30:52

       I think it's quite feasible within the next few years, so we can build systems that can detect these kind of things-

    9. LFLex Fridman30:58

       Mm-hmm.

    10. YBYoshua Bengio30:59

        ... to the extent, unfortunately, that they understand enough a- about the world around us, which is a long time away, but maybe we can initially do this in virtual environments. So, you can imagine, like, a video game-

    11. LFLex Fridman31:12

        Mm-hmm.

    12. YBYoshua Bengio31:13

        ... where agents interact in, in some ways, and then some situations trigger an emotion. Uh, I think we could train machines to detect those situations and predict that a particular emotion, you know, will likely be, uh, felt if a human was playing one of the characters.

14. 31:29 – 34:02

    Machine teaching and BabyAI: humans-in-the-loop and effective teacher–learner interaction

    1. LFLex Fridman31:29

       You, uh, have shown excitement and done a lot of excellent work with, uh, unsupervised learning, but on a supervi- you know, there's been a lot of, uh, success on the supervised learning side.

    2. YBYoshua Bengio31:39

       Yes, yes.

    3. LFLex Fridman31:40

       Uh, and one, one of the things, um, I'm really passionate about is how humans and robots work together, and, uh, uh, in the context of supervised learning, that means the process of annotation. Do you think about the problem of annotation, of, uh, put in a more, uh, interesting way as humans teaching machines?

    4. YBYoshua Bengio32:01

       Yes.

    5. LFLex Fridman32:01

       Is, is, is there a-

    6. YBYoshua Bengio32:02

       Yes. I think it's an important subject. Reducing it to annotation may be useful for somebody building a system tomorrow, but longer term, the process of teaching I think is something that deserves a lot more attention from the machine learning community. And so there are people who have coined the term machine teaching.

    7. LFLex Fridman32:22

       (laughs)

    8. YBYoshua Bengio32:22

       So, what are good strategies for teaching a learning agent?

    9. LFLex Fridman32:25

       Mm-hmm.

    10. YBYoshua Bengio32:26

        And can we, uh, design and train a system that gonna be, is gonna be a good teacher? So, so in my group, we have a project called, uh, Baby AI or Baby AI Game where there is a, a game, uh, or a scenario where there's a, a, a, a learning agent and a teaching agent. Presumably, the teaching agent would eventually be a human, um, but, you know, we're not there yet. Um, and the, um, the role of the teacher is to use its knowledge of the environment, which it can acquire using whatever way, brute force, um, to help the learner learn as quickly as possible. So, the learner is gonna try to learn by itself, maybe be using some exploration and, and whatever, um, but the teacher can choose, can, can, can have a- an influence on the interaction with the learner so as to guide the learner, um, maybe teach it the things that the learner has most trouble with or just at the boundary between what it knows and doesn't know and so on. So, there's, there's a, there's a tradition of these kind of ideas from other fields, and, um, uh, like tutorial systems, for example, and AI, um, and, and of course, people in, in, in the humanities have been thinking about these questions. But I think it's time that machine learning people, um, look at this, because in the future, we'll have more and more, uh, human-machine interaction-

    11. LFLex Fridman33:55

        Mm-hmm.

    12. YBYoshua Bengio33:55

        ... with a human-in-the-loop, and I think understanding how to make this work better is gonna be-

    13. LFLex Fridman34:00

        All the problems around that are very interesting-

    14. YBYoshua Bengio34:02

        Yes.

15. 34:02 – 36:33

    Language, the Turing test, and world knowledge: Winograd schemas and grounding meaning

    1. LFLex Fridman34:02

       ... and not sufficiently addressed. You've done a lot of, of work with language too. What aspect of, um, the traditionally formulated Turing test, a test of natural language understanding and generation, in your eyes is the most difficult of, of conversation? What in your eyes is the hardest part of conversation to solve for machines?

    2. YBYoshua Bengio34:24

       So, I would say it's everything having to do with the non-linguistic knowledge, which implicitly you need in order to make sense of sentences, things like the Winograd schemas. So, these sentences that are semantically ambiguous. In other words, you need to understand enough about the world in order to really interpret properly those sentences. I think these are interesting challenges for machine learning, because they point in the direction of building systems that both understand how the world works and its causal relationships in the world, and associate that knowledge with how to express it in language, either for reading or writing.

    3. LFLex Fridman35:12

       You speak French?

    4. YBYoshua Bengio35:13

       Yes, it's my mother tongue.

    5. LFLex Fridman35:14

       It's one of the romance languages. Uh, do you think passing the Turing test and all the underlying challenges we just mentioned depend on language? Do you think it might be easier in French than it is in English?

    6. YBYoshua Bengio35:26

       No.

    7. LFLex Fridman35:26

       Or is it independent of language?

    8. YBYoshua Bengio35:28

       Hmm, I think it's independent of language. I, I, I would like to build systems that can use the same principles, the same learning mechanisms to learn from human agents, whatever their language.

    9. LFLex Fridman35:47

       Well, certainly us humans can, uh, talk more beautifully and smoothly in poetry. So, I'm Russian originally. I know poetry in Russian is maybe easier to convey complex ideas than it is in English, but, uh, maybe I'm showing my bias, and some people could say that about Fre- uh, Fr- uh, French. But, uh, of course, the goal ultimately is our human brain is able to utilize any kind of those languages to use them as tools to convey meaning.

    10. YBYoshua Bengio36:18

        Yeah. Of course, there are differences between languages, and maybe some are slightly better at some things, but in the grand scheme of things, uh, where we're trying to understand how the brain works, and language, and so on, I think these differences are minute.

16. 36:33 – 42:24

    Personal lessons: surviving the AI winter, gradual progress, and what’s next (RL, GANs, model-based agents)

    1. LFLex Fridman36:33

       So, you've lived, uh, perhaps through an AI winter of sorts.

    2. YBYoshua Bengio36:38

       Yes. (laughs)

    3. LFLex Fridman36:39

       How did you stay warm-

    4. YBYoshua Bengio36:41

       (laughs) With friends.

    5. LFLex Fridman36:42

       ... and, uh, continue with your research? (laughs)

    6. YBYoshua Bengio36:44

       Stay warm with friends.

    7. LFLex Fridman36:45

       With friends, okay. So, f- it's important to have friends, and, uh, what have you learned from the experience?

    8. YBYoshua Bengio36:53

       Listen to your inner voice. Don't, you know, be trying to just please the crowds, and the fashion. And, uh, if you have a strong intuition about something that is not contradicted by actual evidence, go for it. I mean, i- it could be contradicted by people. (laughs)

    9. LFLex Fridman37:16

       (laughs) But not your own instinct of, based on, uh, everything you've learned all the way-

    10. YBYoshua Bengio37:19

        So, of course, of course, you have to adapt your beliefs when your experiments contradict those beliefs. But, but you have to stick to your beliefs. Otherwise, um, it's, it's, it's what allowed me to go through those years. It's what allowed me to persist in directions that, you know, took time, whatever other people think, took time to mature and, you know, bring fruits.

    11. LFLex Fridman37:48

        So, history of AI is marked with these, um ... Of course, it's marked with technical breakthroughs, but it's also marked with these seminal events that capture the imagination of the community. Most recent, I would say, uh, AlphaGo beating the world champion human Go player was one of those, uh, moments. What do you think the next such moment might be?

    12. YBYoshua Bengio38:12

        Okay. So, first, first of all, I think that these so-called seminal events are overrated.

    13. LFLex Fridman38:20

        (laughs)

    14. YBYoshua Bengio38:21

        Uh, as I said, science really moves by small steps. Now, what happens is, you make one more small step, and it's like the, the drop that, uh, you know, allows to, uh, that fills the bucket. And, and, uh, and then you have drastic consequences, because now you're able to do something you were not able to do before. Or now, say, the cost of building some device or solving a problem becomes, um, cheaper than what existed, and, and you have a new market that opens up, right? So, so especially in the world of, uh, commerce and, and applications, the impact of a small scientific progress could be huge. Um, but in the science itself, I think it's very, very gradual, and, um-

    15. LFLex Fridman39:08

        Where are these steps being taken now? So, there is unsupervised-

    16. YBYoshua Bengio39:12

        Yeah. Right.

    17. LFLex Fridman39:13

        ... uh, uh, learning, and-

    18. YBYoshua Bengio39:13

        So, if I look at, if I look at one trend that I like in, uh, in, in my community ... Um, so for example, in, in, at Mila, in my institute, what are the two hottest topics?

    19. LFLex Fridman39:24

        Mm-hmm.

    20. YBYoshua Bengio39:26

        GANs and reinforcement learning, even though, uh, in, in Montreal, in particular, like, reinforcement learning was something pretty much absent just two or three years ago. So, it, it, there's really a big interest from students, and there's a big interest from people like me. (laughs)

    21. LFLex Fridman39:47

        (laughs)

    22. YBYoshua Bengio39:47

        So, I would say this is something where we're gonna see more progress, even though it hasn't yet provided much, in terms of actual industrial fallout. Like, even though there's AlphaGo, there's no ... Like, Google is not making money on this right now. But I think, over the long term, this is really, really important for many reasons. So, in other words, agent ... I would say reinforcement learning may be, uh, more generally agent learning, 'cause it doesn't have to be with rewards. It could be in all kinds of ways that an agent is learning about its environment.

    23. LFLex Fridman40:20

        Now, reinforcement learning, you're excited about. Do you think, uh, do you think GANs could provide something-

    24. YBYoshua Bengio40:28

        Yes.

    25. LFLex Fridman40:28

        ... some moment in, in, uh, that will be remembered?

    26. YBYoshua Bengio40:31

        Well, GANs or other generative models, I believe will be crucial ingredients in building agents that can understand the world. A lot of the, uh, successes in reinforcement learning in the past has been with, um, policy gradient, where you, you just learn a policy. You don't actually learn a model of the world. But there are lots of issues with that. Um, and we don't know how to do model-based RL right now, but I think this is where we have to go in order to build models that can generalize faster and better, like, to new distributions, um, that capture, to some extent, at least, the underlying causal mechanisms in, in the world.

    27. LFLex Fridman41:15

        Last question. What made you fall in love with artificial intelligence? If you look back, what was the first moment in your life when you s- when you were fascinated by either the human mind or the artificial mind?

    28. YBYoshua Bengio41:31

        You know, when I was an adolescent, I was reading a lot, and then I, I, I started reading science fiction. (laughs)

    29. LFLex Fridman41:37

        (laughs) There you go.

    30. YBYoshua Bengio41:38

        There, I got ... Yeah, that's, that's it. That, that, that's, that's where I got hooked. And then, um, and then, you know, I had one of the first, uh, personal computers, and, uh, I, I got hooked in programming. And so it just, you know ...

Episode duration: 42:18