Lex Fridman Podcast - David Ferrucci: IBM Watson, Jeopardy & Deep Conversations with AI

Episode Date: October 11, 2019

David Ferrucci led the team that built Watson, the IBM question-answering system that beat the top humans in the world at the game of Jeopardy. He is also the Founder, CEO, and Chief Scientist of Elemental Cognition, a company working engineer AI systems that understand the world the way people do. This conversation is part of the Artificial Intelligence podcast. If you would like to get more information about this podcast go to https://lexfridman.com/ai or connect with @lexfridman on Twitter, LinkedIn, Facebook, Medium, or YouTube where you can watch the video versions of these conversations. If you enjoy the podcast, please rate it 5 stars on iTunes or support it on Patreon. Here's the outline with timestamps for this episode (on some players you can click on the timestamp to jump to that point in the episode): 00:00 - Introduction 01:06 - Biological vs computer systems 08:03 - What is intelligence? 31:49 - Knowledge frameworks 52:02 - IBM Watson winning Jeopardy 1:24:21 - Watson vs human difference in approach 1:27:52 - Q&A vs dialogue 1:35:22 - Humor 1:41:33 - Good test of intelligence 1:46:36 - AlphaZero, AlphaStar accomplishments 1:51:29 - Explainability, induction, deduction in medical diagnosis 1:59:34 - Grand challenges 2:04:03 - Consciousness 2:08:26 - Timeline for AGI 2:13:55 - Embodied AI 2:17:07 - Love and companionship 2:18:06 - Concerns about AI 2:21:56 - Discussion with AGI

The following is a conversation with David Farochi. He led the team that built Watson, the IBM question answering system that beat the top humans in the world at the game of jeopardy. For spending a couple hours with David, I saw a genuine passion not only for abstract understanding of intelligence, but for engineering it to solve real world problems under real world deadlines and resource constraints. Where science meets engineering is where brilliant, simple ingenuity emerges. People who work at joining it to have a lot of wisdom earned through failures and eventual success.

David is also the founder, CEO, and chief scientist of Elemental Cognition, a company working to engineer AI systems that understand the world the way people do. This is the Artificial Intelligence Podcast. If you enjoy it, subscribe on YouTube, give it 5 stars on iTunes, support it on Patreon, or simply connect with me on Twitter, Alex Friedman spelled F-R-I-D-M-A-N. And now here's my conversation with David Feroci. Your undergrad wasn't biology with a with an eye toward medical school before you

went on for the PhD in computer science. So let me ask you an easy question. What is the difference between biological systems and computer systems in your when you sit back look at the stars and think philosophically. I often wonder, I often wonder whether or not there is a substantive difference. And I think the thing that got me into computer science and interartificial intelligence was exactly this presumposition that if we can get machines to think, or I should say this question, this philosophical question, if we can get machines to think, to understand, to process information the way we do, so if we can describe a procedure, or describe a process, even if that process were the intelligence process itself,

then what would be the difference? So, from philosophical standpoint, I'm not trying to convince that there is. I mean, you can go in the direction of spirituality, you can go in the direction of a soul, but in terms of you know, what we can experience from an intellectual and physical perspective, I'm not sure there is. Clearly, there are different implementations, but if you were to say, is biological information processing system fundamentally more capable than one we might be able to build out of silicon or some other substrate,

I don't know that there is. How distant do you think is the biological implementation? So fundamentally, they may have the same capabilities, but is it really a far mystery where a huge number of breakthroughs are needed to be able to understand it? Or is it something that, for the most part, in the important aspects, echoes of the same kind of characteristics. Yeah, that's interesting. I mean, so, you know, your question presupposes

that there's this goal to recreate, you know, what we perceive as biological intelligence. I'm not, I'm not sure that's the, I'm not sure that's how I would state the goal. I mean, I think that's studying the goal. Good. So, I think there are few goals. I think that understanding the human brain and how it works is important for us to be able to diagnose

and treat issues for us to understand our own strengths and weaknesses, both intellectual, psychological and physical. So, neuroscience and understanding the brain from that perspective has a clear goal there. From the perspective of saying, I want to mimic human intelligence. That one's a little bit more interesting. Human intelligence certainly has a lot of things we envy. It's also got a lot of problems too. So I think we're capable of sort of stepping back

and saying, what do we want out of, what do we want out of an intelligence? How do we want to communicate with that intelligence? How do we want it to behave? How do we want it to perform? Now of course it's somewhat of an interesting argument because I'm sitting here as a human with a biological brain and I'm critiquing this trends and weaknesses of human intelligence and saying that we have the capacity to step back and say, gee, what is intelligence and what do we really want out of it? And that even in and of itself suggests that human intelligence is something quite enviable that it could, you know, it can, it can, it can, um, introspect that. It could introspect that way. And the flaws, you mentioned the flaws. The humans have, yeah, and I think, I think that

flaws that human intelligence has is extremely prejudicial on bias and the way it draws many inferences. Do you think those are sorry to interrupt? Do you think those are features or are those bugs? Do you think the prejudice, the forgetfulness, the fear, what other flaws? List them all. What love? Maybe that's a flaw. Do you think those are all things that can be gotten, get in the weight of intelligence or the essential components of intelligence. Well, again, if you go back and you define intelligence as being able to sort of accurately,

precisely, rigorously reason, develop answers and justify those answers in an objective way, yeah, then human intelligence has these flaws and that it tends to be more influenced by some of the things you said. And it's largely an inductive process, meaning it takes past data, uses that to predict the future, very advantageous in some cases, but fundamentally biased and prejudicial in other cases, because it's going to be strongly influenced by its priors, whether they're right or wrong from some objective reasoning perspective, you're going to favor

them, because those are the decisions or those are the paths that succeeded in the past. And I think that mode of intelligence makes a lot of sense for when your primary goal is to act quickly and survive and make fast decisions. And I think those create problems when you want to think more deeply and make more objective and reason decisions. Of course, humans capable of doing both. They do sort of one more naturally than they do the other, but they humans capable of doing both. They do sort of one more naturally

than they do the other, but they're capable of doing both. You're saying they do the one that responds quickly more naturally? Right. Because that's the thing we kind of need to not be eaten by the predators in the world. For example, but then we've learned to reason through logic. We've developed science. We trained people to do that. I think that's harder for the individual to do. I think it requires training and teaching.

I think we are human mind certainly is capable of it, but we find more difficult. And then there are other weaknesses, if you will, as you mentioned earlier, just memory capacity and how many chains of inference can you actually go through without like losing your way, so just focus. So the way you think about intelligence,

and we're really sort of floating in this philosophical slightly space, but I think you're like the perfect person to talk about this, because we'll get to Jeopardy and Beyond. That's like an incredible, one of the most incredible accomplishments in AI in the history of AI, but hence the philosophical discussion. So let me ask, you've kind of alluded to it, but let me ask again, what is intelligence underlying the discussions we'll have with with jeopardy and beyond? How do you think about intelligence? Is it a sufficiently complicated problem being able to reason your way through solving

that problem? Is that kind of how you think about what it means to be intelligent? So I think of intelligence to primarily two ways. One is the ability to predict. So in other words, if I have a problem, what's gonna, can I predict what's gonna happen next, whether it's to, you know, predict the answer of a question or to say, look, I'm looking at all the market dynamics,

and I'm gonna tell you what's gonna happen next, or you're in a room and somebody walks in and you're gonna to predict what they're going to do next or what they're going to say next. In a highly dynamic environment full of uncertainty be able to lots of the more variables, the more complex, the more possibilities, the more complex. But can I take a small amount of prior data and learn the pattern and then predict what's going to happen next accurately and consistently That's a that's certainly a form of intelligence What do you need for that by the way you need to have an Understanding of the way the world works in order to be able to unroll it into the future right that you will

What do you think is needed to predict? It depends what you mean by understanding. I need to be able to find that function. This is very much like what's a function. Deep learning does, machine learning does, is if you give me enough prior data and you tell me what the output variable is that matters, I'm going to sit there and be able to predict it. And if I can predict it accurately,

so that I can get it right more often than not, I'm smart. If I can do that with less data and less training time, I'm even smarter. If I can figure out what's even worth predicting, I'm smarter, meaning I'm figuring out what path is going to get me toward a goal. What about picking a goal?

So that's interesting. Well, that's interesting about picking a goal. It's sort of an interesting thing. And I think that's where you bring in what do you pre-program to do? We talk about humans and humans are pre-programmed to survive. So it's sort of their primary driving goal. What do they have to do to do that? And that could be very complex, right?

So it's not just figuring out that you need to run away from the ferocious tiger, but we survive in a social context as an example. So understanding the subtleties of social dynamics becomes something that's important for surviving, finding a mate, reproducing, right? So we're continually challenged with complex, excessive variables, complex constraints, rules, if you will, that we, or patterns. And we learn how to find the functions and predict the things, in other words,

represent those patterns efficiently and be able to predict what's going to happen in that form of intelligence. That doesn't really require anything specific, other than ability to find that function and predict that right answer. It's certainly a form of intelligence. But then when we say, well,

do we understand each other? In other words, would you perceive me as intelligent beyond that ability to predict? So now I can predict, but I can't really articulate how I'm going to that process, what my underlying theory is for predicting. And I can't get you to understand what I'm doing so that you can follow, you can figure out how to do this yourself. If you hadn't, or if you did not have, for example, the right pattern matching machinery that I did, and now we potentially have this breakdown.

We're in effect, I'm intelligent, but I'm sort of an alien intelligence relative to you. You're intelligent, but nobody knows about it. Or I can see the output. So you're saying, let's sort of separate the two things. One is you explaining why you were able to predict the future. And the second is me being able to, like, impressing me that you're intelligent, me being able to know that you successfully predicted the future. Do you think that's...

Well, it's not a pressing you that I'm intelligent. In other words, you may be convinced that I'm intelligent in some form. So because of my ability to predict... So I would... When you can't, I say, wow. You're right. You're right more times than I am. You're doing something interesting.

That's a form of intelligence. But then what happens is if I say, how are you doing that? And you can't communicate with me and you can't describe that to me. Now I may label you as a vaunt. I may say, well, you're doing something weird and it's just not very interesting to me because you and I can't really communicate. And so now, this is interesting, right? Because now this is, you're in this weird place where, for you to be recognized as intelligent the way I'm intelligent, then you and I sort of have to be able to communicate. And then we start to understand each other. And then my respect and my appreciation, my ability to relate to, starts to change.

So now you're not an alien intelligence anymore. You're a human intelligence now because you and I can communicate. And so I think when we look at, when we look at, when we look at animals, for example, animals can do things we can't quite comprehend. We don't quite know how they do them, but they can't really communicate with us. They can't put what they're going through in our terms. And so we think of them as sort of low there, these alien intelligences, and they're not really worthness. So we're worth, we don't treat them the same way as a result of that.

But it's hard because who knows what's going on. So just a quick elaboration on that, the explaining that you're intelligent, the explaining the reasoning that went into the prediction is not some kind of mathematical proof. If you look at humans, look at political debates and discourse on Twitter, it's mostly just telling stories. So your task is not to tell an accurate depiction of how you reason, but to tell a story real or not that convinces me that there was a mechanism by which you ultimately that's what a proof is. I mean, even a mathematical proof is that

because ultimately the other mathematicians have to be convinced by your proof otherwise. In fact, there have been that's the mentioned success. Yeah, there have been several proofs out there where mathematicians would study for a long time before they were convinced that it actually proved anything, right? You never know if it proved anything until the community mathematicians decided that it did. So I mean, so it's, but it's, it's a real thing. And, and, and that's sort of the point, right? Is that ultimately on, you know, this notion of understanding us understanding something is ultimately a social concept. In other words, I have to convince enough people that I did this in a reasonable way. I could do this in a way that

other people can understand and replicate and that makes sense to them. So we're very, human intelligence is bound together in that way. We're bound up in that sense. We never really get away with it until we can consider convinced others that are thinking process makes sense. Did you think the general question of intelligence is then also a social construct? If we ask questions of an artificial intelligence system, is this system intelligent? The answer will ultimately be a socially constructed. I think, I think, so I think, I'm making two statements. I'm saying we can try to define intelligence

in a super objective way. That says here, here's this data. I want to predict this type of thing, learn this function, and then if you get it right, often enough, we consider you intelligent. But that's more of a disadvantage. I think it is. It doesn't mean it's not useful. It could be incredibly useful. It could be solving a problem we can't otherwise solve. And can solve it more reliably than we can. But then there's this notion of can humans take responsibility for the decision that you're

making? Can we make those decisions ourselves? Can we relate to the process that you're going through? And now you as an agent, whether you're a machine or another human, frankly, are now obliged to make me understand how it is that you're arriving at that answer and allow me, me, me, me, or the obviously community or judge of people to decide whether or not, whether or not that makes sense. And by the way, that happens with the humans as well.

You're sitting down with your staff, for example, and you ask for suggestions about what to do next. And someone says, well, I think you should buy, and I think you should buy this much, or whatever it is, or I think you should launch the product today or tomorrow, or launch this product versus that product, whatever decision may be. And you ask why, and the person said,

I just have a good feeling about it. And it's not, you're not very satisfied. Now, that person could be, you know, you might say, well, you've been right, you know, before, but I'm going to put the company on the line. Can you explain to me why I should believe this? And that explanation may have nothing to do with the truth. It's not a convinced. The other. The other. The other. The other. The wrong. Yes, still be wrong. She's got to be convincing. But it's got. It's ultimately got to be

convincing. And that's why I'm saying it's, we're bound together, right? Our intelligences are bound together in that sense. We have to understand each other. And, and if, for example, you're giving me an explanation, I mean, this is a very important point, right? You're giving me an explanation. And I'm, and I, and I have, I, I'm not good. And then I'm not good at reasoning well and being objective and following logical paths and consistent paths.

And I'm not good at measuring and sort of computing probabilities across those paths. What happens is collectively we're not going to do, we're not going to do well. How hard is that problem? The second one. I think we'll talk quite a bit about the first on a specific objective metric benchmark performing well, but being able to explain the steps, the reasoning. How hard is that problem? I think that's very hard.

I mean, I think that's, well, it's hard for humans. The thing that's hard for humans, as you know, may not necessarily be hard for computers and vice versa. So, sorry, so, how hard is that problem for computers? I think it's hard for computers. And the reason why I related to, or saying this also hard for humans is because I think when we step back

and we say we wanna design computers to do that, one of the things we have to recognize is we're not sure how to do it well. I'm not sure we have a recipe for that. Even if you wanted to learn it, it's not clear exactly what data we use and what judgements we use to learn that well. What I mean by that is if you look at the entire enterprise of science, sciences supposed to be at about objective reason and reason, right?

So we think about she who's the most intelligent person or group of people in the world. Do we think about the savants who can close their eyes and give you a number? We think about the think tanks or the scientists or the philosophers who kind of work through the details and write the papers and come up with the thoughtful, logical proofs and use the scientific method. I think it's the latter. My point is that how do you train someone to do that? That's what I mean by it's hard.

What's the process of training people to do that? And that's what I mean by it's hard. How do you, what's the process of training people to do that well? That's a hard process. We work as a society. We work pretty hard to get other people to understand our thinking and to convince them of things. Now we could so weigh them, obviously, we talked about this, like human flaws or weaknesses,

we can persuade them through emotional means, but to get them to understand and connect to and follow a logical argument is difficult. We do it as scientists, we try to do it as journalists, we try to do it as, you know, even artists in many forms as writers, as teachers, we go through a fairly significant training process to do that. And we could ask, well, why is that so hard? But it's hard. And for humans, it takes a lot of work. And when we step back and say, well, step back and say, well, how do we get a machine to how do we get a machine to do that? It's a vaccine question.

How would you begin to try to solve that and maybe just a quick pause because there's an optimistic notion in the things you're describing, which is being able to explain something through reason. But if you look at algorithms that recommend things that we look at next, whether it's Facebook, Google, advertisement-based companies, their goal is to convince you to buy things based on anything. So that could be reason, because the best of advertisement is showing you things that you really do need and explain why you need it. But it could also be through emotional manipulation. The algorithm that describes why a certain reason, a certain decision was made, how hard

is it to do it through emotional manipulation? And why is that a good or a bad thing? So you've kind of focused on reason, logic, really showing in a clear way why something is good. One, is that even a thing that us humans do? And two, how do you think of the difference in the reasoning aspect and the emotion of manipulation? Well, you know, so you call it emotional manipulation, but more objectively is essentially saying, you know, there are certain features of things that seem to attract your attention. I mean,

it kind of give you more of that stuff. I mean, manipulation is a bad word. Yeah, I mean, I'm not saying it's good right or wrong. It works to get your attention and it works to get you to buy stuff. And when you think about algorithms that look at the patterns of the, you know, patterns of features that you seem to be spending your money on and is everyone to give you something with a similar pattern. So I'm going to learn that function because the objective is to get you to click on and get you to buy it or whatever it is. I don't know. I mean, it is like it is what it is. I mean, that's what the algorithm does. You can argue whether it's good or bad. It depends what your, you know, what your, what your goal is.

I guess this seems to be very useful for convincing, but telling us to be. It's good because you, because again, this goes back to what is the human behavior like? What is the human brain respond to things? I think there's a more optimistic view of that too, which is that if you're searching for certain kinds of things, you've already reasoned that you need them. And these algorithms are saying, look, that's up to you. The reason whether you need something or not, that's your job. You may have an unhealthy addiction

to this stuff, or you may have a reasoned and thoughtful explanation for why it's important to you. And the algorithms are saying, hey, that's like whatever, like that's your problem. All I know is you're buying stuff like that. You're interested in stuff like that. That could be a bad reason, could be a good reason. That's up to you. I'm going to show you more of that stuff. And so, and I think that that's, it's not good or bad.

It's not a reason or not a reason. The algorithm is doing what it does, which is saying you seem to be interested in this. I'm going to show you more of that stuff. And I think we're seeing this not just in buying stuff, but even in social media, you're reading this kind of stuff. I'm not judging on whether it's good or bad. I'm not reasoning at all. I'm just saying, I'm going to show you other stuff with similar features.

And like in that said, and I wash my hands from it, and I say, that's all that's all that's going on. You know, there is, you know, people are so harsh on AI systems. So one, the bar of performance is extremely high. And yet, we also ask them, in the case of social media, to help find the better angels of our nature and help make a better society. So what do you think about the role of AI?

So that's the interesting dichotomy, right? Because on one hand, we're sitting there and we're sort of doing the easy part, which is finding the patterns. We're not building, the system's not building a theory that is consumable and understandable by other humans that can be explained and justified. And, and so on one hand to say, oh, you know, AI is doing this, why isn't doing this other thing? Well, so other things a lot harder. And it's interesting to think about why it's harder. And because you're interpreting,

you're interpreting the data in the context of prior models. In other words, understandings of what's important in the world, what's not important. We're to all the other abstract features that drive our decision making, what's sensible, what's not sensible, what's good, what's bad, what's moral, what's valuable, what isn't. Where is that stuff?

No one's applying the interpretation. So when I see you clicking on a bunch of stuff, and I look at these simple features, the raw features, the features that are there in the data, like what words are being used, or how long the material is, or other very superficial features, what colors are being used in the material, like I don't know why you're clicking on this stuff, you're looking, or if it's products, what the price is or what the categories, and stuff like that. And I just feed you more of the same stuff.

That's very different than kind of getting in there and saying, what does this mean? The stuff you're reading, like, why are you reading it? What assumptions are you bringing to the table? Are those assumptions sensible? Does the material make any sense? Does it lead you to thoughtful, good conclusions? Again, there's this interpretation, judgment involved in that process

that isn't really happening in the AI today. That's harder. Because you have to start getting at the meaning of the of the of the stuff of the content. You have to get at how humans interpret the content relative to their value system and deeper thought processes. So that's what meaning means is not just some kind of deep, timeless, semantic thing that the statement represents, but also how a large number of people are likely to interpret. So that's, again, even meaning is a social construct.

So you have to try to predict how most people would understand this kind of statement. Yeah, meaning is often relative, but meaning implies that the connections go beneath the surface of the artifacts. If I show you a painting, it's a bunch of colors on a canvas, what does it mean to you? And it may mean different things to different people because of their different experiences. It may mean something even different to the artist who painted it. As we try to get more rigorous with our communication, we try to really nail down that meaning. So we go from abstract art to precise mathematics,

precise engineering drawings and things like that. We're really trying to say, I want to narrow that space of possible interpretations because the precision of the communication ends at becoming more and more important. So that means that I have to specify, and I think that's why this becomes really hard. Because if I'm just showing you an artifact and you're looking at it superficially, whether it's a bunch of words on a page, or whether it's, you know, brushstrokes on a canvas or pixels on a photograph, you can sit there and you can interpret lots of different ways at many, many different levels.

But when I want to align our understanding of that, I have to specify a lot more stuff that's actually not directly in the artifact. Now I have to say, well, how are you interpreting this image and that image and what about the colors and what do they mean to you? What's what perspective are you bringing to the table? What are your prior experiences with those artifacts? What are your fundamental assumptions and values? What is your ability to kind of reason to chain together logical implication as you're sitting there and saying,

well, if this is the case, then I would conclude this. And if that's the case, then I would conclude that. And so your reasoning processes and how they work, your prior models and what they are, your values and your assumptions, all those things now come together into the interpretation. Getting in sync of that is hard. And yet, humans are able to intuit some of that

without any pre, because they have to share experience. And we're not talking about shared to people having shared experience. We have the society. That's correct. We have the shared experience experience and we have similar brains. So we tend to, in other words, part of our shared experiences are shared local

experience like we may live in the same culture, we may live in the same society and therefore we have similar educations, we have similar what we like to call prior models about the world, prior experiences. And we use that as a think of it as a wide collection of interrelated variables. And they're all bound to similar things. And so we take that as our background. And we start interpreting things similarly. But as humans, we have it. We have a lot of shared experience. We do have similar brains, similar goals, similar emotions under similar circumstances, because we're both humans So now one of the early questions you asked well how is biological and you know computer information systems

Fundamentally different. Well one is you know one is Humans come with a lot of pre-programmed stuff. Yeah a ton of program stuff and they're able to communicate because they have a lot of Because they share that stuff. Do you think that shared knowledge, if we can maybe escape the hardware question, how much is encoded in the hardware? Just the shared knowledge in the software, the history, the many centuries of wars and so on that came to today. That shared knowledge. How hard is it to encode? Do you have a hope? Can you speak to how hard is it to encode that knowledge systematically in a way that could be used by a computer? So I think it is possible to learn for a machine to program machine to acquire that knowledge with a similar foundation in other words, a similar interpretive foundation for processing that knowledge.

What do you mean by that? So in other words, we view the world in a particular way. So in other words, we have if you will, as humans, we have a framework for interpreting the world around us. So we have multiple frameworks for interpreting the world around us. But if you're interpreting, for example, social political interactions, you're thinking about whether there's people, there's collections and groups of people, they have goals, goals largely built around survival and quality of life

There are they're fundamental economics around scarcity of resources and when when humans come and start interpreting a situation like that because you brought you brought up like historical events They start interpreting situations like that. They apply a lot of this fundamental framework for interpreting that. Well, who are the people? What were their goals? What reasons did they have? How much power influence did they have over the other?

Does fundamental substrate, if you will, for interpreting and reasoning about that? So, I think it is possible to imbue a computer with that stuff that humans take for granted when they go and sit down and try to interpret things. And then with that foundation, they acquire. They start acquiring the details, the specifics and then given situation, are then able to interpret it with regard to that framework. And then given that interpretation, they can do what? They can predict. But not only can they predict, they can predict now with an explanation that can be given in those terms, in the terms of that underlying framework that most humans share.

Now, you can find humans that come in interpretive events very differently than other humans, because they're like using a different different framework, you know, movie matrix comes to mind where, you know, they decided humans were really just batteries and that's how they interpret the value of humans as a source of election energy. So, but, um, but I think that, you know, for the most part, we have a way of interpreting the events or the social events around us because we have this shared framework. It comes from again, the fact that we're similar beings that have similar goals, similar emotions, and we can make sense out of these frameworks, make sense to us. So, how much knowledge is there, do you think? So you said it's possible.

Well, it's from at this amount of detailed knowledge in the world. You know, you can imagine effectively infinite number of unique situations and unique configurations of these things. But the knowledge that you need, what I refer to as like the frameworks for you need for interpreting them, I don't think. I think those are finite. You think the frameworks are more important than the bulk of the not, so like framing.

Yeah, because what the frameworks do is they give you now the ability to interpret and reason, and to interpret and reason it to interpret and reason over the specifics in ways that other humans would understand. What about the specifics? When you acquire the specifics by reading and by talking to other people. So, mostly, actually, just even if you can focus on

the beginning, the common sense stuff, the stuff that doesn't even require reading or it almost requires playing around with the world or something, just being able to sort of manipulate objects, drink water and so on, all of that. Every time we try to do that kind of thing in robotics or AI, it seems to be like an onion. You seem to realize how much knowledge is really required to perform even some of these basic tasks. Do you have that sense as well? And

if so, how do we get all those details? Are they written down somewhere? Do you have to be learned through experience? So I think when, like you're talking about sort of the physics, the basic physics around us, for example, acquiring information about acquiring how that works. Yeah, I think there's a combination of things going on. I think there's a combination of things going on. I think there is like fundamental pattern matching, like what we were talking about before, where you see enough examples, enough data about something, you start assuming that, and with similar input, I'm going to predict similar outputs. You don't can't necessarily explain it at all. You may learn very quickly that when you let something go, it falls to the ground.

But that's a such a deep idea that if you let something go, like the idea of gravity. I mean, people are letting things go and counting on the falling well before they understood gravity. But that seems to be as exactly what I mean, is before you take a physics class or studying anything about Newton, just the idea of the stuff falls to the ground and then you'll be able to generalize that all kinds of stuff falls to the ground. It just seems like a non,

without encoding it, like hard coding it in, it seems like a difficult thing to pick up. It seems like you have to have a lot of different knowledge to be able to integrate that, into the framework, sort of, into everything else. So both know that stuff falls to the ground and start to reason about social political discourse. So both like the very basic and the high level reasoning decision making. I guess my question is how hard is this problem?

And sorry to linger in it because again, and we'll get to it for sure, as Watson with Jeopardy did, is take on a problem that's much more constrained, but has the same hugeness of scale, at least from the outside of his perspective. So I'm asking the general life question of to be able to be an intelligent being and reasoning in the world about both gravity and politics. How hard is that problem? So, I think it's solvable. Okay, that's beautiful. So, what about time travel?

Okay, I'm not as convinced. Not as convinced. Yeah. Okay. No, I think I think it is I think it is solvable. I mean, I think that it's a learn it's first of all, it's about getting machines to learn. Learning is fundamental. And I think we're ready in a place that we understand, for example, how machines can learn in various ways. Right now, our learning stuff is sort of primitive, in that we haven't sort of taught machines to learn the frameworks. We don't communicate our frameworks because of how

shared, in some cases, we do, but we don't annotate, if you will, all the data in the world with the frameworks that are inherent or underlying our understanding. Instead, we just operate with the data. So if we want to be able to reason over the data in similar terms in the common frameworks, we need to be able to teach the computer, or at least we need to program the computer to acquire, to have access to and acquire, learn the frameworks as well and connect the frameworks to the data. I think this can be done. I think we can start, I think machine learning,

for example, with enough examples, can start to learn these basic dynamics. Will they relate the necessary to gravity, not unless they can also acquire those theories as well, and put the experiential knowledge and connect it back to the theoretical knowledge. I think if we think in terms of these class of architectures that are designed to both learn the specifics, find the patterns, but also acquire the frameworks and connect the data to the frameworks, if we think in terms of robust architectures like this, I think there is a path toward

getting there. In terms of encoding architectures like that, do you think systems that were able to do this will look like neural networks or representing if you look back to the 80s and 90s of the expert systems. So there are more like graphs, systems that are based in logic, able to contain a large amount of knowledge where the challenge was the automated acquisition of that knowledge. I guess the question is, when you collect both the frameworks and the knowledge from the data, what do you think that thing will look like?

Yeah. I mean, I think asking the question they look like neural networks is a bit of a red herring. I mean, I think that they will certainly do inductive or pattern match based reasoning. I've already experimented with architectures that combine both that use machine learning and neural networks to learn certain classes of knowledge in those words to find repeated patterns in order for it to make good inductive guesses. But then ultimately to try to take those learnings and and marry them, in other words, connect them to frameworks

so that it can then reason over that in terms of their humans understand. So for example, at Elemental Cognition, we do both. We have architectures that do both. But both those things, but also have a learning method for acquiring the frameworks themselves and saying, look, ultimately, I need to take this data. I need to interpret it in the form of these frameworks so that it can reason over it. So there is a fundamental knowledge representation

like what you're saying, like these graphs of logic, if you will. There are also neural networks that acquire certain class of information. Then they then align them with these frameworks. But there's also a mechanism to acquire the frameworks themselves. Yeah, so it seems like the idea of frameworks

requires some kind of collaboration with humans. Absolutely. So do you think of that collaboration as the right? Well, and it was to be clear. Only for the express purpose that you're designing your designing a machine, you're designing an intelligence that can ultimately communicate with humans

in terms of frameworks that help them understand things. Right, so now to be really clear, you can create, you can independently create a machine learning system and intelligence that I might call an alien intelligence that does a better job than you with some things, but can't explain the framework to you. That doesn't mean it might be better than you at the thing. It

might be that you cannot comprehend the framework that it may have created for itself that is inexplicable to you. That's a reality. But you're more interested in a case where you can. I am. Yeah. My sort of approach to AI is because I've set the goal for myself. I want machines to be able to ultimately communicate understanding with human.

I want to be able to acquire and communicate, acquire knowledge from humans and communicate knowledge to humans. They should be using what inductive machine learning techniques are good at, which is to observe patterns of data, whether it be in language, or whether it be in images, or videos, or whatever, to acquire these patterns, to induce the generalizations from those patterns, but then ultimately work with humans to connect them to frameworks and interpretations, if you will, that ultimately makes sense to humans. Of course, the machine is going to have the strength that it has the richer and longer memory, but that, you know, it has the more rigorous reasoning

abilities, the deeper reasoning abilities. so it would be an interesting, you know, complementary relationship between the human and the machine. Do you think that ultimately needs to explain ability like a machine? So if you look at study, for example, Tesla autopilot a lot, where humans, I don't know if you've driven the vehicle or are aware of what it is. So you're basically the human and machine are working together there and the human is responsible for their own life to monitor the system. And the system fails every few miles.

And so there's hundreds of millions of those failures a day. And so that's like a moment of interaction. Do you see that? Yeah, that's exactly right. That's a moment of interaction where the machine has learned some stuff. It has a failure, somehow the failure is communicated.

The human is now filling in the mistake, if you will, or maybe correcting or doing something that is more successful in that case, the computer takes that learning. So I believe that the collaboration between in that case, the computer takes that learning. So I believe that the collaboration between you and a machine, I mean, that's sort of a primitive example and sort of a more,

another example is where the machine's literally talking to you and saying, look, I'm reading this thing. I know that like the next word might be this or that, but I don't really understand why. I have my guess, can you help me understand the framework that supports this and then can kind of take, acquire that, take that and reason about it and reuse it. The next time it's reading to try to understand something, not unlike a human student might do. I remember when my daughter was in first grade, she was had a

reading assignment about electricity. And somewhere in the text it says, an electricity is produced by water flowing over turbines or something like that. And then there's a question that says, well, how is electricity created? And so my daughter comes to me and says, I mean, I could create you know, create it and produce their kind of synonyms in this case. So I can go back to the tax and I can copy by water flowing over turbines, but I have no idea what that means. Like, I don't know how to interpret water flowing over turbines and what electricity even is. I mean, I can get the answer right by matching the text, but I don't have any framework for understanding what this means at all. And framework really is, I mean, it's really is a set of not to be mathematical but axioms of Ideas that you bring to the table and interpreting stuff and then you build those up somehow you build them up with the expectation that

There's a shared understanding of what they are share it. Yeah, it's the social the right that us humans Do you have a sense that humans on earth in general share a set of Like how many frameworks are there? I mean it depends on how you bound them right? So in other words how big or small like they're their individual scope But there's lots and there are new ones. I think they're I think the way I think about is kind of a layer. I think the architectures are being layered in that. There's a small set of primitives that allow you the foundation to build frameworks. And then there may be many frameworks, but you have the ability to acquire them. And then you have the ability to reuse them. I mean, one of the most compelling ways of thinking about this is a reasoning by analogy where I can say, oh, wow, I've learned something very similar.

You know, I never heard of this, I never heard of this game soccer, but if it's like basketball in the sense that the goal is like the hoop and I have to get the ball in the hoop and I have guards and I have this and I have that, like where is the, where are the similarities and where the differences? And I have a foundation now for interpreting this new information and then different groups like the millennials will have a framework and then and then well that you know, yeah, well like that Democrats and Republicans well, I mean I think I think right I mean they're talking about political and social ways of interpreting the world around them.

And I think these frameworks are still largely, largely similar. I think they differ in maybe what some fundamental assumptions and values are. Now, from a reasoning perspective, like the ability to process the framework, it might not be that different. The implications of different fundamental values are fundamental assumptions in those frameworks. The implications of different fundamental values or fundamental assumptions in those frameworks frameworks may reach very different conclusions So from so from a a social perspective that conclusions may be very different from an intelligence perspective I you know, I just followed where my assumptions took me Yeah, the process self looks similar, but that's a fascinating idea that frameworks really help carve how a statement will be interpreted. I mean, having a Democrat and a Republican framework and read the exact same statement and

the conclusions that you derive will be totally different from an AI perspective as fascinating. What we would want out of the AI is to be able to tell you that this perspective, one perspective, one set of assumptions is going to lead you here, no other set of assumptions is going to lead you there. And in fact, you know, to help people reason and say, oh, I see where, I see where our differences lie. Yeah. You know, I have this fundamental belief about that.

I have this fundamental belief about that. Yeah, that's quite brilliant. From my perspective, and NLP, there's this idea that there's one way to really understand a statement, but that probably isn't. There's probably an infinite number of ways to understand that. There's a lot of different interpretations. Well, there's a lot of different interpretations. The broader, the broader the content, the richer it is. And so, you and I can have very different experiences

with the same text, obviously. And if we're committed to understanding each other, we start, and that's the other important point, like if we're committed to understanding each other, we start decomposing and breaking down our interpretation to its more and more primitive components. Until we get to that point where we say, oh, I see why we disagree. And we try to understand how fundamental that disagreement really is. But that requires a commitment to breaking down that interpretation in terms of that framework in a logical way.

Otherwise, and this is why I think of AI is really complimenting and helping human intelligence to overcome some of its biases and its predisposition to be persuaded by more shallow reasoning in the sense that we get over this idea, well, I'm right because I'm Republican or I'm right because I'm Democratic and someone labeled as a Democrat at a point of view, or it has to follow on keywords in it. And if the machine can help us break that argument down and say, wait a second, you know, what

do you really think about this? Right? So essentially holding us accountable to doing more critical thinking. I'm going to sit and think about that as fast. That's, I love that. I think that's really empowering use of AI for the public discourse that's completely disintegrating currently, as we learn how to do it on social media. That's right. So one of the greatest accomplishments in the history of AI is Watson competing in the game of

jeopardy against humans. And you were allied in at a critical part of that. Let's start at the very basics. What is the game of jeopardy? The game for us humans, human versus human. Right. So it's to take a question and answer it. Um, the game and Jeff, wow, actually, actually, well, no, but it's not, right? It's really not. It's really, it's really to get a question and answer. But it's, it's what we call a factoid question. So this notion of like, it's, it really relates to some fact that every few people would argue whether the facts are true to some fact that every, few people would argue whether the facts are true or not. In fact, most people wouldn't.

Jeffrey kind of counts on the idea that these statements have factual answers. And the idea is to, first of all, determine whether or not you know the answer, which is sort of an interesting twist. So, first of all, understand the question. You have to understand the question. What is it asking? And that's a good point because the questions are not asked directly, right?

They're all like, the way the questions are asked is non-linear. It's like, it's a little bit witty. It's a little bit playful sometimes. It's a little bit tricky. Yeah, they're asked in exactly, numerous witty tricky ways, exactly what they're asking is not obvious.

It takes an experience, humans a while to go, what is it even asking? And it's sort of an interesting realization that you have. And somebody says, oh, what's the, jeopardy is a question answering show. And then he's like, oh, like I know a lot. And then you read it and you're, you're still trying to process the question and the champions have answered and moved on. There are three questions ahead, but the time you figured out what the question even meant. So there's definitely an ability there to just parse out what the question even is.

So that was certainly challenging. It's interesting historically though, if you look back at the jeopardy games, much earlier, you know, like 60-70 games. The questions were much more direct. They weren't quite like that. They got sort of more and more interesting. The way they asked them, that sort of got more and more interesting and subtle and nuanced and humorous and witty over time, which really required the human

to kind of make the right connections and figuring out what the question was even asking. So yeah, you have to figure out the questions even asking. Then you have to determine whether or not you think you know the answer. And because you have to buzz in really quickly, you sort of have to make that determination as quickly as you possibly can.

Otherwise you lose the opportunity to buzz in. You're going before you really know if you know the answer. I think a lot of humans will assume they'll look at, they'll look at their process very superficially. In other words, what's the topic, what are some keywords, and just say, do I know this area or not, before they actually know the answer? Then they'll buzz in and then they'll buzz in

and think about it. It's interesting what humans do. Now some people who know all things, like Ken Jennings or something, or the more recent big jeopardy player, having those will suppose that. They'll just assume they know all the jeopardy and they'll just suppose that. You know, Watson interestingly didn't even come close to knowing all of jeopardy, right? Watson, we're... Even at the peak, even at the best.

Yeah, so for example, I mean, we had this thing called recall, which is like, how many of all the jeopardy questions, you know, how many did could we even find like, find the right answer for like anywhere? Like, can we come up with if we look, you know, we had a big body of null, some of the order of several terabytes. I mean, from a web scale was actually very small, but from like a book scale, talking about millions of books, right? So they're quote, millions of books, a cycle of pdaries, books, so it's a ton of information. And you know, for, I think it was only 85% was the answer anywhere to be found. So you're already down, you're already down at that level just to get started, right? So

and so it was important to get a very quick sense of, do you think you know the right answer to this question? So we had to compute that confidence as quickly as we possibly could. So in effect, we had to answer it and at least, you know, spend some time essentially answering it and then judging the confidence that we, you know, that our answer was right and then deciding whether or not we were confident enough to buzz in. And that would depend on what else was going on in the game, because it was a risk.

So like if you're really in a situation where I have to take a gas, I have very little to lose, then you'll buzz in with less confidence. So that was accounted for the financial standings of the different competitors. Correct. How much of the game was live? How much time was left? Where you were in the standing? Things like that. How many hundreds of milliseconds that we're talking about here? Do you have a sense of what is...

We targeted... Yeah, we targeted. So, I mean, we targeted the answering in under three seconds. And buzzing it. So, the decision buzz in, and then the actual answering are those two different things. Yeah, they were two different things. In fact, we had multiple stages, whereas we would say, let's estimate our confidence, which was sort of a shallow answering process.

And then ultimately decide to buzz in, and then we may take another second or something to kind of go in there and do that. But by and large, we're saying like, we can't play the game, we can't even compete if we can't, on average, answer these questions in around three seconds or less. So you stepped in, so there's these three humans

playing a game and you stepped in with the idea that IBM Watson would be one of the the replaced one of the humans and compete against to can you tell the story of Watson taking on this game It seems exceptionally difficult. Yeah, so the story Was that um it was it was coming up. I think the 10-year anniversary of a big blue I don't think, deep blue. IBM wanted to do sort of another kind of really fun challenge,

public challenge that can bring attention to IBM research and the kind of the cool stuff that we were doing. I had been working in AI at IBM for some time. I had a team doing what's called open domain factoid question answering, which is, you know, we're not going to tell you what the questions are. We're not even going to tell you what they're about. Can you go off and get accurate answers to these questions?

And it was an area of AI research that I was involved in. And so it was a very specific passion of mine. Language understanding and always been a passion of mine. One sort of narrow slice on whether or not you could do anything with language was this notion of open domain, meaning I could ask anything about anything. Factoid meaning it essentially had an answer and being able to do that accurately and quickly. So that was a research area that my team had already been in.

And so completely independently, several IBM executives, like, what are we gonna do? What's the next cool thing to do? And Ken Jennings was on his winning streak. This was like, whatever was 2004, I think. Was on his winning streak. And someone thought, hey, that'll be really cool if the computer can play Jeopardy. And so this was like in 2004,

they were shopping this thing around. And everyone was telling the research execs No, what like this is crazy And we have some pretty you know senior people in the field and saying now this is crazy and it'll come across my desk And I was like, but that's kind of what what I'm really interested in doing and But there were such this prevailing sense of this is not we're not going to risk IBM's reputation on this We're just not doing it and this happened in 2004 it happened in 2005 at the end of 2006 It was coming around again

And I was coming off of a I was doing it the open domain question answering stuff But I was coming off of a couple a couple other projects Had a lot more time to put into this and I argued that it could be done and I argued it would be crazy not to do this. Can I, you could be honest at this point. So even though you argued for it, what's the confidence that you had yourself privately that this could be done? We just told the story, how you tell stories to convince others.

How confident were you? What was your estimation of the problem at that time? So I thought it was possible and a lot of people thought it was impossible. I thought it was possible. Reason why I thought it was possible is because I did some brief experimentation. I knew a lot about how we were approaching open domain Factoid question asked we've been doing it for some years. I looked at the Jeffery stuff. I said, this is going to be hard for a lot of the points that we mentioned earlier, hard to interpret the question, hard to do it quickly enough, hard to compute an accurate confidence. None of this stuff had been done well enough before. But a lot of the technologies were building with the kinds of technologies that should work. But more to the point what was driving me was, I was an IBM research, I was a senior leader

in IBM research, and this is the kind of stuff we were supposed to do. Basically, we were supposed to do this. The moonshot. I mean, we were supposed to take things and say, this is an active research area. It's our obligation to kind of, if we have the opportunity to push it to the limits. And if it doesn't work, to understand more deeply why we can't do it. And so I was very committed to that notion saying, folks, this is what we do. It's crazy not to do this.

This is an active research here. We've been in this for years. Why wouldn't we take this grand challenge and push it as hard as we can? At the very least, we'd be able to come out and say, here's why this problem is way hard. Here's what we try to hear is how we fail. So I was very driven as the science is from that perspective. And then I also argued, based on what we did,

a feasibility study, a why I thought it was hard but possible, and I showed examples of where it succeeded, where it failed, why it failed, and sort of a high level architectural approach for why we should do it. But for the most part, at that point, the exact sort of really we're just looking for someone crazy enough to say yes,

because for several years at that point, everyone had said no, I'm not willing to risk my reputation and my career, you know, on this day. Clearly you did not have such fears. Okay, I did not. So you dived right in and yet, for what I understand, it was performing very poorly in the beginning. So what were the initial approaches and why did they fail?

Well, there were lots of hard aspects to it. I mean, one of the reasons why prior approaches that we had worked on in the past failed was because of, because the questions were difficult, difficult to interpret, like what are you even asking for, right? Very often like if the question was very direct, like what city, you know, or what, you know, even then it could be tricky, but you know what city or what person is often when it would name it very clearly,

you would know that. And if there were just a small set of them, in other words, we're going to ask about these five types. Like it's going to be an answer, and the answer will be a city in this state or a city in this country. The answer will be a person of this type, right? Like an actor or whatever it is. But it turns out that in jeopardy, there were like tens of thousands of these things. And it was a very, very long tail. Meaning, you know, it just went on and on.

And so even if you focused on trying to encode the types at the very top, like there's five that were the most, let's say, five of the most frequent, you still cover a very small percentage of the data. So you couldn't take that approach of saying, I'm just going to try to collect facts about these five or 10 types or 20 types or 50 types or whatever. So that was like one of the first things like what do you do about that? And so we came up with an approach toward that and the approach to look promising. And we continue to improve our ability to handle that problem throughout the project.

The other issue was that right from the outside, I said, we're not going to, I committed to doing this in three to five years. So we did it in four. So I got lucky. But one of the things that that, putting that like, staking the ground, was I, and I knew how hard the language of the standing problem was I said we're not going to actually understand language of self-discipline problem. We are not going to interpret the question and the domain of knowledge that the question refers to and reason over

that to answer these questions. I'm just going to be doing that. At the same time, simple search wasn't good enough to to confidently answer with this, you know, a single correct answer. First of all, that's brilliant. That's such a great mix of innovation in practical engineering, three, three, four, eight. So you're not trying to solve the general L&LU problem. You're saying, let's solve this in any way possible. Oh, yeah.

No, I was committed to saying, look, we're just solving the open domain question. And some problem were using jeopardy as a driver for the ad. Big benchmark hard enough. Big benchmark exactly. And now we're, how do we do it? We could just like whatever like just figure out what works because I want to be able to go back to the academic science and community and say, here's what we tried. Here's what work. Here's what didn't work. I don't want to go in and say, Oh, I only have one techno, I have a hammer and I'm only going to use this. I'm going to do whatever it takes.

I'm like, I'm going to think out of the box and do whatever it takes. One, and I also, there was another thing I believed. I believed that the fundamental NLP technologies and machine learning technologies would be adequate. And this was an issue of how do we enhance them? How do we integrate them? How do we advance them? So I had one researcher and came to me,

who had been working on question answering with me for a very long time, who had said, we're going to need Maxwell's equations for question answering. And I said, if we, if we need some fundamental formula that breaks new ground and how we understand language, we're screwed. We're not going to get there from here. Like, we, I am not counting. I am, my assumption is I'm not counting on some brand new invention. What I'm counting on is the ability to take everything that has done before to figure out a out an architecture on how to

integrate it well and then see where it breaks and make the necessary advances we need to make and tell this thing works. Yeah, but it should hard to see where it breaks and then pass it over. I mean, that's how people change the world. I mean, that's the Elon Musk approach with rockets, SpaceX, that's the Henry Ford and so on. I love it. And this case, I happen to be right, but we didn't know. But you kind of have to put a statement and say, how are you going to run the project? So yeah, and backtracking to search.

So if you were to do, what's the brute force solution? What would you search over? So you have a question, how would you search the possible space of answers? Look, web searches come a long way even since then. But at the time, you know, you'd, first of all, I mean, there are a couple of other constraints around the problems, interesting. So you couldn't go out to the web. You couldn't search the internet.

In other words, the AI experiment was, we want a self-contained device. The device is as big as a room, fine, it's as big as a room, but we want a self-contained device. Contain device. You're not going out the internet. You don't have a lifeline to anything. So it had to kind of fit in a shoebox, if you will, or at least a size of a few refrigerators, whatever it might be. But also, you couldn't just get out there.

You couldn't go off network to kind of go. So there was that limitation. But then we did, but the basic thing was go do web search. Problem was even when we went and did a web search, I don't remember exactly the numbers, but someone in order of 65% of the time, the answer would be somewhere in the top 10 or 20 documents.

So first of all, that's not even good enough to play jeopardy. You know, even if you could pull the, if you could perfectly pull the answer out of the top 20 documents, top 10 documents, whatever it was, which we didn't know how to do. But even if you could do that, you do that, and you knew it was right, and it was just had enough confidence in it, right? So you'd have to pull out the right answer, you'd have to have confidence it was the right answer.

And then you'd have to do that fast enough to now go buzz in, and you'd still only get 65% of them right, which doesn't even put you in the winter circle. Winter circle, you have to be up over 70, and you have to do it really quickly. But now the problem is, well, even if I had somewhere in the top 10 documents,

how do I figure out where in the top 10 documents that answer is and how do I compute a confidence of all the possible candidates? So it's not like I go in knowing the right answer and have to pick it, I don't know the right answer. I have a bunch of documents, somewhere in there is the right answer. How do I as a machine go out and figure out which one's right?

And then how do I score it? So, and now how do I deal with the fact that I can't actually go out to the web? First of all, if you pause on that, just think about it. If you could go to the web, do you think that problem is solvable? If you just pause on it, just thinking even beyond jeopardy. Do you think the problem of reading text defined where the answer is? Well, we saw that in some definition of solbs given the jeopardy challenge. How did you do it for jeopardy? So how did you take a body of work in a particular topic and extract the key pieces of information?

So, forgetting about the huge volumes that are on the web, right? So now we have to figure out, we did a lot of source research. In other words, what body of knowledge is gonna be small enough, but broad enough to answer jeopardy? And we ultimately did find the body of knowledge. It did that. I mean, it included Wikipedia and a bunch of other stuff.

So like inside of the pdf type of stuff, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it.

In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it.

In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak to it. In fact, I don't know if you can speak seeds using those seeds for other searches and then expanding that.

So using these expansion techniques, we went out and found enough content and we're like, okay, this is good. And even up until the end, we had a thread of research that's always trying to figure out what content could we efficiently include. I mean, there's a lot of popular, like what is the church lady? Well, I think it was one of the, like, what, I guess that's probably an encyclopedia. So I guess that's what we would take that stuff when we would go out and we would expand.

In other words, we go find other content that wasn't in the core resources and expanded. The amount of content grew it by an order of magnitude, but still, again, from a web scale perspective, this is very small amount of content. It's very select. We then took all that content, we pre-analyzed a crap out of it, meaning we parsed it, broke it down into all this individual words, and we did semantic, satanic and semantic

parses on it, had computer algorithms that annotated it, and we indexed that in a very rich and very fast index. So we have a relatively huge amount of, let's say the equivalent of, for the sake of argument, two to five million bucks, we've now analyzed all that, blowing up at size even more because now it bulls metadata, and we then we richly indexed all of that. And by the way, in a giant in memory cache. So Watson did not go to disk. So the infrastructure component there, if you could just speak to it, how tough it,

I mean, I know 2000, maybe this is 2008, 2009, you know, that's kind of a long time ago. Right. How hard is it to use multiple machines? How hard is the infrastructure for the hardware component? So we used IBM. So we used IBM hardware. We had something like I forgot exactly about 2000, close to 3000 cores completely connected. So you had a switch where, you know, every CPU was connected to every other thing. You know, we're sharing

memory in some kind of way. It's kind of cleverly. Shared memory, right? And all this data was pre-analyzed and put into a very fast indexing structure. That was all in memory. And then we took that question, we would analyze the question. So all the content was now pre-analyzed. So if I went and tried to find a piece of content, it would come back with all the content was now pre-analyzed. So if I went and tried to find a piece of content, it would come back with all the metadata that we had pre-computed. How do you shove that question?

How do you connect the big knowledge base of the metadata and that's indexed to the simple little witty, confusing question? Right. So therein lies, you know, the Watson Arches, so we would take the question. We would analyze the question, so which means that we would parse it and interpret it a bunch of different ways. We try to figure out what is it asking about?

So we would come, we had multiple strategies to kind of determine what was it asking for. That might be represented as a simple string and character string, or something we would connect back to different semantic types that were from existing resources. So anyway, the bottom line is we would do a bunch of analysis in the question. And question analysis had to finish,

and had to finish fast. So we do the question analysis because then from the question analysis, we would now produce searches. So we would, and we had built using open source search engines, we modified them. But we had a number of different search engines we would use that had different characteristics. We went in there and engineered and modified those search engines. Ultimately, to now take our question analysis, produce multiple queries based on different interpretations of the question, and fire out a whole bunch of searches in parallel.

And they would come back with passages. So these were passage search algorithms. They would come back with passages. And so now you, let's say you had a thousand passages. Now for East passage, you parallelize again. So you went out and you parallelized the search. Each search would now come back with a whole bunch of passages. Maybe you had a total of a thousand or five thousand, whatever passages. For each passage now,

you'd go and figure out whether or not there was a candidate, we'd call a candidate answer in there. So you had a whole bunch, another, a whole bunch of other algorithms that would find candidate answers, possible answers to the question. And so you had candidate answer, called candidate answers generators, a whole bunch of those. So for every one of these components,

the team was constantly doing research coming up better ways to generate search queries from the questions, better ways to analyze the question, better ways to generate candidates. And speed. So better is accuracy and speed. Correct. So right, speed and accuracy for the most part were separated. We handled that sort of in separate ways, like I focus purely on accuracy and to an accuracy. Are we ultimately getting more questions and producing more accurate conferences?

And then a whole other team that was constantly analyzing the workflow to find the bottlenecks, and then figuring out about parallelize and drive the algorithm speed. But anyway, so now think of it like you have this big fan hour now, right? Because you have multiple queries now you have now you have thousands of candidate answers.

For each candidate answer, you're gonna score it. So you're going to use all the data that built up. You're going to use the question analysis. You're going to use how the query was generated. You're going to use the passage itself and you're going to use the candidate answer that was generated and you're going to score that. So now we have a group of researchers coming up with scores. There are hundreds of different scores. So now you're getting a fan out of it again, from whoever many candidate answers you have, to all the different scores. So if you have a 200 different scores and you have a thousand candidates, you have 200,000 scores. And so now you've got to figure out, you know, how do I now rank these answers based on the scores

that came back? And I want to rank them based on the likelihood that they're correct the answer to the question. So every score was its own research project. What do you mean by score? So is that the annotation process of basically a human being saying that this answer has a quality of? Think of it if you want to think of it, what you're doing, process of basically human being saying that this answer has quality.

Think of it. If you want to think of it, what you're doing, if you want to think about what a human would be doing, human would be looking at a possible answer. They'd be reading the Emily Dixon to consent. They'd be reading the passage in which that occurred. They'd be looking at the question and they'd be making a decision of how likely it is that Emily Dickinson, given this evidence in this passage, is the right answer to that question.

Got it. So that's the annotation task. That's the scoring task. So but scoring implies zero to one kind of continuous. That's right. You give it zero to one score. So it's not a binary. No, so it's a score.

You give it a zero, yeah, exactly zero. So humans give different scores so that you have to somehow normalize and all that kind of stuff that deal with all that depends on what your strategy is. We both, we could be relative to it could be. We actually looked at the raw scores as well, standardized scores because humans are not involved in this. Humans are not involved. Sorry.

So I'm misunderstanding the the process here. There's just passages. Where is the ground truth coming from? Grant truth is only the answers to the questions. So it's end to end. It's end to end. So we all so I was always driving end to end. And performance was a very interesting a very interesting, you know, engineering approach and ultimately scientific and research approach always driving intent. Now, that's not to say we wouldn't make hypotheses that individual component performance was related in some way to end that performance.

Of course, we would because people would have to build individual components. But ultimately, to get your components into the system, you have to show impact on end-to-end performance, question answering performance. There's many, very smart people working on this and they're basically trying to sell their ideas as a component that should be part of the system. That's right. And they would do research on their component and they would say things like, you know, I'm going to improve this as a candidate generator.

Or I'm going to improve this as a question score or as a passive score. I'm going to improve this. Or as a parser. And I can improve it by 2% on its component metric, like a better parse or better candidate or a better type estimation, whatever it is. And then I would say, I need to understand how the improvement on that component metric

is gonna affect the end-to-end performance. If you can't estimate that and can't do experiments that demonstrate that, it doesn't get in. That's like the best run AI project I've ever heard. This is awesome. Okay. What breakthrough would you say? Like, I'm sure there's a lot of day-to-day breakthroughs,

but was there like a breakthrough that really helped improve performance? Like, wait, what people began to believe? Or was it just a gradual process? Well, I think it was a gradual process, but one of the things that I think gave people confidence that we can get there was that as we follow this procedure of different ideas, build different components, plug them into the architecture, run the system, see how we do, do the air analysis, start off new research projects to improve things, and the very important idea

that the individual component work did not have to deeply understand everything that was going on with every other component. And this is where we leverage machine learning in a very important way. So while individual components could be statistically driven machine learning in a very important way. So while individual components could be statistically driven machine learning components, some of them were heuristic, some of them were machine learning components, the system has a whole combined all the scores

using machine learning. This was critical because that way you can divide and conquer. So you can say, okay, you work on your candidate generator or you work on this approach to answer scoring or you work on this approach to answer scoring. You work on this approach to type scoring. You work on this approach to passage search or to passage selection and so forth. But when we just plug it in and we had enough training data to say now we can we can train and figure out how do we weigh all the scores relative to each other based on the predicting

the outcome, which is right, right or wrong on jeopardy. And we had enough training data to do that. So this enabled people to work independently and to let the machine learning do the integration. Beautiful. So the, yeah, the machine learning is doing the fusion and then it's a human orchestrated ensemble. That's right. I'm trying to frame the approaches. That's great. So yeah, the machine learning is doing the fusion and then it's a human orchestrated ensemble I friend approaches. That's great Still impressive. You're able to get it done a few years

That that not obvious to me that it's doable if I just put myself in that mindset But when you look back at the Jeopardy challenge Again when you're looking up at the stars, what are you most proud of? It's looking back at those days. I'm most proud of my commitment and my team's commitment to be true to the science, to not be afraid to fail. That's beautiful because there's so much pressure because it is a public event. It is a public show that you were dedicated to the idea.

That's right. Do you think it was a success? In the eyes of the world, it was a success. By your, I'm sure, exceptionally high standards. Is there something you regret you would do differently? It was a success. It was a success for our goal. Our goal was to build the most advanced, open the main question answering system.

We went back to the old problems that we used to try to solve, and we did dramatically better on all of them, as well as we beat jeopardy. So we won a jeopardy. So it was a success. It was, I worry that the world would not understand it as success because it came down to only one game. And I knew statistically speaking, this can be a huge technical success and we could still lose that one game. And that's a whole other theme of this, of the journey. But it was a success. It was not a

success in natural language understanding, but that was not the goal. Yeah, that was the, but I would argue, I understand what you're saying in terms of the science, but I would argue that the inspiration of it, right, the, not a success in terms of solving natural language understanding. There was a success of being an inspiration to future challenges. Absolutely.

The drive future efforts. What's the difference between how human being compete in jeopardy and how Watson does it? That's important in terms of intelligence. Yeah, so that actually came up very early on in the project also. In fact, I had people who wanted to be on the project who were early on who sort of approached me once I committed to do it. I wanted to think about how humans do it. And they were, you know, from a cognition perspective, like human cognition and how that should play. And I would not take them on the project because another assumption or another

state I put in the ground was, I don't really care human's do this. I, at least in the context of this project. I need to build in the context of this project in NLU and in building an AI that understands how it needs to ultimately communicate with humans, I very much care. So it wasn't that I didn't care in general. In fact, as an AI scientist, I care a lot about that, but I'm also a practical engineer and I committed to getting this thing done, and I wasn't going to get distracted. I had to kind of say, look, if I'm going to get this thing done. And I wasn't gonna get distracted. I had to kind of say like, if I'm gonna get this done, I'm gonna chart this path.

And this path says, we're gonna engineer a machine that's gonna get this thing done. And we know what search and NLP can do, we have to build on that foundation. If I come in and take a different approach and start wondering about how the human mind might or might not do this

I'm not gonna get there from here in the time and you know in the time frame I think that's a great way to lead the team But now there's done and then one you're back right so analyze what's the difference actually right? So so I was a little bit surprised actually to discover Over time as this would come up from time to time and we'd reflect on it, that and talking to Ken Jennings a little bit and hearing Ken Jennings talk about how he answered questions, that it might have been closer to the way humans answer questions than I might have imagined previously.

Because humans are probably in the game of jeopardy at the level of congeny things are probably also cheating. They're a weight to winning right? Well, they're doing shallow analysis shallow. They're doing their fastest possible. They're doing shallow analysis. So they are very quickly analyzing the question and coming up with some you know key you know key vectors or cues, if you will. They're taking those cues and they're very quickly going through their library of stuff, not deeply reasoning about what's going on. And then sort of like, a lot's of different, like what we call these scores, what's kind

of score in a very shallow way, and then say, oh, boom, you know, that's what it is. And so it's interesting as we reflected on that. So we may be doing something that's not too far off from the way humans do it, but we certainly certainly didn't approach it by saying, you know, how would a human do this. Now in elemental cognition, like the project I'm leading now, we ask those questions all the time, because ultimately, we're trying to do something that is to make the intelligence and the machine and the intelligence of the human very compatible. Well, compatible in a sense, they can communicate with one another, and they can reason with

their shared understanding. So how do they think about things and how they build answers, how they build explanations, becomes a very important question to consider. So, what's the difference between this open domain but cold, constructed question answering of jeopardy and more, something that requires understanding for shared communication with humans and machines. Yeah, well, this goes back to the interpretation of what we were talking about before. The system's not trying to interpret the question,

and that's not interpreting the content that's reusing and with regard to any particular framework. I mean, it is parsing it and parsing the content and using grammatical cues and stuff like that. So if you think of grammar as a human framework in some sense, it is parsing it and parsing the content and using grammatical cues and stuff like that. So if you think of grammar as a human framework in some sense, it has that. But when you get into the richest semantic frameworks, what do people, how do they think? What motivates them? What are the events that are occurring and why are they occurring and what causes what else

to happen and where are things and time and space? And like when you started thinking about how humans formulate and structure the knowledge that they acquire in their head and wasn't doing any of that. What do you think are the essential challenges of like free flowing communicate, free flowing dialogue versus question answering even with with a framework, with the interpretation. Dialogue. Yep.

Do you see free-flowing dialogue as a fundamentally more difficult than question answering, even with shared interpretation? Yeah, so dialogue is important in a number of different ways. I mean, it's a challenge. So first of all, when I think about the machine that understands language and ultimately can reason in an objective way, that can take the information

that it perceives through language or other means and connect it back to these frameworks, reason and explain itself, that system ultimately needs to be able to talk to humans, right? It needs to be able to interact with humans. So in some sense, it needs to dialogue. That doesn't mean that it, that, like sometimes people talk about dialogue and they think, you know, how do humans talk, you know, how do humans talk to like talk to each other

in a casual conversation? And you can mimic casual conversations. We're not trying to mimic casual conversations. We're really trying to produce a machine, this goal is to help you think and help you reason about your answers and explain why. So instead of like talking to your friend down the street about having a small talk conversation with your friend down the street, this is more about like you would be communicating to the computer on Star Trek, where,

what do you wanna think about? Like what do you wanna reason about? I'm gonna tell you the information I have, I'm gonna have to summarize it. I'm gonna ask you questions, you're gonna answer those questions. I'm gonna go back and forth with you. I'm gonna figure out what your mental model is.

I'm gonna now relate that to the information I have and present it to you in a way that you can understand that and we can ask follow up questions. So it's that type of dialogue that you want to construct. It's more structured, it's more goal oriented, but it needs to be fluid. In other words, it can't, it can't, it has to be engaging and fluid. It has to be productive and not distracting. So there has to be a model of, in other words, the machine has to have a model of how humans think through things and discuss them. So basically a productive rich conversation. productive rich conversation, unlike this podcast, what I like to think it's more similar to this podcast is joking. I'll ask you about humor as well, actually. But what's the hardest part of that? Because it seems we're quite far away

as a community from that still, to be able to, so one is having a shared understanding. That's, I think, a lot of the stuff you said with frameworks is quite brilliant, but just creating a smooth discourse, and I feel clunky right now. What, which aspects of this whole problem that you just specified of having a

productive conversation is the hardest and that were, or maybe any aspect of it you can comment on because it's so shrouded in mystery. So I think to do this you have to be creative in the following sense. If I were to do this, it's purely a machine learning approach. And someone said, learn how to have a good, fluent, structured, knowledge acquisition conversation. I'd go out and say, okay, I have to collect a bunch of data of people doing that. People reasoning well, having a good, structured conversation that both acquires knowledge efficiently as well as produces answers and explanations as part of the process.

And you struggle. I don't know to collect the data, to collect the data because I don't know how much data is like that. Okay, okay, this one, there's a humorous commenter on the lack of rational discourse. But also, even if it's out there, say it was out there, how do you actually, how do you, how do you, I think, successful, then.

Right, so I think any, like any problem like this, where you don't have enough data to represent the phenomenon you want to learn, in other words, you want, if you have enough data, you could potentially learn the pattern. And an example like this, it's hard to do. It's sort of a human sort of thing to do. What recently came out IBM was the debate or projects and it's interesting, right? Because now you do have these structured dialogues, these debate things where they did use machine

learning techniques to generate these debates. Dialogues are a little bit tougher in my opinion than generating a structured argument where you have lots of other structured arguments like this. You can potentially annotate that data and you can say this is a good response, it's a bad response in a particular domain. Here, I have to be responsive and I have to be opportunistic with regard to what is the human saying?

So I'm goal oriented and saying, I want to solve the problem. I want to acquire the knowledge necessary, but I also have to be opportunistic and responsive to what the human is saying. So I think that it's not clear that we could just train on the body of data to do this, but we can bootstrap it. In other words, we can be creative. And we can say, what do we think? What do we think the structure of a good dialogue is that does this well? And we can start to create that. If we can create that more programmatically, at least to get this process started.

And I can create a tool that now engages humans effectively. I can start both, I can start generating data. I can start the human learning process, and I can update my machine. But I also start the automatic learning process as well. But I have to understand what features to even learn over. So I have to bootstrap the process a little bit first. And that's a creative design task that I could then use as input into a more automatic learning task. So some creativity in, yeah, in boost trapping.

Right. What elements of conversation do you think you would like to see? So one of the benchmarks for me is humor, right? That seems to be one of the hardest. And to me, the biggest contrast was Watson. So one of the greatest sketches of comedy sketches of all time, right, is the SNL celebrity jeopardy with Alex DeBrec and Sean Connery and Bert Reynolds and so on, with Sean Conner commentating on Alex DeBk's mother a lot. And I think all of them are in a negative point, what's why? So they're clearly all losing in terms of the gaming jeopardy, but they're winning in terms of comedy.

So what do you think about humor in this whole interaction in the dialogue that's productive? Or even just whatever what humor it represents to me is the same idea that you're saying about framework, because humor only exists within a particular human framework. So what do you think about humor? What do you think about things like humor that connect to the kind of creativity you mentioned that's needed? I think there's a couple things going on there., I sort of feel like, and I might be too optimistic this way, but I think that there are, we did a little bit about with this and with puns and in jeopardy.

We literally sat down and said, well, you know, how to puns work. And you know, it's like word play and you could formalize these things. So, I think there's a lot aspects of humor that you could formalize. You could also learn humor, you could just say what do people laugh at. And if you have enough, again, if you have enough data to represent the phenomenon, you might be able to weigh the features

and figure out what humans find funny and what they don't find funny. The machine might not be able to explain why do you want to find a good body unless we sit back and think about that more formally. I think, again, I think you do a combination of both. And I'm always a big proponent of that. I think robust architectures and approaches are always a little bit combination of us

reflecting and being creative about how things are structured and how to formalize them, and then taking advantage of large data and doing learning and figuring out how to combine these two approaches. I think there's another aspect to human though, which goes to the idea that I feel like I can relate to the person telling the story. Telling this person telling the story. And I think that's an interesting theme in the whole AI theme, which is, do I feel differently when I know it's a robot? is, do I feel differently when I know it's a robot? And when I know, when I imagine that the robot is not conscious the way I'm conscious, when I imagine the robot does not

actually have the experiences that I experience, do I find it funny? Or do, because it's not as relate, I don't imagine that the person is relating it to it, the way I relate to it. I think this also, you see this in the arts and in an entertainment where, sometimes you have savants who are remarkable at a thing, whether it's sculpture, it's music or whatever,

but the people who get the most attention are the people who can evoke a similar emotional response who can get you to emote, right, about the way they are. In other words, who can basically make the connection from the artifact, from the music or the painting of the sculpture to the emotion and get you to share that emotion with them. And then, and that's when it becomes compelling. So they're communicating at a whole different level. They're just not communicating the artifact.

They're communicating their emotional response to the artifact and then you feel like, oh wow, I can relate to that person. I can connect to that person. So I think humor has that aspect as well. So the idea that you can connect to that person, person being the critical thing. But we're also able to anthropomorphize objects, pretty robots and AI systems pretty well. So we're almost looking to make them human. Maybe from your experience with Watson, maybe you can comment on, did you consider that as part, well, obviously the problem of the jeopardy doesn't require anthropomorphization, but nevertheless. Well, there was some interest in doing that, and that's another thing I didn't want to

do, because I didn't want to distract from the actual scientific task. Science. But you're absolutely right. Humans do anthropomorphize, and without necessarily a lot of work, I mean, just put some eyes in a couple of eyebrow movements and you're getting humans to react emotionally. And I think you can do that. So I didn't mean to suggest that that connection

cannot be mimicked. I think that connection can be mimicked and can produce that emotional response. I just wonder though, if you're told what's really going on, if you know that the machine is not conscious, not having the same richness of emotional reactions and understanding that it doesn't really share the understanding, but it's essentially just moving its eyebrow,

or drooping its eyes, or making them bigger, whatever it's doing, just getting the emotional response, will you still feel it? Interesting. I think you probably would for a while. And then when it becomes more important that there's a deeper, a deeper share of understanding that it may run flat. But I don't know. I'm, I'm, I'm pretty, I'm pretty confident that it will, that majority of the world,

even if you tell them how it works. No matter. Well, it will not matter, especially if the machine herself says that she is conscious. That's very possible. So you were the scientist that made the machine, is saying that this is how the algorithm works. Everybody will just assume you're lying and that there's a conscious being there. So you're deep into the science fiction genre now, but yeah, I don't think it's actually psychology. I think it's not science fiction.

I think it's reality. I think it's a really powerful one that will have to be exploring in the next few decades. It's not exactly. It's a very interesting element of intelligence. So what do you think we've talked about social constructs of intelligence and frameworks in the way humans kind of interpret information?

What do you think is a good test of intelligence in your view? So there's the Alan Turing with the Turing test. Watson accomplished something very impressive with jeopardy. What do you think is a test that would impress the heck out of you that you saw that a computer could do? They would say, this is crossing a kind of threshold that gives me pause in a good way. My expectations for AI are generally high. What does high look like, by the way?

So not the threshold. Test is a threshold. What does high look like, by the way? So not the threshold test is a threshold. What do you think is the destination? What do you think is the ceiling? I think machines will in many measures will be better than us, will become more effective, in other words, better predictors about a lot of things than ultimately we can do. I think where they're going to struggle is what we've talked about before, which is relating to communicating with and understanding humans in deeper ways. I think that's a key point.

We can create the superparad, what I mean by the superparad is, given enough data, machine can mimic your emotional response, can even generate language that will sound smart and what someone else might say under similar circumstances. Like I would just pause on that, like that's a superpar it, right? So given similar circumstances, it's faces in similar ways, changes its tone of voice in similar ways, produces strings of language that, you know, would similar that a human might say, not necessarily

being able to produce a logical interpretation or understanding that would ultimately satisfy a critical interrogation or a critical understanding. I think you just described me in a nutshell. I think philosophical speaking, you could argue that that's all we're doing is human beings to where so far. I was going to say it's very possible humans do behave that way too. And so upon deeper probing and deeper interrogation, you may find out that there isn't a shared understanding because I think humans do both. deeper probing and deeper interrogation, you may find out that there isn't a shared understanding.

Because I think humans do both. Like humans are statistical language model machines and they are capable reasoners. You know, they're both. And you don't know which is going on, right? So, and I think it's an interesting problem we talked earlier about like where we are And I think it's an interesting problem. We talked earlier about like where we are in our social and political landscape.

Can you distinguish someone who can string words together and sound like they know what they're talking about from someone who actually does? Can you do that without dialogue, with that interrogative appropriating dialogue? So it's interesting because humans are really good at in their own mind Justifying or explaining what they hear because they project They're understanding on onto yours

So you could say you could put together a string of words and and Someone will sit there and interpret in the way this extremely bias to the way they want interpret They want to assume you're an idiot and they'll interpret it one way. They will all see your genius and interpret it Another way that suits their needs. So this is tricky business So I think to answer your question As AI gets better and better better and better mimic we create the super parrots We're challenged just as we are with, we're challenged with demons. Do you really know what you're talking about? Do you have a meaningful interpretation,

a powerful framework that you could reason over and justify your answers, justify your predictions and your beliefs? Why do you think they make sense. Can you convince me what the implications are? Can you reason intelligently and make me believe that those, the implications of your prediction and so forth? So what happens is it becomes reflective. My standard for judging your intelligence depends a lot on mine. But you're saying there should be a large group of people with a certain standard of intelligence that would be convinced by this particular AI system than the past.

There should be, but I think one of the, depending on the content, one of the problems we have there is that if that large community of people are not judging it with regards to a rigorous standard of objective logic and reason, you still have a problem like masses of people can be persuaded. The millennials, yeah, to turn their brains off. Right. Right. Okay. Sorry.

I have nothing against the millennium. No, I don't. Just. So you're a part of one of the great benchmarks, challenges of AI history. What do you think about Alpha Zero, Open AI 5, Alpha Star accomplishments on video games recently, which are also, I think, at least in the case of Go, with Alpha Go and Alpha Zero playing Go, was a monumental accomplishment as well. What are your thoughts about that challenge?

I think it was a giant landmark for AI. I think it was phenomenal. I mean, as one of those other things, nobody thought like solving Go was going to be easy, particularly because it's getting it's hard for particularly hard for humans, um, hard for humans to learn, hard for humans to excel at. And so it was a, another measure, a measure of intelligence. Um, it's very cool. I mean, it's very interesting, you know, what they did. I mean, and I love how they solved like the data problem, which is, again, they bootstrapped it and got the machine

to play itself to generate enough data to learn from. I think that was brilliant. I think that was great. And, and of course, the result speaks for itself. I think it makes us think about, again, is it okay, what's intelligence? What aspects of intelligence are important? Can the Go machine help me make me a better go player? Is it an alien intelligence?

You know, is, is, am I even capable of like, again, if we, if we put in very simple terms, it found the function, it found the go function. Can I even copy hand the go function? Can I talk about the go function? Can I conceptualize the go function like whatever it might be? So one of the interesting ideas of that system is that it plays against itself, right? But there's no human in the loop there. So like you're saying, it could have by itself

created an alien intelligence. Toward a goal, I imagine you're sentencing, you're judging your sentencing people, or you're setting policy, or you're making medical decisions and you can't explain. You can't get anybody to understand what you're doing or why. So it's an interesting dilemma for the applications of AI. Do we hold AI to this accountability that says, you know, humans have to be, humans have to be able to take responsibility, you know, for the decision. In other words, can you explain

why you would do this thing? Will you, will you get up and speak to other humans and convince them that this was a smart decision? Is the AI enabling you to do that? Can you get behind the logic that was made there? Do you think, sorry, the link on this point, because it's a fascinating one. It's a great goal for AI. Do you think it's achievable in many cases? Or, okay, there's two possible worlds that we have in the future. One is where AI systems do like medical diagnosis

or things like that, or drive a car without ever explaining to you why it fails when it does. That's one possible world, and we're okay with it. Or the other where we are not okay with it and we really hold back the technology from getting too good before it's able to explain. Which of those worlds are more likely to think and which are concerning to you or not.

I think the reality is it's going to be a mix. I'm not sure I have a problem with that. I mean, I think there are tasks that are perfectly fine with machines show a certain level of performance and that level of performance is already better than humans. So for example, I don't know that I take driverless cars. If driverless cars learn how to be more effective driver's than humans, but can't explain what they're doing, but bottom line, statistically speaking, they're 10 times safer than humans. I don't know that I care.

I think when we have these edge cases, when something bad happens and we want to decide who's liable for that thing and who made that mistake and what do we do about that and I think in those edge cases are interesting cases. And now do we go to designers of the AI and the AI says, I don't know, that's what it learned to do and it says, well, you didn't train it properly. You know, you were negligent in the training data that you gave that machine. Like, how do we drive down the real life? Oh, so I think those are, I think those are interesting questions. And so the optimization problem there,

sorry, is to create an ass system, it's able to explain the lawyers away. Yeah, there you go. There you go. I think that, I think it's gonna be interesting. I mean, I think this is where technology and social discourse are gonna get like deeply intertwined And how we start thinking about

Problems decisions and problems like that. I think in other cases it becomes more obvious where You know, it's like I Like why did you decide to give that person, you know a longer sentence or or to deny deny them parole Again policy decisions or why did you pick that treatment like that treatment ended up killing that guy? Like why was that a reasonable choice to make? So and people are gonna demand explanations. Now there's a reality though here. And the reality is that it's not,

I'm not sure humans are making reasonable choices when they do these things. And the reality is that it's not, I'm not sure humans are making reasonable choices when they do these things. They are using statistical hunches, biases, or even systematically using statistical averages to make calls. This is what happened to my dad and a few of us were talking about that. But, you know, I mean, they decided that my father was brain dead. He had went into cardiac arrest and it took a long time

for the ambulance to get there and he wasn't not resuscitated right away. And so forth, and they came, they told me he was brain dead. And why was he brain dead? Because essentially, they gave me a purely statistical argument under these conditions with these four features, 98% chance he's brain dead. And I said, but can you just tell me not inductively, but deductively go there and tell me his brain

Stop functioning is the way for you to do that and and and they're the protocol in response was no, this is how we make this decision I said this is adequate for me. I understand the statistics and I don't have you know There's a 2% chance he's told I like I just don't know the specifics I need the specifics of this case And I want the specifics of this case. And I want the deductive logical argument about why you actually know he's brained. So I wouldn't sign the do not resuscitate. And I don't know it was like they went through lots of procedures, a big long story, but the bottom was a fascinating story by the way, but how I reasoned and how the doctors reasoned

through this whole process. But I don't know, somewhere around 24 hours later or something, he was sitting up in bed with zero brain damage. He will lessons you draw from that. Story. That experience. That the data that the data that's being used to make statistical inferences doesn't adequately reflect the phenomenon. So in other words, you're getting shit wrong. Sorry, you're getting stuff wrong because your model is not robust enough and you might be better off not using statistical inferences and statistical averages in certain cases

when you know the model is insufficient and that you should be reusing it about the specific case more logically and more deductively, and hold yourself accountable to doing that. And perhaps AI has a role to say the exact thing we just said, which is perhaps this is a case you should think for yourself, you should reason deductively. Well, it's hard because it's hard to know that. You'd have to go back and you'd have to have enough data to essentially say, and this goes back to how do we, this goes back to the case of how do we decide whether AI is good enough to do a particular task.

And regardless of whether or not it produces an explanation. So, and what standard do we hold? Right? For that. So, um, you know, if you look at, you look more broadly, for example, as my father as a medical case, the medical system ultimately hoped them a lot throughout his life. Without it, he probably would have died much sooner. So overall, sort of, you know, work for him in sort of a net, net kind of way. Actually, I don't know if that's fair.

But it may be not in that particular case, but overall. Like, the medical system overall does more good than that. Yeah, medical system overall was doing more good than that. Now there's another argument that suggests that there wasn't the case, but for the sake of argument let's say like that's, let's say a net positive. And I think if you sit there and take that into consideration. Now you look at a particular use case, like for example making this decision. Have you done enough studies to know how good that prediction really is? And have you

done enough studies to compare it to say, well, what if we, what if we dug in in a more direct, you know, let's get the evidence. Let's, let's do the deductive thing and not use statistics here. How often would that have done better, right? So you have to do the studies to know how good the AI actually is. And it's complicated because it depends how fast you have to make decision.

So if you have to make the decision super fast, do you have no choice, right? If you have more time, right? But if you're ready to pull the plug and this is a lot of the argument that I had with a doctor, I said, what's he going to do if you do it? What's going to happen to him in that room? If you do it my way, you know, if you do, well, he's going to die anyway. So let's do it my way then.

I mean, it raises questions for our society to struggle with as with the case with your father, but also when things like race and gender start coming into play, when certain, when judgments are made based on things that are complicated in our society, at least in discourse. And it starts, you know, I think, I think I'm safe to say that most of the violent crimes committed by males. So if you discriminate based, you know, it's a male versus female saying that

if it's a male, more likely to commit the crime. So this is one of my very positive and optimistic views of why the study of artificial intelligence, the process of thinking and reasoning, logically and statistically and how to combine them is so important for the discourse today because it's causing a, regardless of what state AI devices are or not,

it's causing this dialogue to happen. This is one of the most important dialogues that in my view, the human species can have right now, which is how to think well, how to reason well, how to understand our own cognitive biases and what to do about them. That has got to be one of the most important things we as a species can be doing honestly. we as a species can be doing honestly. We have created an incredibly complex society. We've created amazing abilities to amplify noise

faster than we can amplify signal. We are challenged. We are deeply, deeply challenged. We have big segments of the population getting hit with enormous amounts of information. Do they know how to do critical thinking? Do they know how to objectively reason? Do they understand what they are doing,

never mind what their AI is doing? This is such an important dialogue to be having. And we are fundamentally, our thinking can be easily becomes fundamentally biased. And there are fundamentally, our thinking can be and easily becomes fundamentally biased. And there are statistics, and we shouldn't blind us, we shouldn't discard statistical inference, but we should understand the nature of statistical inference. As a society, as we decide to reject statistical inference,

to favor individual favor understanding and deciding on the individual. Yes We we consciously make that choice. So even if the statistics said Even if the statistics said males are more likely to have, you know, to be violent criminals. We still take each person as an individual and we treat them based on the logic and the knowledge of that situation. We purposefully and intentionally reject the statistical inference. We do that at a respect for the individual. Yeah, and that requires reasoning and correct thinking

Looking forward what grand challenges would you like to see in the future because The the jeopardy challenge You know captivated the world Alpha-go alpha-zero captivated the world debal-Go, Alpha-Zero, Cap-Date of the World, De-Baloo, certainly beating Casparov, Gary's bitterness aside, and Cap-Date of the World. What do you think, do you have ideas for next grand challenges for future challenges of that?

You know, I, look, I mean, I think there are lots of really great ideas for grand challenges. I'm particularly focused on one right now, which is, can you demonstrate that they understand, that they could read and understand, that they can acquire these frameworks and communicate, you know, reasoning communicate with humans. So it is kind of like the Turing test, but it's a little bit more demanding than a Turing test.

It's not enough, it's not enough to convince me that you might be human because you can power into conversation. I think the standard is a little bit higher. For example, can you, the standard is higher? I think one of the challenges of devising this grand challenge is that we're not sure what intelligence is, we're not sure how to determine whether or not two people actually understand each other and in what depth they understand it, to what depth they understand each other. So the challenge becomes something along the lines of, can you satisfy me that we have a shared understanding?

So if I were to probe and you probe me, can machines really act like thought partners where they can satisfy me, that we a share, our understanding is shared enough that we can collaborate and produce answers together and that they can help me explain and justify those answers. So maybe you hear an idea. So we'll have AI system run for president and convince.

That's too easy. I'm sorry, go ahead. No, you have to convince the voters that they should vote? So like, I guess what does winning again, I thought why I think this is such a challenge because we go back to the emotional persuasion. We go back to, you know, now we're checking off an aspect of human cognition that is in many ways weak or flawed, right? We're so easily manipulated. Our minds are drawn for often the wrong reasons, right? Not the reasons that ultimately mattered us, but the reasons that can easily persuade us. I think we can

be persuaded to believe one thing or another for reasons that ultimately don't serve us well in the long term. And a good benchmark should not play with those elements of emotional manipulation. I don't think so. And I think that's where we have to set the set the higher standard for ourselves of what you know what the mean, this goes back to rationality and it goes back to objective thinking and can you produce, can you acquire information and produce reasoned arguments and to those reasons arguments pass a certain amount of muster. And is it, and can you acquire new knowledge,

you know, can you, can you under, for example, can you reason, I have acquired new knowledge, can you identify where it's consistent or contradictory with other things you've learned? And can you explain that to me and get me to understand that? So I think another way to think about it perhaps is kind of machine teach you. Can I help you? That's a really nice, nice one to put it. Can I help you understand something that you didn't really understand before?

Where is it? It's taking you to, so you're not, again, it's almost like, can it, can it teach you? Can it help you learn? And, and in an arbitrary space, so it can open those domain space. So can you tell the machine, and again, this borrows from some science fiction. So but can you go off and learn about this topic that I'd like to understand better? And then work with me to help me understand it. That's quite brilliant.

What, the machine that passes that kind of test, do you think it would need to have self-awareness or even consciousness? What do you think about consciousness and the importance of it, maybe in relation to having a body, having a presence, an entity? Do you think that's important? You know, people used to ask me

if Watson was conscious, and I just think, I think they're conscious of what, exactly. I mean, I think, you know, it depends what it is that you're conscious of. I mean, so, you know, if you, you know, it's certainly easy for it to answer questions about, it would be trivial to program it. So answer questions about whether or not it was playing jeopardy. I mean, it could certainly answer questions that will imply that it was aware of things. Exactly. What does it mean to be aware and what does it mean

to conscious? It's sort of interesting. I mean, I think that we differ from one another based on what we're conscious of. But wait, wait, wait, yes, for sure. There's degrees of conscious in there. So it's just areas like it's not just degrees. What do you what do you will wear of like what are you not aware of it? But nevertheless there's a very subjective element to our experience. Let me even not talk about consciousness. Let me talk about another to me really interesting topic of mortality, fear or mortality. The Watson, as far as I could tell, did not have a fear of death. Certainly not. Most, most humans do. It wasn't conscious of death. It wasn't, it was

not. So there's an element of finiteness to our existence that I think, like we, like you mentioned, survival that adds to the whole thing. I mean, consciousness is tied up with that, that we are a thing. It's a subjective thing that ends. And that seems to add a color and flavor to our motivations in a way that seems to be fundamentally important for intelligence, or at least the kind of human intelligence.

Well, I think for generating goals, again, I think you could have, you could have an intelligence capability and a capability to learn and capability to predict, but I think without, I mean, again, you get a fear, but essentially without the goal to survive. So you think you can just encode that without having to really. I think you can. I mean, you can create a robot now and you could say, you know, plug it in and say, protect your power source, you know, and give it some capabilities and we'll sit there

and operate to try to protect this power source and survive. I mean, I, so I don't know that that's philosophically a hard thing to demonstrate. It sounds like a fairly easy thing to demonstrate that you can give it that goal. We'll come up with that goal by itself. And I think you have to program that goal in. But there's something because I think as we touched on intelligence is kind of like a social construct.

The fact that a robot will be protecting its power source would add depth and grounding to its intelligence in terms of us being able to respect it. I mean, ultimately, it boils down to us acknowledging that it's intelligent. And the fact that it can die, I think, is an important part of that. The interesting thing to reflect on is how trivial that would be. And I don't think if you knew how trivial that was, you would associate that with being intelligence. I mean, I literally put in a statement of code that says, you know, you have the following actions you can take, you give it a bunch of actions, like, maybe you mount a laser gun

on it or you make your ability to scream and to screech or whatever. And you say, if you see your power source threatened and you could program that in, and you're gonna take these actions to protect it. You know, you teach it, train it on a bunch of things. So, and now you can look at that and you say, well, that's intelligence,

which is protecting the power source. Maybe, but that's again at this human bias that says, the thing I identify, my intelligence and my conscious, so fundamentally with the desire or at least the behavior is associated with the desire to survive, that if I see another thing doing that, I'm going to assume it's intelligence. What timeline year will society have something that you would be comfortable calling

an artificial general intelligence system? Well, what's your intuition? Nobody can predict the future, certainly not next few months or 20 years away, but what's your intuition? How far away are we? I don't know. I just want to make these predictions. I mean, I would be, you know, I would be guessing. And there's so many different variables, including just how much we want to invest in it and how important it, you know, and how important we think it is.

What kind of investment we're willing to make in it? What kind of talent we end up bringing to the table, all, you know, the incentive structure, all these things. So I think it is possible to do this sort of thing. I think it's, I think trying to sort of ignore many of the variables and things like that. Is it a 10 year thing? Is it a 23? It's probably closer to a 20 year thing, I guess. But not several hundred years. No, I don't think it's several hundred years. I don't think it's several hundred years. But again, so much depends on how committed we are to investing and incentivizing this type of work, this type of work. And it's sort of interesting. Like I don't, I don't think it's obvious.

How incentivized we are. I think from a task perspective, if we see business opportunities to take this technique or that technique to solve that problem, I think that's the main driver for many of these things. From a general intelligence kind of an interesting question. Are we really motivated to do that and? Like we just struggled ourselves right now to even define what it is So it's hard to incentivize when we don't even know what it is we're incentivized to create and if you said mimic a human intelligence I just think there are so many Challenges with the the significance and meaning of that. There's not a clear directive.

There's no clear directive to do precisely that thing. So assistance in larger and larger number of tasks. So being able to a system that's particularly able to operate my microwave and making a grilled cheese sandwich, I don't even know how to make one of those. And then the same system would be doing the vacuum cleaning and then the same system would be doing the vacuum cleaning. And then the same system would be teaching my kids that I don't have math. I think that went went when you get into a general intelligence for learning physical tasks. And again, I want to go back to your body questions. I think your body question was interesting, but you want to go back to your body question, because I think your body question was interesting, but You want to go back to you know learning abilities to physical tasks. You might have

We might get I imagine in that time frame We will get better and better at learning these kinds of tasks whether it's mowing your lawn or driving a car or whatever it is I think we will get better and better at that where it's learning how to make predictions over large bodies of data I think we're gonna continue to get better and better at that, where it's learning how to make predictions over large bodies of data. I think we're going to continue to get better and better at that. And machines will out, you know, outpace humans and a variety of those things. The underlying mechanisms for doing that may be the same, meaning that, you know, maybe these are deep nets. There's infrastructure to train them, reusable components to get them to do different classes

of tasks, and we get better and better at building these kinds of machines. You could see argued that the general learning infrastructure in there is a form of a general type of intelligence. I think what starts getting harder is this notion of, can we effectively communicate and understand and build that shared understanding because of the layers of interpretation that are required to do that and the need for the machine to be engaged with humans at that level in a continuous basis. So how do you get in there?

How do you get the machine in the game? How do you get the machine in the intellectual game? Yeah, and to solve a GI, you probably have to solve that problem. You have to get the machine in the intellectual game? Yeah. And to solve a GI, you probably have to solve that problem. You have to get the machine. So it's a little bit of a bootstrapping thing. Can we get the machine engaged in the intellectual, like a colonial game? But in the intellectual dialogue with the humans, are the humans sufficiently an intellectual dialogue with each other to generate enough data in this context. How do you bootstrap that?

Because every one of those conversations, every one of those conversations, those intelligent interactions require so much prior knowledge that is the challenge to bootstrap it. The question is, how committed, so I think that's possible, but when I go back to, are we incentivized to do that? I know one incentivized to do the former. Are we incentivized to do the latter significantly enough? Do people understand what the latter really is well enough? Part of the elemental cognition mission is to try to articulate that better and better through demonstrations and through trying to craft these grand challenges and get people

to say, look, this is a class of intelligence. This is a class of AI. Do we want this? What is the potential of this? What's the business potential? What's the society of the potential to that? And to build up that incentive system around that? Yeah, I think if people don't understand yet, I think they will.

I think there's a huge business potential here So it's exciting that you're working on it Kind of skipped over but I'm a huge fan of physical presence of things. Do you think? You know Watson had a body. Do you think? having a body as to the interactive element between the AI system and a human or just in general to intelligence? So I think going back to that shared understanding bit, humans are very connected to their bodies.

I mean, one of the reasons, one of the challenges in getting an AI to kind of be a compatible human intelligence is that our physical bodies are generating a lot of features that make up the input. So in other words, our bodies are the tool we use to affect output, but they also generate a lot of input for our brains. So we generate emotion, we generate all these feelings, we generate all these signals that machines don't have. So it's machines don't have this, it's the input data.

And they don't have the feedback. It says, okay, I've gotten this, I've gotten this emotion, or I've gotten this idea, I now want a process set, and then affects me as a physical being, and then I can then affects me as a physical being and then I and I can play that out. In other words, I could realize the implications of that.

So implications again on my mind, mind, body complex. I then process that and the implications again are internal features are generated. I learn from them. They have an effect on my mind body complex. So it's interesting when we think do we want a human intelligence? Well, if we want a human compatible intelligence, probably the best thing to do is to embed it

in a human body. Just to clarify, and both concepts are beautiful, is a humanoid robot. So a robot that look like humans is one, or did you mean actually sort of what Elon Musk was working with your link, really embedding intelligence systems to write along human bodies? No, writing along is different. I meant like if you want to create an intelligence that is human compatible,

meaning that it can learn and develop a shared understanding of the world around it. You have to give it a lot of the same substrate. Part of that substrate is the idea that it generates these kinds of internal features, like sort of emotional stuff, it has similar senses, it has to do a lot of the same things with those same senses. So I think if you want that, again,

I don't know that you want that. Like, yeah, like that's not my specific goal. I think that's a fascinating scientific goal. I think it has all kinds of other implications. That's sort of not the goal. I want to create, I think if it has like create insulutral thought partners for humans, that kind of intelligence.

I know there are other companies that are creating physical thought partners, physical partners for you, but that's kind of not where I'm at. But the important point is that a big part of what we process is that physical experience of the world around us? On the point of thought partners, what role does an emotional connection or forgive me, love, have to play in that thought partnership? Is that something you're interested in, put another way, sort of having a deep connection beyond intellectual.

With the AI? Yeah, with the AI between human and AI. Is that something that gets in the way of the rational discourse? Is that something that's useful? I worry about biases, obviously. So in other words, if you develop an emotional relationship with the machine, do all of a sudden you start

are more likely to believe what it's saying even if it does make any sense. So I you know, I worry about that But at the same time I think the opportunity to use machines to provide human companionship is actually not crazy and To intellectual and social companionship is not crazy idea Do you have concerns as a few people do? Elon Musk, Sam Harris, about long-term existential threats of AI and perhaps short-term threats of AI? We talked about bias, we talked about different misuses, but do you have concerns about thought

partners, systems that are able to help us make decisions together with humans, somehow having a significant negative impact on society in the long term. I think there are things to worry about. I think the giving machines too much leverage is a problem. What I mean by leverage is too much control over things that can hurt us, whether it's socially, psychological, intellectually, or physically, and if you give the machines too much control, I think that's a concern. You forget about the AI,

just when you give them too much control, human bed actors can hack them and produce havoc. So, that's a problem, and you imagine hackers taking over the driverless car network and, you know, creating all kinds of havoc. But you could also imagine given the ease that humans could repersuade in one way or the other, and now we have algorithms that can easily take control over that, over that and amplify noise and move people one direction or another.

I mean, humans do that, other humans all the time. And we have marketing campaigns, we have political campaigns that take advantage of our emotions or our fears. And this is done all the time. But with machines, machines are like giant megaphones, right? We can amplify this in orders of magnitude and fine-tune its control. So we can tell her the message. We can now vary rapidly and efficiently tell her the message to the audience, taking advantage of their biases and amplifying them and using them to pursue

a way them in one direction or another in ways that are not fair, not logical, not objective, not meaningful, and humans, machines, and power to that. So that's what I mean by leverage. It's not new, but wow, it's powerful because machines can do it more effectively, more quickly, and we see that already going on in social media and other places. And other places. That's scary.

And that's why I'm, that's why I go back to saying, one of the most important public dialogues we could be having is about the nature of intelligence and the nature of inference and logic and reason and rationality and us understanding our own biases, us understanding our own cognitive biases and how they work and then how machines work and how do we use them to complement it basically so that in the end we have a stronger overall system. That's just incredibly important.

I don't think most people understand that. So like telling your kids or telling your students, this goes back to the cognition. Here's how your brain works. Here's how easy it is to trick your brain. Right, There are fundamental cognitive, but you should appreciate the different, the different types of thinking and how they work and what you're prone to and, you know, and what

and what do you prefer? And on the what conditions does this make sense versus that makes sense? And then say, here's what AI can do. Here's how it can make this worse. And here's how it can make this better. And that's where the AI has a role is to reveal that trade off. So if you imagine a system that is able to beyond any definition of the touring test of the benchmark, really an AGI system as a thought partner,

that you one day will create. What question, what topic of discussion, if you get to pick one, would you have with that system? What would you ask, and you get to find out the truth together? So you threw me a little bit was finding the truth at the end, but Because the truth is all another topic, but the I think the beauty of it. I think what excites me is the beauty of it is if I really have that system I don't have to pick.

So in other words, I can go to and say, this is what I care about today. And that's what we mean by this general capability. Go out, read this stuff in the next three milliseconds. And I want to talk to you about it. I want to draw analogies, I want to understand how this affects the decision, that decision, what if this were true, what if that were true, what knowledge should I be aware of that

could impact my decision? Here's what I'm thinking is the main implication. Can you find, can you prove that out? Can you give me the evidence that supports that? Can you give me evidence that supports this other thing? Boy, that would that be incredible. Would that be just incredible? Just a long discourse.

To be just just to be part of whether it's a medical diagnosis or whether it's, you know, the various treatment options or whether it's a legal case or whether it's a social problem that people are discussing, like be part of the dialogue. One that holds itself and us accountable to reasons and objective dialogue. You know, I just, I could goosebumps talking about it, right? I mean, it's like this is what I want. So when you created, please come back on the podcast and we can have a discussion together and make it even longer.

This is a record for the longest conversation. Other is an honor. It was a pleasure, David. Thank you so much. Thanks so much. A lot of fun. you