David Ferrucci: IBM Watson, Jeopardy & Deep Conversations with AI

[0] The following is a conversation with David Ferrochi.

[1] 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.

[2] 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.

[3] Where science meets engineering is where brilliant, simple ingenuity emerges.

[4] People who work adjoining it to have a lot of wisdom earned through failures and eventual success.

[5] 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.

[6] This is the Artificial Intelligence Podcast.

[7] If you enjoy it, subscribe on YouTube, give it five stars on iTunes, support it on Patreon, or simply connect with me on Twitter, at Lex Friedman, spelled FRID, M -A -N.

[8] And now here's my conversation with David Ferruji.

[9] Your undergrad was in biology with an eye toward medical school before you went on for the PhD in computer science.

[10] So let me ask you an easy question.

[11] What is the difference between biological systems and computer systems in your, when you sit back, look at the stars and think philosophically.

[12] I often wonder, I often wonder whether or not there is a substantive difference.

[13] I mean, I think the thing that got me into computer science and into artificial intelligence was exactly this presupposition 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, 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 sure I'm convinced that there 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 what we can experience from an intellectual and physical perspective I'm not sure there is.

[14] Clearly there are implement there are different implementations but if you were to say is a biological information processing system fundamentally more capable than one we might be able to build out of silicon or 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?

[15] Yeah, that's interesting.

[16] I mean, so, you know, your question presupposes that there's this goal to recreate, you know, what we perceive as biological intelligence.

[17] I'm not, I'm not sure that's the, I'm not sure that that's how I would state the goal.

[18] I mean, I think that's studying, good.

[19] So I think there are a few goals.

[20] 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 intellectuals.

[21] actual, psychological, and physical.

[22] So neuroscience and understanding the brain from that perspective, there's a clear, clear goal there.

[23] From the perspective of saying, I want to, I want to, I want to mimic human intelligence.

[24] That one's a little bit more interesting.

[25] Human intelligence certainly has a lot of things we envy.

[26] It's also got a lot of problems too.

[27] 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?

[28] How do we want to communicate with that intelligence?

[29] How do we want it to behave?

[30] How do we want it to perform?

[31] 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 the strengths 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?

[32] And that in and of itself suggests that, human intelligence is something quite enviable, that it could, you know, it can, it can, it can, it can, it can, um, introspect that way.

[33] And the flaws, you mentioned the flaws.

[34] So humans have, yeah, and I think, I think that flaws that human intelligence has is extremely, um, prejudicial and bias in the way it draws many inferences.

[35] Do you think those are, sorry to interrupt, do you think those are features or are those bugs?

[36] Do you think the, the prejudice, the forgetfulness, the fear what other flaws list them all what love maybe that's a flaw 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 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.

[37] 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, you know, objective reasoning perspective, you're going to favor them because those are the decisions or those are the paths that succeeded in the past.

[38] 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.

[39] And I think those create problems when you want to think more deeply and make more objective and reasoned decisions.

[40] Of course, humans capable of doing both.

[41] They do one more naturally than they do the other, but they're capable of doing both.

[42] You're saying they do the one that responds quickly more naturally?

[43] Right.

[44] Because that's the thing we kind of need to not be eaten by the predators in the world.

[45] For example, but then we've learned to reason through logic.

[46] We've developed science.

[47] We train people to do that.

[48] I think that's harder for the individual to do.

[49] I think it requires training and teaching.

[50] I think we are, the human mind certainly is capable of it, but we find it more difficult.

[51] 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 losing your way.

[52] So just focus and.

[53] 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.

[54] 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 jeopardy and beyond.

[55] How do you think about intelligence?

[56] Is it a sufficiently complicated problem being able to reason your way through solving that problem?

[57] Is that kind of how you think about what it means to be intelligent?

[58] So I think of intelligence, primarily two ways.

[59] One is the ability to predict.

[60] So in other words, if I have a problem, can I predict what's going to 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 going to tell you what's going to happen next, or you're in a room and somebody walks in and you're going to predict what they're going to do next or what they're going to say next.

[61] So in a highly dynamic environment full of uncertainty, be able to predict.

[62] You know, the more variables, the more complex, the more possibilities, the more complex.

[63] But can I take a small amount of prior data?

[64] and learn the pattern and then predict what's going to happen next accurately and consistently.

[65] That's certainly a form of intelligence.

[66] What do you need for that, by the way?

[67] You need to have an understanding of the way the world works in order to be able to unroll it into the future, right?

[68] What do you think is needed to predict?

[69] Depends what you mean by understanding.

[70] I need to be able to find that function, and this is very much what's a function.

[71] 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.

[72] And if I can predict it accurately, so I can get it right more often than not, I'm smart.

[73] If I can do that with less data and less training time, I'm even smarter.

[74] 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.

[75] What about picking a goal?

[76] So that's interesting about picking a goal, sort of an interesting thing.

[77] I think that's where you bring in, what are you pre -programmed to do?

[78] We talk about humans, and well, humans are pre -programmed to survive.

[79] So it's sort of their primary, you know, driving goal.

[80] What do they have to do to do that?

[81] And that can be very complex, right?

[82] So it's not just figuring out that you need to run away from the ferocious tiger, but we survive in a social context.

[83] as an example.

[84] So understanding the subtleties of social dynamics becomes something that's important for surviving, finding a mate, reproducing.

[85] So we're continually challenged with complex sets of variables, complex constraints, rules, if you will, or patterns.

[86] And we learn how to find the functions and predict the things.

[87] In other words, represent those patterns efficiently and be able to predict what's going to happen.

[88] And that's a form of intelligence.

[89] That doesn't really require anything specific other than the ability to find that function and predict that right answer.

[90] It's certainly a form of intelligence.

[91] But then when we say, well, do we understand each other?

[92] In other words, would you perceive me as intelligent beyond that ability to predict?

[93] 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, if you did not have, for example, the right pattern matching machinery that I did.

[94] And now we potentially have this breakdown where in effect, I'm intelligent, but I'm sort of an alien intelligence relative to you're intelligent but nobody knows about it or i can see the i can see the output like so 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 and the second is me being able to 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 impressing you that i'm In other words, you may be convinced that I'm intelligent in some form.

[95] So how would convince?

[96] Because of my ability to predict.

[97] So I would look at the metrics.

[98] I said, wow, you're right all.

[99] You're right more times than I am.

[100] You're doing something interesting.

[101] That's a form of intelligence.

[102] But then what happens is if I say, how are you doing that?

[103] And you can't communicate with me. And you can't describe that to me. Now I may label you a savant.

[104] 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.

[105] And so now, so this is interesting, right?

[106] 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.

[107] And then my, we start to understand each other and then my respect and my appreciation, my ability to relate to you starts to change.

[108] So now you're not an alien intelligence anymore.

[109] You're a human intelligence now because you and I can communicate.

[110] And so I think when we look at, when we look at animals, for example, animals can do things.

[111] We can't quite comprehend.

[112] We don't quite know how they do them, but they can't really communicate with us.

[113] They can't put what they're going through in our terms.

[114] and so we think of them as sort of low there are these alien intelligences and they're not really worth necessarily what we're worth, we don't treat them the same way as a result of that but it's hard because who knows what you know 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 we're looking Look at humans.

[115] Look at political debates and discourse on Twitter.

[116] It's mostly just telling stories.

[117] So you, your task is, sorry, 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.

[118] Ultimately, that's what a proof is.

[119] I mean, even a mathematical proof is that.

[120] Because ultimately the other mathematicians have to be convinced by your proof.

[121] Otherwise, in fact, there have been...

[122] That's the metric of success, yeah.

[123] 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?

[124] You never know if it proved anything until the community of mathematicians decided that it did.

[125] So, I mean, so it's, but it's a real thing, right?

[126] And that's sort of the point, right?

[127] Is that ultimately, you know, this notion of understanding us, understanding something, is ultimately a social concept.

[128] In other words, I have to convince enough people that I did this in a reasonable way.

[129] I did this in a way that other people can understand and replicate and that it makes sense to them.

[130] So our human intelligence is bound together in that way.

[131] We're bound up in that sense.

[132] We sort of never really get away with it until we can sort of convince others that our thinking process, you know, makes sense.

[133] So do you think the general question of intelligence is then also a social construct?

[134] So if we task, ask questions of an artificial intelligence system, is this system intelligent?

[135] The answer will ultimately be a socially constructed concept.

[136] So I think, so I think, you know, I'm making two statements.

[137] I'm saying we can try to define intelligence in a super objective way that says, here's this data.

[138] I want to predict this type of thing.

[139] learn this function and then if you get it right often enough we consider you intelligent but that's more like a savant 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 that you're making can we make those decisions ourselves can relate to the process that you're going through.

[140] 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 or obviously a community or judge of people to decide whether or not, whether or not that makes sense.

[141] And by the way, that happens with the humans as well.

[142] You're sitting down with your staff, for example, and you ask for suggestions about what to do next, and someone says, oh, I think you should buy, and I should think you should buy this much or whatever, or sell, 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.

[143] And it's not, you're not very satisfied.

[144] 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.

[145] Can you explain to me why I should believe this?

[146] Right.

[147] And that explanation may have nothing to do with the truth.

[148] It's just the ultimate...

[149] It's got to convince the other person.

[150] It's still be wrong.

[151] She's got to be convincing.

[152] But it's ultimately got to be convinced.

[153] And that's why I'm saying it's we're bound together, right?

[154] Our intelligences are bound together in that sense.

[155] We have to understand each other.

[156] And if for example, you're giving me an explanation.

[157] I mean, this is a very, very important point, right?

[158] You're giving me an explanation.

[159] And I'm, and I, and I have, I, I, I'm not good.

[160] And then I'm not good at reasoning well and being objective and following logical paths and consistent paths.

[161] And I'm not good at measuring and sort of computing probabilities across those paths.

[162] What happens is collectively, we're not going to do, we're not going to do well.

[163] How hard is that problem?

[164] The second one.

[165] So I think we'll talk quite a bit about the first on a specific, objective, metric benchmark performing well.

[166] But being able to explain the steps, the reasoning, how hard is that problem?

[167] I think that's very hard.

[168] I mean, I think that that's, well, it's hard for humans.

[169] The thing that's hard for humans, as you know, may not necessarily be hard for computers and vice versa.

[170] So, sorry, so how hard is that problem for computers?

[171] I think it's hard for computers.

[172] And the reason why I related to, or saying that it's also hard for humans, is because I think when we step back and we say we want to design computers to do that, one of the things we have to recognize is we're not sure how to do it well.

[173] I'm not sure we have a recipe for that.

[174] And even if you wanted to learn it, it's not clear exactly what data we use and what judgments we use to learn that well.

[175] And so what I mean by that is if you look at the entire enterprise of science, science is supposed to be at a bad objective reason and reason right so we think about 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'd 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 proves and use the scientific method and I think it's the latter.

[176] And my point is that how do you train someone to do that?

[177] And that's what I mean by it's hard.

[178] How do you, what's the process of training people to do that well?

[179] That's a hard process.

[180] We work as a society, we work pretty hard to get other people to understand our thinking and to convince them of things.

[181] Now we could persuade them, obviously you talked about this, like human flaws or weaknesses, we can persuade them through emotional means.

[182] But to get them to understand and connect to and follow a logical argument is difficult.

[183] We try it, we do it as scientists, we try to do it as journalists.

[184] We, you know, we try to do it as, you know, even artists in many forms as writers, as teachers.

[185] We go through a fairly significant training process to do that.

[186] And then we could ask, well, why is that so hard?

[187] But it's hard.

[188] And for humans, it takes a lot of work.

[189] And when we step back and say, well, how do we get a machine to do that?

[190] It's a vexing question.

[191] How would you begin to try to solve that?

[192] 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.

[193] But if you look at algorithms that recommend things that we look at next, whether it's Facebook, Google, advertisement -based companies, you know, their goal is to convince you to buy things based on anything.

[194] 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.

[195] The algorithm that describes why a certain reason, a certain decision was made, how hard is it to do it through emotional manipulation?

[196] And why is that a good or a bad thing?

[197] So you've kind of focused on reason, logic, really showing in a clear way why something is good.

[198] One, is that even a thing that us humans do?

[199] And two, how do you think of the difference in the reasoning aspect and the emotional manipulation?

[200] 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.

[201] and kind of give you more of that stuff.

[202] Manipulation is a bad word.

[203] Yeah, I mean, I'm not saying it's right or wrong.

[204] It works to get your attention and it works to get you to buy stuff.

[205] And when you think about algorithms that look at the patterns of the patterns of features that you seem to be spending your money on and say, I'm going to give you something with a similar pattern.

[206] So I'm going to learn that function because the objective is to get you to click on it or get you to buy it or whatever it is.

[207] I don't know.

[208] I mean, that it is like it is what it is.

[209] I mean, that's what the algorithm does.

[210] You can argue whether it's good or bad.

[211] It depends what your goal is.

[212] I guess this seems to be very useful for convincing, for telling a story.

[213] For convincing humans, it's good because, again, this goes back to what is the human behavior like?

[214] What does the human brain respond to things?

[215] 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.

[216] And these algorithms are saying, look, that's up to you.

[217] The reason whether you need something or not, that's your job.

[218] 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.

[219] And the algorithms are saying, hey, that's like whatever.

[220] Like, that's your problem.

[221] All I know is you're buying stuff like that.

[222] You're interested in stuff like that.

[223] Could be a bad reason, could be a good reason.

[224] That's up to you.

[225] I'm going to show you more of that stuff.

[226] And so, and I, and I think that that's, it's not good or bad.

[227] It's not reasoned or not reason.

[228] The algorithm is doing what it does, which is saying, you seem to be interested in this.

[229] I'm going to show you more of that stuff.

[230] And I think we're seeing this not just in buying stuff, but even in social media, you're reading this kind of stuff.

[231] I'm not judging on whether it's good or bad.

[232] I'm not reasoning at all.

[233] I'm just saying, I'm going to show you other stuff with similar features.

[234] And, you know, and like, in that's it.

[235] And I wash my hands from it.

[236] And I say, that's all, you know, that's all that's going on.

[237] You know, there is, people are so harsh on AI systems.

[238] So one, the bar of performance is extremely high.

[239] And yet, we also ask them to, in the case of social media, to help find the better angels of our nature and help make a better society.

[240] So what do you think about the role of AI?

[241] So that, I agree with you, that's, that's the interesting dichotomy, right?

[242] Because Because on one hand, we're sitting there and we're sort of doing the easy part, which is finding the patterns.

[243] We're not building a, the system's not building a theory that it's consumable and understandable by other humans that can be explained and justified.

[244] And so on one hand to say, oh, you know, AI is doing this, why isn't doing this other thing?

[245] Well, those other things are a lot harder.

[246] And it's interesting to think about why it's harder.

[247] and because you're interpreting, you're interpreting the data in the context of prior models.

[248] In other words, understandings of what's important in the world, what's not important.

[249] What are all the other abstract features that drive our decision making?

[250] What's sensible?

[251] What's not sensible?

[252] What's good?

[253] What's bad?

[254] What's moral?

[255] What's valuable?

[256] What isn't?

[257] Where is that stuff?

[258] No one's applying the interpretation.

[259] 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.

[260] Like, I don't know why you're clicking on the stuff you're clicking, or if it's products, what the price is, or what the categories and stuff like that.

[261] And I just feed you more of the same stuff.

[262] That's very different than kind of getting in there and saying, what does this mean?

[263] what the stuff you're reading like why are you reading it what assumptions are you bringing to the table are those assumptions sensible uh does the material make any sense does it does it lead you to thoughtful good conclusions again there's judgment there's interpretation judgment involved in that process that isn't really happening in in the AI today that's harder right because you have to start getting at the meaning of the style of the content.

[264] You have to get at how humans interpret the content relative to their value system and deeper thought processes.

[265] 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.

[266] So it's a Again, even meaning is a social construct.

[267] So you have to try to predict how most people would understand this kind of statement.

[268] Yeah, meaning is often relative, but meaning implies that the connections go beneath the surface of the artifacts.

[269] If I show you a painting, it's a bunch of colors on a canvas, what does it mean to you?

[270] And it may mean different things of different people because of their different experiences.

[271] It may mean something even different to the artist who painted it.

[272] As we try to get more rigorous with our communication, we try to really nail down that meaning.

[273] So we go from abstract art to precise mathematics, precise engineering drawings and things like that.

[274] We're really trying to say, I want to narrow that space of possible interpretations because the precision of the communication ends up becoming more and more important.

[275] And 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 brushstrokes on a canvas or pixels in a photograph, you can sit there and you can interpret lots of different ways at many, many different levels.

[276] But when I want to, when I want to align our understanding of that, I have to specify a lot more stuff that's actually not in, not directly in the artifact.

[277] Now, I have to say, well, how are you interpreting this image and that image?

[278] And what about the colors and what do they mean to you?

[279] What perspective are you bringing to the table?

[280] What are your prior experiences with those artifacts?

[281] What are your fundamental assumptions and values?

[282] 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.

[283] And if that's the case, then I would conclude that.

[284] 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.

[285] Getting and sick of that is hard.

[286] And yet humans are able to intuit some of that without any.

[287] pre because they have the shared experience me and we're not talking about shared two people having shared experience me as a society that's correct we have the shared 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 education we have similar what we like to call prior models about the word 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.

[288] And so we take that as our background, and we start interpreting things similarly.

[289] But as humans, we have a lot of shared experience.

[290] We do have similar brains, similar goals, similar emotions under similar circumstances, because we're both humans.

[291] So now one of the early questions you asked, well, how is biological and, you know, computer information systems fundamentally different?

[292] Well, One is humans come with a lot of pre -programmed stuff, a ton of program stuff, and they're able to communicate because they have a lot of, because they share that stuff.

[293] 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?

[294] Do you have a hope?

[295] Can you speak to?

[296] How hard is it to encode that knowledge systematically in a way that could be used by a computer?

[297] So I think it is possible to learn for a machine, to program a machine, to acquire that knowledge with a similar foundation, in other words, a similar interpretative foundation for process.

[298] processing that knowledge.

[299] What do you mean by that?

[300] So in other words, we view the world in a particular way.

[301] So in other words, we have a, if you will, as humans, we have a framework for interpreting the world around us.

[302] So we have multiple frameworks for interpreting the world around us.

[303] 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, The goals are largely built around survival and quality of life.

[304] They're fundamental economics around scarcity of resources.

[305] And when humans come and start interpreting a situation like that, because you brought up historical events, they start interpreting situations like that.

[306] They apply a lot of this fundamental framework for interpreting that.

[307] Well, who are the people?

[308] What were their goals?

[309] What resources did they have?

[310] how much power influence that they have over the other, like this fundamental substrate, if you will, for interpreting and reasoning about that.

[311] So I think it is possible to imbue a computer with that stuff that humans like take for granted when they go and sit down and try to interpret things.

[312] And then with that foundation, they acquire.

[313] They start acquiring the details, the specifics and in a given situation, are then able to interpret it with regard to that framework, and then given that interpretation, they can do what?

[314] They can predict.

[315] 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.

[316] Now, you could find humans that come and interpret events very differently than other humans because they're like using a different framework.

[317] You know, the movie Matrix comes to mind where, you know, they decided humans were really just batteries, and that's how they interpreted the value of humans as a source of electrical energy.

[318] But I think that, you know, for the most part, we have a way of interpreting the events or these social events around us, because we have this shared framework.

[319] 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.

[320] These frameworks make sense to us.

[321] So how much knowledge is there, do you think?

[322] So you said it's possible.

[323] Well, there's a tremendous amount of detailed knowledge in the world.

[324] You can imagine, you know, effectively infinite number of unique situations and unique configurations of these things.

[325] But the knowledge that you need, what I referred to as like the frameworks for you need for interpreting them, I don't think.

[326] I think that's those are finite.

[327] You think the frameworks are more important than the bulk of the now.

[328] So like framing.

[329] Yeah, because what the frameworks do is they give you now the ability to interpret and reason and to interpret and reason to interpret and reason over the specifics in ways that other humans would understand.

[330] What about the specifics?

[331] You know, you acquired the specifics by reading and by talking to other people.

[332] So I'm mostly actually just even if you can focus on even 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 that.

[333] Every time we try to do that kind of thing in robotics or AI, it seems to be like an onion.

[334] You seem to realize how much knowledge is really required to perform even some of these basic tasks.

[335] Do you have that sense as well?

[336] And so how do we get all those details?

[337] Are they written down somewhere?

[338] Do they have to be learned through experience?

[339] So I think when, like if you're talking about sort of the physics, the basic physics around us, for example, acquiring information about acquiring how that works.

[340] Yeah, I mean, I think there's a combination of things going.

[341] I think there's a combination of things going on.

[342] I think there is like fundamental pattern matching, like what we were talking about before.

[343] or you see enough examples, enough data about something, you start assuming that.

[344] And with similar input, I'm going to predict similar outputs.

[345] You can't necessarily explain it at all.

[346] You may learn very quickly that when you let something go, it falls to the ground.

[347] That's such a...

[348] But you can't necessarily explain that.

[349] But that's such a deep idea that if you let something go, like the idea of gravity.

[350] I mean, people are letting things go and counting on them falling, before they understood gravity.

[351] But that seems to be, that's exactly what I mean, is before you take a physics class or study anything about Newton, just the idea that stuff falls to the ground and then you be able to generalize that all kinds of stuff falls to the ground.

[352] It just seems like a non, without encoding it, like hard -coding it in, it seems like a difficult thing to pick up.

[353] 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.

[354] So both know that stuff falls to the ground and start to reason about social political discourse.

[355] So both like the very basic and the high level reasoning decision making.

[356] I guess so a question is how hard is this problem?

[357] and sorry to linger in it because again and we'll get to it for sure as what 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 outsider's 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?

[358] So, I think it's solvable.

[359] Okay, now beautiful.

[360] So what about time travel?

[361] Okay, on that time, I'm just the same answer.

[362] Not as convinced.

[363] Not as convinced yet, okay.

[364] No, I think it is, I think it is solvable.

[365] I mean, I think that it's a, first of all, it's about getting machines to learn.

[366] Learning is fundamental.

[367] And I think we're already in a place that we understand, for example, how machines can learn in various ways.

[368] Right now, our learning stuff is sort of primitive in that we haven't sort of taught machines to learn the frameworks.

[369] We don't communicate our frameworks because of how shared there in some cases we do.

[370] But we don't annotate, if you will, all the data in the world with the frameworks that are inherent or underlying.

[371] our understanding.

[372] Instead, we just operate with the data.

[373] 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.

[374] I think this can be done.

[375] I think we can start, I think machine learning, for example, with enough examples, can start to learn these basic dynamics.

[376] Will they relate them necessarily 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?

[377] 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.

[378] In terms of encoding architectures like that, do you think systems that were able to do this will look like neural networks?

[379] or representing, if you look back to the 80s and 90s of the expert systems, so they're 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.

[380] 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?

[381] Yeah, so I mean, I think asking the question that they look like neural networks is a bit of a red herring.

[382] I mean, I think that they will certainly do inductive or pattern -match -based reasoning.

[383] And I've already experimented with architectures that combine both that use machine learning and neural networks to learn certain classes of knowledge, in other words, to find repeated patterns in order for it to make good inductive guesses.

[384] But then ultimately to try to take those learnings and marry them, in other words, connect them to frameworks so that it can then reason over them.

[385] that in terms other humans understand.

[386] So, for example, at elemental cognition, we do both.

[387] 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 they can reason over it.

[388] So there is a fundamental knowledge representation, like you're saying like these graphs of logic, if you will.

[389] There are also neural networks that acquire a certain class of information, they then they align them with these frameworks.

[390] But there's also a mechanism to acquire the frameworks themselves.

[391] Yeah, so it seems like the idea of frameworks requires some kind of collaboration with humans.

[392] Absolutely.

[393] So do you think of that collaboration as well, and let's be clear, let's be a machine designing and intelligence that can ultimately communicate with humans in the terms of frameworks that help them understand things.

[394] So now, to be really clear, you can independently create a machine learning system and an 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.

[395] That doesn't mean it might be better than you at the thing.

[396] it might be that you cannot comprehend the framework that it may have created for itself that is inexplicable to you.

[397] That's a reality.

[398] But you're more interested in a case where you can.

[399] I am.

[400] My sort of approach to AI is because I've set the goal for myself.

[401] I want machines to be able to ultimately communicate understanding with humans.

[402] I want to be able to acquire and communicate, acquire knowledge from humans and communicate knowledge to humans, they should be using what, you know, 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 to work with humans to connect them to frameworks, interpretations, if you will, that ultimately make sense to humans.

[403] Of course, the machine is going to have the strength that it has the richer and longer memory, it has the more rigorous reasoning abilities, the deeper reasoning abilities.

[404] So it'll be an interesting, you know, complementary relationship between the human and the machine.

[405] Do you think that ultimately needs explainability like a machine?

[406] So if we look, we study, for example, Tesla autopilot a lot, where humans, I don't know if you've driven the vehicle are aware of what...

[407] 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, you know, the system fails every few miles.

[408] And so there's hundreds of, there's millions of those failures a day.

[409] And so that's like a moment of interaction.

[410] Do you see...

[411] Yeah, that's exactly right.

[412] That's a moment of interaction where you know, 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.

[413] So I believe that the collaboration between human and 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, I'm reading this thing.

[414] I know that the next word might be this or that, but I don't really understand why.

[415] I have my guess.

[416] Can you help me understand the framework that supports this and then can kind of acquire that, take that, and reason about it, and reuse it.

[417] The next time it's reading to try to understand something.

[418] Not unlike a human student might do.

[419] I mean, I remember when my daughter was in first grade and she had a reading assignment, about electricity.

[420] And, you know, somewhere in the text it says, and electricity is produced by water flowing over turbines or something like that.

[421] And then there's a question that says, well, how is electricity created?

[422] And so my daughter comes to me and says, I mean, I could, you know, created and produced are kind of synonyms in this case.

[423] So I can go back to the text and I can copy by water flowing over turbines.

[424] But I have no idea what that means.

[425] Like, I don't know how to interpret water flowing over turbines and what electricity even in.

[426] I mean, I can get the answer right by matching the text.

[427] But I don't have any framework for understanding what this means at all.

[428] And framework really, I mean, it's 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.

[429] You build them up with the expectation that there's a shared understanding of what they are.

[430] It's the social, that us humans.

[431] Do you have a sense that humans on Earth, in general, share a set of, like, how many frameworks are there?

[432] I mean, it depends on how you bound them, right?

[433] So, in other words, how big or small, like, their individual scope.

[434] But there's lots, and there are new ones.

[435] I think the way I think about is kind of in a layer.

[436] I think of the architectures are being layered in that.

[437] There's a small set of primitives that allow you the foundation to build frameworks.

[438] and then there may be, you know, many frameworks, but you have the ability to acquire them and then you have the ability to reuse them.

[439] 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.

[440] You know, I never heard of this game soccer, but if it's like basketball in the sense that the goal's like the hoop and I have to get the bowl in the hoop and I have guards and I have this and I have that, like where does the, where are the, where are the, similarities and where the difference is.

[441] And I have a foundation now for interpreting this new information.

[442] And then different groups, like the millennials will have a framework.

[443] And then then never, you know, Democrats and Republicans, millennials, nobody wants that framework.

[444] Well, I mean, I think.

[445] I mean, I think we're talking about political and social ways of interpreting the world around them.

[446] And I think these frameworks are still largely, largely similar.

[447] I think they differ in maybe what some fundamental assumptions and values are.

[448] Now, from a reasoning perspective, like the ability to process the framework, it might not be that different.

[449] The implications of different fundamental values or fundamental assumptions in those frameworks may reach very different conclusions.

[450] So from a social perspective, the conclusions may be very different.

[451] From an intelligence perspective, I just followed where my assumptions took me. Yeah, the process itself looks similar.

[452] But that's a fascinating idea that frameworks really help carve how a statement will be interpreted.

[453] I mean, having a Democrat and a Republican framework and read the exact same statement.

[454] And the conclusions that you derive will be totally different from an AI perspective is fascinating.

[455] What we would want out of the AI is to be able to tell you that this perspective, one perspective, that one set of assumptions is going to lead you here, another set of suspensions is going to leave you there.

[456] And in fact, you know, to help people reason and say, oh, I see where our differences lie.

[457] Yeah.

[458] You know, I have this fundamental belief about that.

[459] I have this fundamental belief about that.

[460] Yeah, that's quite brilliant.

[461] From my perspective, NLP, there's this idea that there's one way to really understand a statement.

[462] But there probably isn't.

[463] There's probably an infant number of ways to understand a statement.

[464] Well, there's lots of different interpretations.

[465] and the broader, you know, the broader to the content, the richer it is.

[466] And so, you know, you and I can have very different experiences with the same text, obviously.

[467] 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.

[468] 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.

[469] But that requires a commitment to breaking down that interpretation in terms of that framework in a logical way.

[470] Otherwise, you know, and this is why, like, I think of AI is really complementing and helping human intelligence to overcome some of its biases and its predisposition to be persuaded by, you know, by, you know, by, by, but more shallow reasoning in the sense that like we get over this idea well i you know you know i'm right because i'm republican or i'm right because i'm democratic and someone labeled this as a democratic point of view or it has the following keywords in it and of 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 we're not to sit and think about that's That's, I love that.

[471] 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.

[472] That's right.

[473] So, one of the greatest accomplishments in the history of AI is Watson competing in the Game of Jeopard Against Humans.

[474] And you were a lead in that, a critical part of that.

[475] Let's start the very basics.

[476] What is the game of Jeopardy?

[477] The game for us, humans, human versus human.

[478] Right.

[479] So it's to take a question and answer it.

[480] It's just the opposite.

[481] Actually, well, no, but it's not, right?

[482] It's really not.

[483] It's really to get a question and answer.

[484] But it's what we call a factoid question.

[485] So this notion of like it's really relates to some fact that a few people would argue.

[486] you whether the facts are true or not.

[487] In fact, most people, what, in jeopardy kind of counts on the idea that these statements have factual answers.

[488] And the idea is to, first of all, determine whether or not you know the answer, which is sort of an interesting twist.

[489] So first of all, understand the question.

[490] You have to understand the question.

[491] What is it asking?

[492] And that's a good point, because the questions are not asked directly, right?

[493] 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 it's a little bit tricky yeah they're asked in and exactly in numerous witty tricky ways exactly what they're asking is not obvious it takes it takes inexperienced humans a while to go what is it even asking right and it's sort of an interesting realization that you have when 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 still trying to process the question, and the champions have answered and moved on.

[494] There are three questions ahead by the time you figured out what the question even met.

[495] So there's definitely an ability there to just parse out what the question even is.

[496] So that was certainly challenging.

[497] It's interesting historically, though, if you look back at the Jeopardy Games, much earlier, you know, like 60s, 70s, that kind of thing.

[498] The questions were much more direct.

[499] They weren't quite like that.

[500] They got sort of more and more interesting.

[501] 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 in figuring out what the question was even asking.

[502] So yeah, you have to figure out the questions even asking.

[503] Then you have to determine whether or not you think you know the answer.

[504] And because you have to buzz in really quickly, you sort of have to make that determination as quickly as you possibly can.

[505] Otherwise, you lose the opportunity to.

[506] buzz in.

[507] Even before you really know if you know the answer.

[508] I think a lot of humans will assume they'll look at, they'll look at it, they're processed it very superficially.

[509] In other words, what's the topic?

[510] What are some keywords and just say, do I know this area or not?

[511] Before they actually know the answer, then they'll buzz in and then they'll buzz in and think about it.

[512] So it's interesting what humans do.

[513] Now some people who know all things like Ken Jennings or something or the more recent big jeopardy player, I mean, those just buzz in.

[514] Those are 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 what watson 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 could we come up with if we look you know we had a big body of knowledge some of the order of several terabytes i mean from from a web scale was actually very small But from like a book scale, I was talking about millions of books, right?

[515] So the equivalent millions of books, encyclopedias, dictionaries, books.

[516] So it's a ton of information.

[517] And, you know, for, I think it was only 85 % was the answer anywhere to be found.

[518] So you're ready down, you're ready down at that level just to get started, right?

[519] And so it was important to get a very quick sense of, do you think you know the right answer to this question?

[520] So we had to compute that confidence as quickly as we possibly could.

[521] 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.

[522] And that would depend on what else was going on in the game because it was a risk.

[523] So like if you're really in a situation where I have to take a guess, I have very little to lose, then you'll buzz in with less confidence.

[524] So that was accounted for the financial standings of the different competitors.

[525] Correct.

[526] How much of the game was laughed?

[527] How much time was laughed?

[528] Where you were in the standing and things like that.

[529] How many hundreds of milliseconds that we're talking about here?

[530] Do you have a sense of what is...

[531] We targeted...

[532] Yeah, what's the target?

[533] So, I mean, we targeted answering in under three seconds.

[534] And...

[535] Busing in.

[536] So the decision to buzz in and then the actual answering, Are those two different stages?

[537] Yeah, they were two different things.

[538] In fact, we had multiple stages, whereas, like, we would say, let's estimate our confidence, which was sort of a shallow answering process.

[539] 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.

[540] But by and large, we're saying, like, we can't play the game.

[541] We can't even compete if we can't, on average, answer these questions in around three seconds or less.

[542] So you stepped in.

[543] So there's this, there's these three humans playing a game and you stepped in with the idea that IBM Watson would be one of, replace one of the humans and compete against two.

[544] Can you tell the story of Watson taking on this game?

[545] Sure.

[546] It seems exceptionally difficult.

[547] Yeah.

[548] So the story was that it was coming up, I think, to the 10 -year anniversary of Big Blue, not Big Blue.

[549] Deep Blue.

[550] Deep Blue.

[551] IBM, wanted to do sort of another kind of really, you know, fun challenge, public challenge that can bring attention to IBM research and the kind of the cool stuff that we were doing.

[552] I had been working in AI at IBM for some time.

[553] 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.

[554] We're not even going to tell you what they're about.

[555] Can you go off and get accurate answers to these questions?

[556] and it was an area of AI research that I was involved in.

[557] And so it was a very specific passion of mine.

[558] Language understanding had always been a passion of mine.

[559] One sort of narrow slice on whether or not you could do anything was 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.

[560] So that was a research area that my team had already been in.

[561] And so completely independently, several, you know, IBM executives, like, what are we going to do?

[562] What's the next cool thing to do?

[563] And Ken Jennings was on his winning streak.

[564] This was like, whatever it was, 2004, I think, was on his winning streak.

[565] And someone thought, hey, that would be really cool if the computer can play Jeopardy.

[566] And so this was like in 2004, they were shopping this thing around.

[567] And everyone was telling the research execs, no way.

[568] Like, this is crazy.

[569] And we had some pretty, you know, senior people in the field and they're saying, well, this is crazy.

[570] And it'll come across my desk and I was like, but that's kind of what I'm really interested in doing.

[571] And, but there was such this prevailing sense of, this is nuts, we're not going to risk IBM's reputation on this.

[572] We're just not doing it.

[573] And this happened in 2004.

[574] It happened in 2005.

[575] At the end of 2006, it was coming around again.

[576] And I was coming off of a, I was doing the open domain question answering stuff, but I was coming off a couple other projects had a lot more time to put into this.

[577] And I argued that it could be done.

[578] And I argue it would be crazy not to do this.

[579] Can I, you could be honest at this point.

[580] So even though you argued for it, what's the confidence that you had yourself privately that this could be done?

[581] Was we just told the story, how you tell stories to convince others?

[582] How confident were you?

[583] What was your estimation of the problem at that time?

[584] So I thought it was possible, and a lot of people thought it was impossible.

[585] I thought it was possible.

[586] The reason why I thought it was possible was because I did some brief experimentation.

[587] I knew a lot about how we were approaching open domain factoid question answering, I've been doing it for some years.

[588] I looked at the Jefferty stuff.

[589] I said, this is going to be hard for a lot of the points that we mentioned earlier.

[590] Hard to interpret the question.

[591] Hard to do it quickly enough, hard to compute an accurate confidence.

[592] none of this stuff had been done well enough before.

[593] But a lot of the technologies were building with the kinds of technologies that should work.

[594] But more to the point, what was driving me was, I was in IBM research.

[595] I was a senior leader in IBM research, and this is the kind of stuff we were supposed to do.

[596] In other words, we were basically supposed to take things and say, this is an active research area.

[597] It's our obligation to kind of, if we have the opportunity, to push it to the limits.

[598] And if it doesn't work, to understand more deeply why we can't do it.

[599] And so I was very committed to that notion saying, folks, this is what we do.

[600] It's crazy not to do it.

[601] This is an active research here.

[602] We've been in this for years.

[603] Why wouldn't we take this grand challenge and push it as hard as we can?

[604] At the very least, we'd be able to come out and say, here's why this problem is way hard.

[605] Here's what we've tried and here's how we fail.

[606] So I was very driven as a scientist from that perspective.

[607] And then I also argued, based on what we did a feasibility study, why I thought it was hard but possible.

[608] 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.

[609] But for the most part, at that point, the execs really were just looking for someone crazy enough to say yes.

[610] 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 thing.

[611] Clearly, you did not have such fears.

[612] Okay, I did not.

[613] So you dived right in, and yet, for what I understand, it was performing very poorly in the beginning.

[614] So what were the initial approaches and why did they fail?

[615] Well, there were lots of hard aspects to it.

[616] I mean, And 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.

[617] Like, what are you even asking for, right?

[618] 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, but, you know, what city or what person often when it would name it very clearly, you would know that.

[619] And if there were just a small set of them, in other words, we're going to ask about these five types.

[620] Like, it's going to be an answer, and the answer will be a city in this state or a city in this country.

[621] The answer will be a person of this type, right?

[622] Like an actor or whatever it is.

[623] But it turns out that in jeopardy, there were like tens of thousands of these things.

[624] And it was a very, very long tail, meaning, you know, it just went on and on.

[625] 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 are the most frequent, you still cover a very small percentage of the data.

[626] So you couldn't take that approach of saying, I'm just going to try to collect facts about these five or ten types or 20 types or 50 types or whatever.

[627] So that was like one of the first things like, what do you do about that?

[628] And so we came up with an approach toward that.

[629] and the approach to look promising and we continue to improve our ability to handle that problem throughout the project.

[630] 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.

[631] So we did it in four, so I got lucky.

[632] But one of the things that that, putting that stake in the ground was I knew how hard the language understanding problem was.

[633] I said, we're not going to actually understand understand language to solve this problem.

[634] We are not going to interpret the question and the domain of knowledge that the question refers to in reason over that to answer these questions.

[635] Obviously, we're not going to be doing that.

[636] At the same time, simple search wasn't good enough to confidently answer with a single correct answer.

[637] First of all, that's brilliant.

[638] That's such a great mix of innovation in practical engineering, three, four, eight.

[639] So you're not you're not trying to solve the general NLU problem.

[640] You're saying, let's solve this in any way possible.

[641] Oh, yeah.

[642] No, I was committed to saying, look, we're going to solving the open domain question answering problem.

[643] We're using Jeopardy as a driver for that.

[644] Big benchmark.

[645] Hard enough, big benchmark, exactly.

[646] And now we're...

[647] How do we do it?

[648] We could just like whatever, like, just figure out what works because I want to be able to go back to the academic science community and say, here's what we tried, here's what worked, here's what didn't work.

[649] That's great.

[650] I don't want to go.

[651] go in and say, oh, I only have one technology.

[652] I have a hammer.

[653] I'm only going to use this.

[654] I'm going to do whatever it takes.

[655] I'm like, I'm going to think out of the box and do whatever it takes.

[656] One, and I also, there was another thing I believed.

[657] I believe that the fundamental NLP technologies and machine learning technologies would be adequate.

[658] And this was an issue of how do we enhance them, how do we integrate them, how do we advance them?

[659] So I had one researcher and came to me who had been working on question answering with me for 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 yeah 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 an architecture on how to integrate it well and then see where it breaks and make the necessary advances we need to make until this thing works.

[660] Yeah, push it hard to see where it breaks and then patch it off.

[661] I mean, that's how people change the world.

[662] I mean, that's the Yon Musk approach of the rockets, SpaceX, that's the Henry Ford and so on.

[663] I love it.

[664] And I happen to be, in this case, I happen to be right, but like we didn't know.

[665] But you kind of have to put a stake in the ground to say, how are you going to run the project?

[666] So, yeah, and backtracking to search.

[667] So if you were to do, what's the brute force solution?

[668] What would you search over?

[669] So you have a question, how would you search the possible space of answers?

[670] Look, web search has come a long way even since then.

[671] But at the time, like, you know, first of all, I mean, there are a couple of other constraints around the problem, which is interesting.

[672] So you couldn't go out to the web.

[673] You couldn't search the internet.

[674] In other words, the AI experiment was, we want a self -containting device.

[675] If device is as big as a room, fine, it's as big as a room, but we want a self -contained device.

[676] You're not going out to the internet.

[677] You don't have a lifeline to anything.

[678] So it had to kind of fit in a shoebox, if you will, or at least the size of a few refrigerators, whatever it might be.

[679] But also, you couldn't just get out there.

[680] You couldn't go off network, right, to kind of go.

[681] So there was that limitation.

[682] But then we did, but the basic thing was go go do a web search.

[683] The problem was, even when we went and did a web search, I don't remember exactly the numbers, but someone in the order of 65 % of the time, the answer would be somewhere, you know, in the top 10 or 20 documents.

[684] So first of all, that's not even good enough to play Jeopardy.

[685] You know, the words, even if you could pull the, if you could perfectly pull the answer out of the top 20 documents, top 10 documents, whatever was, which we didn't know how to do.

[686] But even if you could do that, you'd be and you knew it was right unless you 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 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 right winter circle you have to be up over 70 and you have to do it really 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 computer confidence of all the possible candidates.

[687] So it's not like I go in knowing the right answer and have to pick it.

[688] I don't know the right answer.

[689] I have a bunch of documents, somewhere in there is the right answer.

[690] How do I, as a machine, go out and figure out which one's right?

[691] And then how do I score it?

[692] And now how do I deal with the fact that I can't actually go out to the web?

[693] First of all, if you pause on that, just think about it.

[694] If you could go to the web, do you think that problem is solvable?

[695] If you just pause on it, just thinking even beyond Jeopardy?

[696] Do you think the problem of reading text to find where the answer is?

[697] Well, we solved that in some definition of solved, given the Jeopardy Challenge.

[698] How did you do it for Jeopardy?

[699] So how did you take a body of work in a particular topic and extract the key pieces of information?

[700] So now forgetting about the huge volumes that are on the web, right?

[701] So now we have to figure out we did a lot of source research.

[702] In other words, what body of knowledge is going to be small enough but brought enough to answer jeopardy.

[703] And we ultimately did find the body of knowledge that did that.

[704] I mean, it included Wikipedia and a bunch of other stuff.

[705] So like encyclopedia type of stuff.

[706] I don't know if you can speak to it.

[707] Cyclopedia is dictionaries, different times of semantic resources, like WordNet and other types of semantic resources like that, as well as like some web crawls.

[708] In other words, where we went out and took that content and then expanded it based on producing statistical seed, you know, statistically producing seeds, using those seeds for other searches, and then expanding that.

[709] So using these, like, expansion techniques, we went out and had found enough content that were like, okay, this is good.

[710] And even up until the end, we had a threat of research.

[711] It was always trying to figure out what content could we efficiently include.

[712] I mean, there's a lot of popular, like, what is the church lady, well, I think was one of the end.

[713] Like, what, what do you, I guess, that's probably, probably in encyclopedias.

[714] So I guess...

[715] Encyclopedia, but then we would take that stuff when we would go out and we would expand.

[716] In other words, we'd go find other content that wasn't in the core resources and expand it.

[717] You know, 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.

[718] It's very select.

[719] We then took all that content.

[720] We pre -analyzed the crap out of it, meaning we parsed it, you know, broke it down into all its individual words, and we did semantic, satactic and semantic parses on it, you know, had computer algorithms that annotated it, and we indexed that in a very rich and very fast index.

[721] So we have a relatively huge amount of, let's say the equivalent of, for the sake of argument, two to five million bucks.

[722] We've now analyzed all that, blowing up its size even more because now we've all this metadata, and then we richly indexed all of that.

[723] And by way, in a giant in memory cache.

[724] So Watson did not go to disk.

[725] So the infrastructure component there, if you could just speak to it, how tough it, I mean, I know 2000, maybe this is 2008, 9, you know, that's kind of a long time ago.

[726] Right.

[727] How hard is it to use multiple machines?

[728] How hard is it the infrastructure component, the hardware component?

[729] So we used IBM.

[730] So we used IBM hardware.

[731] We had something like, I forget exactly, but close to 3 ,000 cores.

[732] completely connected.

[733] So you had a switch where, you know, every CPU was connected to every other.

[734] And they were sharing memory in some kind of way.

[735] Large shared memory, right?

[736] And all this data was pre -analyzed and put into a very fast indexing structure that was all in memory.

[737] And then we took that question.

[738] We would analyze the question.

[739] So all the content was now pre -analyzed.

[740] 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.

[741] How do you shove that question?

[742] How do you connect the big stuff with the big knowledge base of the metadata and that's indexed to the simple little witty, confusing question?

[743] Right.

[744] So therein lies, you know, the Watson architecture.

[745] So we would take the question.

[746] We would analyze the question, which means that we would parse it.

[747] and interpreted a bunch of different ways.

[748] We'd try to figure out what is it asking about.

[749] So we had multiple strategies to kind of determine what was it asking for.

[750] 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.

[751] So anyway, the bottom line is we would do a bunch of analysis in the question.

[752] And question analysis had to finish and had to finish fast.

[753] So we do the question analysis because then from the question analysis, question analysis, we would now produce searches.

[754] 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.

[755] 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.

[756] And they would come back with passages.

[757] So these were passage search algorithms.

[758] They would come back with passages.

[759] And so now, let's say you had a thousand passages.

[760] Now, for each passage, you parallelized again.

[761] So you went out and you paralyzed the search.

[762] Each search would now come back with a whole bunch of passages.

[763] Maybe you had a total of 1 ,000 or 5 ,000, whatever passages.

[764] For each passage now, you'd go and figure out whether or not there was a candidate would call a candidate answer in there.

[765] So you had a whole bunch of other algorithms that would find candidate answers, possible answers to the question.

[766] And so you had candidate answer called candidate answers generators, a whole bunch of those.

[767] 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.

[768] And speed.

[769] so better is accuracy and speed.

[770] Correct.

[771] So, right, speed and accuracy, for the most part, we're separated.

[772] We handle that sort of in separate ways.

[773] Like, I focus purely on accuracy and to inaccuracy.

[774] Are we ultimately getting more questions and producing more accurate confidences?

[775] And then a whole other team that was constantly analyzing the workflow to find the bottlenecks and then figuring out of both parallelize and drive the algorithm speed.

[776] But anyway, so now think of it, like, you have this big fan out now, right?

[777] because you have multiple queries, now you have thousands of candidate answers.

[778] For each candidate answer, you're going to score it.

[779] So you're going to use all the data that built up.

[780] You're going to use the question analysis.

[781] You're going to use how the query was generated.

[782] 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.

[783] So now we have a group of researchers coming up with scorers.

[784] There are hundreds of different scorers.

[785] So now you're getting a fan out of again from however many candidate answers you have to all the different scores.

[786] So if you have 200 different scores and you have 1 ,000 candidates, now you have 200 ,000 scores.

[787] 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?

[788] And I want to rank them based on the likelihood that there are a correct answer to the question.

[789] So every scorer was its own research project.

[790] What do you mean by score?

[791] So is that the annotation process of basically a human being saying that this answer has a quality of it.

[792] Think of it.

[793] If you want to think of it, what you're doing, you know, if you want to think about what a human would be doing, human would be looking at a possible answer.

[794] They'd be reading the, you know, Emily Dickson, Dickinson, they'd be reading the passage in which that occurred.

[795] 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 the this passage is the right answer to that question got it so that that's the annotation task that's the scoring task so but scoring implies zero to one kind of that's right you give it a zero to one score so it's not a binary no give it a score give it a zero to yeah exactly so but humans do 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 uh we we actually looked at the raw scores as well, standardized scores, because humans are not involved in this.

[796] Humans are not involved.

[797] Sorry, so I'm misunderstanding the process here.

[798] This is passages.

[799] Where is the ground truth coming from?

[800] Grand truth is only the answers to the questions.

[801] So it's end to end.

[802] It's end to end.

[803] So I was always driving end to end performance.

[804] It was a very interesting, a very interesting, you know, engineering approach and ultimately scientific and research approach, are always driving into end.

[805] Now, that's not to say we wouldn't make hypotheses that individual component performance was related in some way to end -to -end performance.

[806] Of course we would, because people would have to build individual components.

[807] But ultimately, to get your component integrated into the system, you had to show impact on end -to -end performance, question -answering performance.

[808] So there's many very smart people work on this, and they're basically.

[809] trying to sell their ideas as a component that should be part of the system.

[810] That's right.

[811] 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 a better candidate, or a better type estimation, whatever it is.

[812] And then I would say, I need to understand how the improvement on that component metric is going to affect the end -to -end performance.

[813] If you can't estimate that and can't do experiments to demonstrate that, it doesn't get in.

[814] That's like the best run AI project I've ever heard.

[815] That's awesome.

[816] Okay.

[817] 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?

[818] Like, wait, where people began to believe, or is it just a gradual process?

[819] 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, as we follow this procedure of different ideas, build different components, plug them into the architecture, or run the system, see how we do, do the error 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.

[820] And this is where we leverage machine learning in a very important way.

[821] So while individual components could be statistically driven, machine learning components.

[822] Some of them were heuristic.

[823] Some of them were machine learning components.

[824] The system has a whole combined all the scores using machine learning.

[825] This was critical because that way you can divide and conquer.

[826] So you can say, okay, you work on your candidate generator or you work on this approach to answer scoring.

[827] You work on this approach to type scoring.

[828] You work on this approach to passage search or the passage selection and so forth.

[829] 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 uh 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.

[830] So, yeah, the machine learning is doing the fusion, and then it's a human orchestrated ensemble of different approaches.

[831] That's great.

[832] Still, impressive that you're able to get it done in a few years.

[833] That's not obvious to me that it's doable if I just put myself in that mindset.

[834] But when you look back at the Jeopardy Challenge, again, when you're looking up at the stars, what are you most proud of?

[835] It's looking back at those days.

[836] I'm most proud of my commitment and my team's commitment to be true to the science, to not be afraid to fail.

[837] 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.

[838] That's right.

[839] Do you think it was a success?

[840] In the eyes of the world, it was a success.

[841] By your, I'm sure, exceptionally high standards.

[842] Is there something you regret you would do differently?

[843] It was a success.

[844] It was a success for our goal.

[845] Our goal was to build the most advanced open domain question answering system.

[846] 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.

[847] So we wanted Jeopardy.

[848] So it was a success.

[849] It was, I worry that the world would not understand it as a success because it came down to only one game.

[850] And I knew statistically speaking, this can be a huge technical success and we could still lose that one game.

[851] And that's a whole other theme of this, of the journey.

[852] But it was a success.

[853] It was not a success in natural language.

[854] understanding.

[855] But that was not the goal.

[856] Yeah, that was, 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.

[857] There was a success of being an inspiration to future challenges.

[858] Absolutely.

[859] That drive future efforts.

[860] What's the difference between how human being compete in jeopardy and how Watson does it.

[861] That's important in terms of intelligence.

[862] Yeah.

[863] So that actually came up very early on in the project also.

[864] 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.

[865] I'd wanted to think about how humans do it.

[866] And they were, you know, from a cognition perspective, like human cognition and how that should play.

[867] And I would not take them on the project.

[868] because another assumption or another stake I put in the ground was I don't really care.

[869] Are humans do this?

[870] At least in the context of this project.

[871] 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.

[872] So it wasn't that I didn't care in general.

[873] In fact, as an AI scientist, I care a lot about that.

[874] 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 like if I'm going to get this done I'm going to chart this path and this path says we're going to engineer a machine that's going to 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 going to get there from here in the time, you know, in the time frame.

[875] I think that's a great way to lead the team.

[876] But now there's done and then one when you look back.

[877] Right.

[878] Analyze, what's the difference, actually, between.

[879] 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.

[880] It might have been closer to the way humans answer questions than I might have imagined previously.

[881] Because humans are probably in the Game of Jeopardy at the level of Ken Jennings are probably also cheating their way to winning, right?

[882] Well, they're doing shallow analysis.

[883] They're doing the fastest possible.

[884] They're doing shallow analysis.

[885] So they are very quickly analyzing the question and coming up with some key vectors or cues, if you will and they're taking those cues and they're very quickly going through like their library of stuff not deeply reasoning about what's going on and then sort of like a lots of different like what we call these these scorers what's kind of score that in a very shallow way and then say oh boom you know that's what it is and 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?

[886] 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 of the machine and the intelligence of the human very compatible.

[887] Well, compatible in the sense they can communicate with one another and they can reason with this shared understanding.

[888] So how they think about things and how they build answers, how they build explanations, becomes a very important question to consider.

[889] 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?

[890] Yeah, well, this goes back to the interpretation of what we were talking about before.

[891] framework.

[892] Jeopardy, the system's not trying to interpret the question, and it's not interpreting the content that's reusing, and with regard to any particular framework.

[893] I mean, it is parsing it and parsing the content and using grammatical cues and stuff like that.

[894] So if you think of grammar as a human framework in some sense, it has that.

[895] But when you get into the richer 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?

[896] And, and, and, and, And where are things in time and space?

[897] And like when you start thinking about how humans formulate and structure the knowledge that they acquire in their head, and wasn't doing any of that.

[898] What do you think are the essential challenges of like free -flowing communication, free -flowing dialogue versus question answering, even with the framework, with the interpretation, dialogue?

[899] Yep.

[900] Do you see free -flowing dialogue?

[901] as fundamentally more difficult than question answering, even with shared interpret.

[902] Yeah, so dialogue is important in a number of different ways.

[903] I mean, it's a challenge.

[904] So first of all, when I think about the machine that, when I think about a 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 or needs to be able to interact with humans so in some sense needs to dialogue that doesn't mean that 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 could mimic casual conversations we're not trying to mimic casual conversations We're really trying to produce a machine whose goal as it is to help you think and help you reason about your answers and explain why.

[905] 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 like what do you want to think about?

[906] Like what do you want to reason about?

[907] I'm going to tell you the information I have.

[908] I'm going to have to summarize it.

[909] I'm going to ask you questions.

[910] You're going to answer those questions.

[911] I'm going to go back and forth with you.

[912] I'm going to figure out what your mental model is.

[913] I'm going to now relate that to the information I have and present it to you in a way that you can understand it, and then we could ask follow -up questions.

[914] So it's that type of dialogue that you want to construct.

[915] It's more structured.

[916] It's more goal -oriented, but it needs to be fluid.

[917] In other words, it can't, it has to be engaging in fluid.

[918] It has to be productive and not distract.

[919] acting.

[920] 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.

[921] So basically a productive, rich conversation, unlike this podcast.

[922] What?

[923] I'd like to think it's more similar to this podcast.

[924] I was joking.

[925] 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 a I think a lot of the stuff you said with frameworks is quite brilliant but just creating a smooth discourse yeah it feels clunky right now 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.

[926] So I think to do this, you kind of have to be creative in the following sense.

[927] If I were to do this as 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.

[928] And you struggle.

[929] I don't know to collect the data.

[930] Because I don't know how much data is like that.

[931] Okay.

[932] 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, like, how do you, like, successful examples?

[933] 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, you, if you have enough data, you could potentially learn the pattern.

[934] In an example like this, it's hard to do.

[935] This is a, you know, sort of a human sort of thing to do.

[936] what recently came out at IBM was the debater projects 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 than generating a structured argument where you have lots of other structured arguments like this you can potentially annotate that data and you could say this is a good response is bad response in a particular domain I mean, 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.

[937] So I think that it's not clear that we could just train on the body of data to do this, but we could bootstrap it.

[938] In other words, we can be creative, and we could say, what do we think, what do we think the structure of a good dialogue is that does this well.

[939] And we can start to create that.

[940] 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 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.

[941] But I have to understand what features to even learn over.

[942] So I have to have to bootstrap the process a little bit first.

[943] And that's a creative design task that I could then use as input into a more automatic learning task.

[944] Some creativity in, yeah, and boostrapping.

[945] Right.

[946] What elements of a conversation do you think you would like to see?

[947] So one of the benchmarks for me is humor, right?

[948] That seems to be one of the hardest.

[949] And to me, the biggest contrast is Watson.

[950] So one of the greatest sketches of comedy sketches of all time, right, is the SNL Celebrity Jeopardy with Alex Trebek and Sean Connery and Bert Reynolds and so on, with Sean Connery commenting on Alex Trebek's mother a lot.

[951] And I think all of them are in the negative point wise.

[952] So they're clearly all losing in terms of the game of Jeopardy, but they're winning in terms of comedy.

[953] So what do you think about humor in this whole interaction in the dialogue that's productive?

[954] Or even just whatever, what humor represents to me is, the same idea that you're saying about framework, because humor only exists within a particular human framework.

[955] So what do you think about humor?

[956] What do you think about things like humor that connect to the kind of creativity you mentioned that's needed?

[957] I think there's a couple of things going on there.

[958] So I, 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 in jeopardy.

[959] We literally sat down and said, well, you know, how do puns work?

[960] And, you know, it's like word play.

[961] And you could formalize these things.

[962] So I think there's a lot aspects of humor that you could formalize.

[963] You could also learn humor.

[964] You could just say, what do people laugh at?

[965] And if you have enough, again, if you have enough data to represent the phenomenon, might be able to, you know, weigh the features and figure out, you know, what humans find funny and what they don't find funny.

[966] You might, the machine might not be able to explain why the human finding anybody unless we, unless we sit back and think about that more formally.

[967] I think, again, I think you do a combination of both.

[968] And I'm always a big proponent of that.

[969] I think, you know, 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 how to combine these two approaches.

[970] I think there's another aspect to humor, though, which goes to the idea that I feel like I can relate to the person telling the story, to the person telling the story.

[971] 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?

[972] And when I know, when I imagine that the robot is not conscious the way I'm unconscious.

[973] When I imagine the robot does not actually have the experiences that I experience, do I find it funny?

[974] Or do, because it's not as related, but I don't imagine that the person's relating it to it the way I relate to it.

[975] I think this also, you see this in the arts and in entertainment where, like, you know, sometimes you have savants who are remarkable at a thing, whether it's sculpture, it's music or whatever.

[976] But the people who get the most attention are the people who can evoke a similar emotional response who can get you to emot, right, about the way they are.

[977] 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.

[978] And that's when it becomes compelling.

[979] So they're communicating at a whole different level.

[980] They're just not communicating the artifact.

[981] They're communicating their emotional response to the artifact.

[982] and then you feel like, oh, wow, I can relate to that person, I can connect to that, I can connect to that person.

[983] So I think humor has that aspect as well.

[984] 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.

[985] So we're almost looking to make them human.

[986] Maybe from your experience with Watson, maybe you can comment on, did you consider that as part, well, obviously the problem of jeopardy doesn't require anthropomorphization, but nevertheless?

[987] 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.

[988] But you're absolutely right.

[989] I mean, humans do anthropomorphize, and without necessarily a lot of work.

[990] I mean, you 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.

[991] So I didn't mean to suggest that that connection cannot be mimicked.

[992] I think that connection can be mimicked and can get you to, can produce that emotional response.

[993] 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 big or whatever it's doing, just getting the emotional response, will you still feel it?

[994] Interesting, I think you probably would for a while, and then when it becomes more important that there's a deeper shared understanding, it may run flat, but I don't know.

[995] I'm pretty confident that the majority of the world, even if you tell them, it won't matter well it will not matter especially if the machine herself says that she is conscious that's very possible so you 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 you're deep into the science fiction genre now but yeah i don't think it's it's actually psychology i think it's not science fiction i think it's reality i think it's a it's a really powerful one that will have to be exploring in the next few decades.

[996] It's a very interesting element of intelligence.

[997] So what do you think, we've talked about social constructs of intelligence and frameworks in the way humans kind of interpret information.

[998] What do you think is a good test of intelligence in your view?

[999] So there's the Alan Turing with the Turing test.

[1000] Watson accomplished something very impressive with Jeopardy.

[1001] What do you think, is a test that would impress the heck out of you, that you saw that a computer could do.

[1002] They would say, this is crossing a kind of threshold that gives me pause in a good way.

[1003] My expectations for AI are generally high.

[1004] What does high look like, by the way?

[1005] So not the threshold.

[1006] Test is a threshold.

[1007] What do you think is the destination?

[1008] What do you think is the ceiling?

[1009] I think machines will, in many measures, will be better than us, will become more effective.

[1010] In other words, better predictors about a lot of things than ultimately we can do.

[1011] I think where they're going to struggle is what we talked about before, which is relating to communicating with and understanding humans in deeper ways.

[1012] And so I think that's a key point.

[1013] Like, we can create the super parrot.

[1014] What I mean by the super parrot is, given enough data, a machine can mimic your emotional response, can even generate language that will sound smart and what someone else might say under similar circumstances.

[1015] Like, I would just pause on that.

[1016] Like, that's a super parrot, right?

[1017] So given similar circumstances, moves its 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.

[1018] I think you just described me in a nutshell.

[1019] So I think philosophically speaking, you could argue that that's all we're doing as human beings, too.

[1020] So I was going to say it's very possible humans do behave that way too.

[1021] 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.

[1022] Like, humans are statistical language model machines and they are capable reasoners.

[1023] You know, they're both.

[1024] And you don't know which is going on, right?

[1025] So, and I think it's an interesting problem.

[1026] We talked earlier about, like, where we are in our.

[1027] 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?

[1028] Can you do that without dialogue, without interrogative or probing dialogue?

[1029] So it's interesting because humans are really good in their own mind justifying or explaining what they hear because they project their understanding onto yours.

[1030] So you could say you could put together a string of word.

[1031] and someone will sit there and interpret it in a way that's extremely biased to the way they want to interpret it.

[1032] They want to assume you're an idiot and they'll interpret it one way.

[1033] They will all assume you're a genius and they'll interpret it another way that suits their needs.

[1034] So this is tricky business.

[1035] So I think to answer your question, as AI gets better and better mimic, we create the super parrots, we're challenged, just as we are with, we're challenged with humans.

[1036] Do you really know what you're talking about.

[1037] 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 you think they make sense?

[1038] Can you convince me what the implications are?

[1039] You know, can you, so can you reason intelligently and make me believe that those, the implications of your prediction and so forth.

[1040] So what happens is it becomes reflective.

[1041] My standard for judging your intelligence depends a lot on mine.

[1042] 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, then it would pass.

[1043] There should be, but I think, depending on the content, one of the problems we have there is that, if that large community of people are not judging it with regard to a rigorous standard of objective, logic, and reason, you still have a problem.

[1044] Like, masses of people can be persuaded.

[1045] The millennials, yeah.

[1046] To turn their brains off.

[1047] Right, okay.

[1048] Sorry.

[1049] By the way, I have nothing against the one.

[1050] I don't know.

[1051] I'm just, just, just, so you're a part of one of the great, benchmarks, challenges of AI history, what do you think about Alpha Zero, Open AI5, 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.

[1052] What are your thoughts about that challenge?

[1053] I think it was a giant landmark for AI.

[1054] I think it was phenomenal.

[1055] I mean, as one of those other things, nobody thought like solving Go was going to be easy, particularly because Again, it's hard for, particularly hard for humans.

[1056] Hard for humans to learn.

[1057] Hard for humans to excel at.

[1058] And so it was another measure, a measure of intelligence.

[1059] It's very cool.

[1060] I mean, it's very interesting, you know, what they did.

[1061] I mean, and I loved how they solved, like, the data problem, which, again, they bootstrapped it and got the machine to play itself to generate enough data to learn from.

[1062] I think that was brilliant.

[1063] I think that was great.

[1064] And, of course, the result speaks for itself.

[1065] I think it makes us think about, again, it is, okay, what's intelligence, what aspects of intelligence are important?

[1066] Can the go machine help me make me a better go player?

[1067] Is it an alien intelligence?

[1068] Am I even capable of, like again, if we put in very simple terms, it found the function, it found the go function, can I even comprehend the go function?

[1069] Can I talk about the go function?

[1070] Can I conceptualize the go function, like whatever it might be?

[1071] So one of the interesting ideas of that system is that it plays against itself, right?

[1072] Yeah.

[1073] But there's no human in the loop there.

[1074] So like you're saying, it could have by itself created an alien intelligence.

[1075] How toward a goal, like imagine you're sentencing, you're a judge and you're sentencing people or you're setting policy or you're, you know, you're making medical decisions.

[1076] And you can't explain.

[1077] You can't get anybody to understand.

[1078] what you're doing or why.

[1079] So it's, it's, it's an interesting dilemma for the applications of AI.

[1080] 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, for the decision?

[1081] In other words, can you explain why you would do the thing?

[1082] 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?

[1083] Can you get behind the logic that was made there?

[1084] Do you think, sorry to link down this point, because it's a fascinating one, it's a great goal for AI.

[1085] Do you think it's achievable in many cases?

[1086] Okay, there's two possible worlds that we have in the future.

[1087] One is where AI systems do like medical diagnosis or things like that drive a car without ever explaining to you why it fails when it does.

[1088] That's one possible world and we're okay with it.

[1089] 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.

[1090] Which of those worlds are more likely do you think and which are concerning to you or not?

[1091] I think the reality is it's going to be a mix.

[1092] I'm not sure I have a problem with that.

[1093] 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, is already better than humans.

[1094] So, for example, I don't know that I take driverless cars.

[1095] If driverless cars learn how to be more effective drivers than humans but can't explain what they're doing, but bottom line, statistically speaking, they're, you know, 10 times safer than humans, I don't know that I care.

[1096] 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?

[1097] And I think those educators are interesting cases.

[1098] And now do we go to designers of the AI?

[1099] And the AI says, I don't know.

[1100] That's what it learned to do.

[1101] And it says, well, you didn't train it properly.

[1102] You know, you were negligent in the training data that you gave that machine.

[1103] Like, how do we drive down the reliable?

[1104] So I think those are, I think those are interesting questions.

[1105] So the optimization problem there, sorry, is to create an AI system, it's able to explain the lawyers away.

[1106] There you go.

[1107] I think that, I think it's going to be interesting.

[1108] I mean, I think this is where technology and social discourse are going to get like deeply intertwined and how we start thinking about problems, decisions, and problems like that.

[1109] I think in other cases, it becomes more obvious where, you know, it's like, why did you decide to give that person, you know, a longer sentence or to deny them parole?

[1110] 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 so and people are going to demand uh 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 they are using statistical hunches uh biases or even systematically using statistical averages to make calls.

[1111] I mean, this is what happened.

[1112] My dad, and if you saw that talk I gave about that, but, you know, I mean, they decided that my father was brain dead.

[1113] He had, went into cardiac arrest and it took to, took a long time for the ambulance to get there and he wasn't not resuscitated right away and so forth.

[1114] And they came, they told me he was brain dead.

[1115] And why was he brain dead?

[1116] Because essentially they gave me a purely statistical argument under these conditions with these four features, 98 % chance he's brain dead.

[1117] I said, but can you just tell me, not inductively, but deductively, go there and tell me his brain's not functioning is the way for you to do that.

[1118] And the protocol and response was, no, this is just how we make this decision.

[1119] I said, this is adequate for me. I understand the statistics, and I don't know how, you know, there's a 2 % chance he's to a lot.

[1120] Like, I just don't know the specifics.

[1121] I need the specifics of this case, and I want the deductible.

[1122] the logical argument about why you actually know he's brained it.

[1123] So I wouldn't sign that do not resuscitate.

[1124] And I don't know, it was like they went through lots of procedures.

[1125] It was a big, long story.

[1126] But the bottom was a fascinating story, by the way, but how I reasoned and how the doctor's reasoned through this whole process.

[1127] But I don't know, somewhere around 24 hours later or something, he was sitting up in bed with zero brain damage.

[1128] I mean, what lessons do you draw from that story, that experience?

[1129] That the data that's being used to make statistical inferences doesn't adequately reflect the phenomenon.

[1130] So in other words, you're getting shit wrong, sorry, you're getting stuff wrong because your model is not robust enough.

[1131] And you might be better off not using statistical inference and statistical averages in certain cases when you know the model is insufficient and that you should be reasoning about the specific case, more logical.

[1132] and more deductively, and hold yourself responsible or hold yourself accountable to doing that.

[1133] And perhaps AI has a role to say the exact thing what you just said, which is perhaps this is a case you should think for yourself.

[1134] You should reason deductively.

[1135] Well, it's hard, so it's hard because it's hard to know that.

[1136] You know, 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 the AI is good enough to do a particular task.

[1137] And regardless of whether or not it produces an explanation.

[1138] So, and what standard do we hold, right, for that?

[1139] So, you know, if you look at, you look more broadly, for example, as my father as a medical case, the medical system ultimately helped him a lot.

[1140] throughout his life without it it 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 that's fair um but it may be not in that particular case but overall like oh the medical system overall does more good than bad you know was doing more more good than bad now there's another argument that suggests that wasn't the case but for the sake of argument, let's say, like, that's, let's say, a net positive.

[1141] And I think you have to sit there and there and take that into consideration.

[1142] Now you look at a particular use case, like, for example, making this decision.

[1143] Have you done enough studies to know how good that prediction really is?

[1144] Right.

[1145] And have you done enough studies to compare it to say, well, what if we dug in in a more direct, you know, let's get the evidence, let's do the deductive thing and not use statistics here, how often would that have done better, right?

[1146] You just, so you have to do the studies to know how good the AI actually is.

[1147] And it's complicated because it depends how fast you have to make decision.

[1148] So if you have to make the decision super fast, do you have no choice?

[1149] Right.

[1150] If you have more time, right.

[1151] 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 my, what's going to happen to him in that room if you do it my way?

[1152] You know, if you do it.

[1153] Well, he's going to die anyway.

[1154] So let's do it my way, then.

[1155] I mean, it raises questions for our society to struggle with, as 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 the discourse.

[1156] And it starts, you know, I think, I think I'm safe to say that most of the violent crimes committed.

[1157] by males.

[1158] So if you discriminate based, you know, this is a male versus female saying that if it's a male, more likely to commit the crime.

[1159] 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.

[1160] Because it's causing a, regardless of what state, AI devices are or not, it's causing this dialogue to happen.

[1161] 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.

[1162] That has got to be one of the most important things we as a species can be doing, honestly.

[1163] we are reached we've 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 you know 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 you know to be having and and you know we are fundamentally our thinking can be and easily becomes fundamentally bias 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 you know we decide to reject statistical inference to favor to favor understanding and deciding on the individual.

[1164] Yes.

[1165] We consciously make that choice.

[1166] 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.

[1167] We do that at a respect for the individual.

[1168] Yeah, and that requires reasoning and thinking.

[1169] Looking forward, what grand challenges would you like to see in the future?

[1170] Because the jeopardy challenge, you know, captivated the world.

[1171] alpha go alpha zero captivated the world Deep Blue certainly beating Kasparov Gary's bitterness aside and captivated the world what do you think do you have ideas for next grand challenges for future challenges of that?

[1172] 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 know can demonstrate that they understand that they could read and understand that they can acquire these frameworks and communicate, you know, reason and communicate with humans.

[1173] So it is kind of like the Turing test, but it's a little bit more demanding than the Turing test.

[1174] It's not enough to convince me that you might be human because you could, you know, you can parrot a conversation.

[1175] I think, you know, the standard is a little bit higher.

[1176] Is, for example, can you, you know, the standard is higher.

[1177] And I think one of the challenges of devising this grand challenge is that we're not sure what intelligence is.

[1178] We're not sure how to determine whether or not to people actually understand each other and in what depth they understand it.

[1179] You know, to what depth they understand each other.

[1180] So the challenge becomes something along the lines of can you satisfy me that we have a shared.

[1181] understanding.

[1182] So if I were to probe and probe and you probe me, can can can can machines really act like thought partners where they can satisfy me that they that we have a share our understanding is shared enough that we can collaborate and produce answers together and that you know they can they can help me explain and justify those answers.

[1183] So maybe here's an idea.

[1184] So we'll have AI system run for president and convince...

[1185] That's too easy.

[1186] I'm sorry, go ahead.

[1187] Well, no, you have to convince the voters that they should vote.

[1188] So, like, I guess what does winning look like?

[1189] Again, that's why I think this is such a challenge, because we go back to the emotional persuasion.

[1190] We go back to, you know...

[1191] Now we're checking off an aspect of human cognition that is in many ways weak or flawed, right?

[1192] We're so easily manipulated.

[1193] Our minds are drawn for often the wrong reasons, right?

[1194] Not the reasons that ultimately matter to us, but the reasons that can easily persuade us.

[1195] 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.

[1196] And a good benchmark should not play with those elements of emotional manipulation?

[1197] I don't think so.

[1198] And I think that's where we have to set the higher standard for ourselves of what, you know, what does mean?

[1199] 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?

[1200] And is it a, and can you acquire new knowledge?

[1201] You know, can you, can you under, for example, can you reasonable, I have acquired new knowledge, can you identify where it's consistent or contradictory with other things you've learned?

[1202] And can you explain that to me and get me to understand that?

[1203] So I think another way to think about it perhaps is can a machine teach you?

[1204] Can it help you?

[1205] Oh, that's a really nice way to put it.

[1206] Can it help you understand something that you didn't really understand before?

[1207] Where it's, you know, it's taking you.

[1208] So you're not, again, it's almost like, can it teach you?

[1209] Can it teach you?

[1210] Can it?

[1211] Can it?

[1212] it help you learn?

[1213] And in an arbitrary space, so it can open those domain space.

[1214] So can you tell the machine, and again, this borrows from some science fiction, 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?

[1215] That's quite brilliant.

[1216] Well, the machine that passes that kind of test, do you think it would need to have self -awareness or even consciousness.

[1217] What do you think about consciousness and the importance of it, maybe in relation to having a body, having a presence, an entity?

[1218] Do you think that's important?

[1219] You know, people used to ask me if Watson was conscious, and I used to say conscious of what, exactly?

[1220] I mean, I think, you know, of self.

[1221] It depends what it is that you're a conscious of.

[1222] I mean, like, so, you know, did it, if you, you know, it's certainly easy for it to answer questions about, it would be trivial to program it so to answer questions about whether or not it was playing jeopardy.

[1223] I mean, it could certainly answer questions that would imply that it was aware of things.

[1224] Exactly, what does it mean to be aware and what does it mean to consciousness?

[1225] It's sort of interesting.

[1226] I mean, I think that we differ from one another based on what we're conscious of.

[1227] But wait, wait, wait, yes, for sure.

[1228] There's degrees of consciousness and there's so.

[1229] well and there's just areas like it's not just degrees what do you what are you aware of like what are you not aware of 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 Watson as far as I could tell did not have a fear of death certainly not most most humans do uh wasn't conscious of death it wasn't yeah 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 that 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.

[1230] Well, I think for generating goals.

[1231] Again, I think you could have an intelligence capability and a capability to learn, a capability to predict, but I think without, I mean, again, you get a fear, but essentially without the goal to survive.

[1232] So you think you can just encode that?

[1233] I think you can just encode that.

[1234] 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 its power source and survive.

[1235] I mean, so I don't know that that's philosophically a hard thing to demonstrate.

[1236] It sounds like a fairly easy thing to demonstrate that you can give it that goal.

[1237] We'll come up with that goal by itself.

[1238] I think you have to program that goal in.

[1239] But there's something because I think, as we touched on, intelligence is kind of like a social construct.

[1240] the fact that a robot would be protecting its power source would add depth and grounding to its intelligence in terms of us being able to respect it.

[1241] I mean, ultimately, it boils down to us acknowledging that it's intelligent.

[1242] And the fact that it can die, I think, is an important part of that.

[1243] The interesting thing to reflect on is how trivial that would be.

[1244] And I don't think if you knew how trivial that was, you would associate that with being intelligence.

[1245] I mean, I literally put in a statement of code that says, you know, you have the following actions you can take.

[1246] You give it a bunch of actions like maybe you mount a laser gun on it or you may, you know, you give the ability to scream or screech or whatever.

[1247] And, you know, and you say, you know, if you see your power source threatened and you could program that in and, you know, you're going to protect, you're going to take these actions to protect it.

[1248] you know, you teach it, train it on a bunch of things.

[1249] So, and now you're going to look at that and you can say, well, you know, that's intelligence because it's protecting the power source.

[1250] Maybe, but that's, again, 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 intelligent.

[1251] What timeline year will society have something that would, that you would be comfortable calling an artificial general intelligence system?

[1252] What's your intuition?

[1253] Nobody can predict the future, certainly not the next few months or 20 years away, but what's your intuition?

[1254] How far away are we?

[1255] It's hard to make these predictions.

[1256] 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 are 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's probably closer to a 20 -year thing, I guess.

[1257] But not several hundred years.

[1258] No, I don't think it's several hundred years.

[1259] I don't think it's several hundred years.

[1260] But again, so much depends on how committed we are to investing and incentivizing this type of work.

[1261] And it's sort of interesting.

[1262] Like, I don't think it's obvious how incentivized we are.

[1263] I think from a task perspective...

[1264] you know, 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, for many of these things.

[1265] From a general intelligence thing, it's kind of an interesting question.

[1266] Are we really motivated to do that?

[1267] And like, we just struggled ourselves right now to even define what it is.

[1268] So it's hard to incentivize when we don't even know what it is.

[1269] We're incentivized to create.

[1270] And if you said mimic a human intelligence, I just think there are so.

[1271] so many challenges with the significance and meaning of that.

[1272] There's not a clear directive.

[1273] There's no clear directive to do precisely that thing.

[1274] So assistance in a larger and larger number of tasks.

[1275] 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.

[1276] And then the same system would be doing the vacuum cleaning.

[1277] And then the same system would be teaching my kids that I don't have math.

[1278] I think that when you get into a general intelligence for learning physical tasks, and again, I 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 the abilities to physical tasks.

[1279] 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.

[1280] I think we'll 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 outpace humans and a variety of those things the underlying mechanisms for doing that may be the same meaning that 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 building these kinds of machines.

[1281] You could argue that the general learning infrastructure in there is a form of a general type of intelligence.

[1282] I think what starts getting harder is this notion of, you know, 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.

[1283] So how do you get in there?

[1284] How do you get the machine in the game?

[1285] How do you get the machine in the intellectual game?

[1286] Yeah, and to solve AGI, you probably have to solve that problem.

[1287] You have to get the machine.

[1288] So it's a little bit of a bootstrapping thing.

[1289] Can we get the machine engaged in the intellectual, I can call you a game, but in the intellectual dialogue with the humans?

[1290] Are the humans sufficiently an intellectual dialogue with each other to generate enough data in this context?

[1291] And how do you bootstrap that?

[1292] because every one of those conversations, every one of those conversations, those intelligent interactions, require so much prior knowledge that is a challenge to bootstrap it.

[1293] So that's, so the question is, and how committed?

[1294] So I think that's possible, but when I go back to, are we incentivized to do that?

[1295] I know we're incentivized to do the former.

[1296] Are we incentivized to do the latter significantly enough?

[1297] Do people understand what the latter really is well enough?

[1298] part of the elemental cognition mission is to try to articulate that better and better through demonstrations and to try 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?

[1299] What is the potential of this?

[1300] What's the business potential?

[1301] What's the society of the potential to that?

[1302] And to build up that incentive system around that.

[1303] Yeah, I think if people don't understand yet, I think they will.

[1304] I think there's a huge business potential here.

[1305] So it's exciting that you're working on it.

[1306] You've kind of skipped over, but I'm a huge fan of physical presence of things.

[1307] Do you think, you know, Watson had a body?

[1308] Do you think having a body adds to the interactive element between the AI system and a human or just in general to intelligence?

[1309] So I think, I think going back to that, shared understanding bit.

[1310] Humans are very connected to their bodies.

[1311] 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.

[1312] So in other words, our bodies are the tool we use to affect output, but they're also, but they also generate a lot of input for our brain.

[1313] And so we generate emotion, we generate all these feelings, we generate all these signals that machines don't have.

[1314] So machines that have this is the input data.

[1315] And they don't have the feedback that says, okay, I've gotten this, I've gotten this emotion or I've gotten this idea, I now want to process it, and then I can, it then affects me as a physical being, and then I can play that out.

[1316] In other words, I could realize the implications of that, because implications, again, on my mind -body complex, I then process that, and the implications, again, are internal features are generated.

[1317] I learn from them.

[1318] They have an effect on my mind -body complex.

[1319] So it's interesting when we think, do we want a human intelligence?

[1320] Well, if we want a human -compatible intelligence, probably the best thing to do is to embed it, embed it in a human body.

[1321] Just to clarify, and both concepts are beautiful, is humanoid robots.

[1322] So robots that look like humans is one.

[1323] or did you mean actually sort of what Elon Musk is working with NeurLink, really embedding intelligence systems to ride along human bodies?

[1324] No, I mean, riding along is different.

[1325] 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.

[1326] Part of that substrate, you know, is the idea that it generates these kinds of internal features, like sort of emotional stuff.

[1327] It has similar senses.

[1328] It has to do a lot of the same things with those same senses.

[1329] Right.

[1330] So I think if you want that, again, I don't know that you want that.

[1331] Like, that's not my specific goal.

[1332] I think that's a fascinating scientific goal.

[1333] I think it has all kinds of other implications.

[1334] That's sort of not the goal.

[1335] Like, I want to create, I think of it as like create intellectual thought.

[1336] partners for humans, so that kind of, that kind of intelligence.

[1337] I know there are other companies that are creating physical thought partner.

[1338] It's physical partners for humans, but that's kind of not where, you know, I'm at.

[1339] But the important point is that a big part of how, of what we process is that physical experience of the world around us.

[1340] On the point of thought partners, what role does an emotional connection or forgive me love have to play in that thought partnership.

[1341] Is that something you're interested in?

[1342] Put another way, sort of having a deep connection beyond intellectual.

[1343] With the AI?

[1344] Yeah, with the AI between human and AI.

[1345] Is that something that gets in the way of the rational discourse?

[1346] Is that something that's useful?

[1347] I worry about biases, you know, obviously.

[1348] So no, In other words, if you develop an emotion relationship with a machine, do all of a sudden you start, are more likely to believe what it's saying, even if it does make any sense.

[1349] So, you know, I worry about that.

[1350] But at the same time, I think the opportunity to use machines to provide human companionship is actually not crazy.

[1351] And intellectual and social companionship is not crazy idea.

[1352] Do you have concerns, as a few people do, Elon Musk, Sam Harris, about long.

[1353] long -term existential threats of AI, and perhaps short -term threats of AI.

[1354] 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?

[1355] I think there aren't things to worry about.

[1356] I think the giving machines too much leverage is a problem, and what I mean by averages, is too much control over things that can hurt us, whether it's socially, psychologically, intellectually, or physically.

[1357] And if you give the machines too much control, I think that's a concern.

[1358] You forget about the AI.

[1359] Just when you give them too much control, human bad actors can hack them and produce havoc.

[1360] So, you know, that's a problem.

[1361] And you imagine hackers taking over the driverless car network and, you know, creating all kinds of havoc.

[1362] But you could also imagine given given the ease at which humans could be persuaded one way or the other, and now we have algorithms that can easily take control over that and amplify noise and move people one direction or another.

[1363] I mean, humans do that to 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.

[1364] But with machines, machines are like giant megaphones, right?

[1365] We can amplify this in orders of magnitude and fine -tune its control so we can tailor the message.

[1366] We can now very rapidly and efficiently tailor the message to the audience, taking advantage of their biases and amplifying them and using them to pursue away them in one direction or another in ways that are not fair, not logical, not objective, not meaningful.

[1367] And humans, machines empower that.

[1368] So that's what I mean by leverage.

[1369] Like, it's not new, but wow, it's powerful because machines can do it more effectively, more, you know, more quickly.

[1370] And we see that already going on in social media and other places.

[1371] That's scary.

[1372] And that's why, like, I'm, 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 and to basically so that in the end we have a stronger overall system.

[1373] that's just incredibly important.

[1374] I don't think most people understand that.

[1375] So like telling your kids or telling your students, this goes back to the cognition.

[1376] Here's how your brain works.

[1377] Here's how easy it is to trick your brain.

[1378] There are fundamental cognitive.

[1379] You should appreciate the different types of thinking and how they work and what you're prone to.

[1380] and what do you prefer and under 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 that tradeoff so if you imagine a system that is able to beyond any definition of the towing test to the benchmark really an AGI system as a thought partner that you one day will create.

[1381] What question, what topic of discussion, if you get to pick one, would you have with that system?

[1382] What would you ask and you get to find out the truth together?

[1383] So you threw me a little bit with finding the truth at the end, but because the This is a whole other topic.

[1384] But 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.

[1385] So, in other words, I can go to it and say, this is what I care about today.

[1386] And that's what we mean by, like, this general capability.

[1387] Go out, read this stuff in the next three milliseconds.

[1388] And I want to talk to you about it.

[1389] I want to draw analogies.

[1390] I want to understand how this affects this decision or that decision.

[1391] what if this were true what if that were true what what knowledge should I be aware of that could impact my decision um 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 supports this other thing boy that would that be incredible would that be just incredible just a long discourse just to be 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.

[1392] You know, I get goosebumps talking about it, right?

[1393] I mean, it's like, this is what I want.

[1394] So when you create it, please come back on the podcast and we can have a discussion together and make it even longer.

[1395] This is a record for the longest conversation.

[1396] It was an honor.

[1397] It was a pleasure, David.

[1398] Thank you so much for talking.

[1399] Thanks so much.

[1400] A lot of fun.