Jeff Hawkins: Thousand Brains Theory of Intelligence

[0] The following is a conversation with Jeff Hawkins.

[1] He's the founder of the Redwood Center for Theoretical Neuroscience in 2002 and Numenta in 2005.

[2] In his 2004 book titled On Intelligence and in the research before and after, he and his team have worked to reverse engineer the NeuroCortex and propose artificial intelligence, architectures, approaches, and ideas that are inspired by the human brain.

[3] These ideas include hierarchical temporal memory, H .T .M. from 2004 and new work, the thousands brain's theory of intelligence from 2017, 18, and 19.

[4] Jeff's ideas have been an inspiration to many who have looked for progress beyond the current machine learning approaches, but they have also received criticism for lacking a body of empirical evidence supporting the models.

[5] This is always a challenge when seeking more than small incremental steps forward in AI.

[6] Jeff is a brilliant mind, and many of the ideas he has developed and aggregated from neuroscience are worth understanding and thinking about.

[7] There are limits to deep learning as it is currently defined.

[8] Forward Progress in AI is shrouded in a mystery.

[9] My hope is that conversations like this can help provide an inspiring spark for new ideas.

[10] This is the Artificial Intelligence Podcast.

[11] If you enjoy it, subscribe on YouTube, iTunes, or simply connect with me on Twitter at Lex Friedman, spelled F -R -I -D.

[12] And now here's my conversation with Jeff Hawkins.

[13] are you more interested in understanding the human brain or in creating artificial systems that have many of the same qualities but don't necessarily require that you actually understand the underpinning workings of our mind so there's a clear answer to that question my primary interest is understanding the human brain no question about it but um i also firmly believe that we will not be able to create fully intelligent machines until we understand how the human brain works.

[14] So I don't see those are separate problems.

[15] I think there's limits so what can be done with machine intelligence if you don't understand the principles by which the brain works.

[16] And so I actually believe that studying the brain is actually the fastest way to get to machine intelligence.

[17] And within that, let me ask the impossible question.

[18] How do you not define, but at least think about what it means to be intelligent?

[19] So I didn't try to answer that question first.

[20] We said, let's just talk about how the brain works and let's figure out how it's certain parts of the brain, mostly the near cortex, but some other parts too.

[21] The parts of the brain most associated with intelligence.

[22] And let's discover the principles by how they work.

[23] Because intelligence isn't just like some mechanism and it's not just some capabilities.

[24] It's like, okay, we don't even know where to begin on this stuff.

[25] And so now that we've made a lot of progress on this, after we've made a lot of progress on how the near cortex works, and we can talk about that, I now have a very good idea what's going to be required.

[26] to make intelligent machines.

[27] I can tell you today, some of the things are going to be necessary, I believe, to create intelligent machines.

[28] So we'll get there.

[29] We'll get to the neocortex and some of the theories of how the whole thing works.

[30] And you're saying, as we understand more and more about the neocortex, about our own human mind, we'll be able to start to more specifically define what it means to be intelligent.

[31] It's not useful to really talk about that until...

[32] I don't know if it's not useful.

[33] Look, there's a long history of AI.

[34] you know.

[35] Right.

[36] And there's been different approaches taken to it.

[37] And who knows, maybe they're all useful.

[38] Right.

[39] Right.

[40] So, you know, the good old -fashioned AI, the expert systems, the current convolution neural networks, they all have their utility.

[41] They all have a value in the world.

[42] But I would think almost everyone agree that none of them are really intelligent in a set of a deep way that humans are.

[43] And so it's just the question of how do you get from where those systems were or are today to where a lot of people think we're going to go.

[44] And there's a big, big gap there, a huge gap.

[45] And I think the quickest way of bridging that gap is to figure out how the brain does that.

[46] And then we can sit back and look and say, oh, what's of these principles that the brain works on are necessary and which ones are not.

[47] Clearly, we don't have to build this and tell the machines aren't going to be built out of you know, organic living cells.

[48] But there's a lot of stuff that goes on the brain that's going to be necessary.

[49] So let me ask me, before we get into the fun details, let me ask maybe a depressing or a difficult question.

[50] Do you think it's possible that we will never be able to understand how our brain works, that maybe there's aspects to the human mind, like we ourselves cannot introspectively get to the core, that there's a wall you eventually hit?

[51] Yeah.

[52] I don't believe that's the case.

[53] I have never believed that's the case.

[54] There's not been a single thing.

[55] We've not been a single thing we've ever, humans have ever put their minds to us.

[56] We've said, oh, we reached the wall.

[57] We can't go any further.

[58] People keep saying that.

[59] People used to believe that about life, you know, Alain Vitale, right?

[60] There's like, what's the difference in living matter and non -living matter?

[61] Something special, you never understand.

[62] We no longer think that.

[63] So there's no historical evidence that's the justice is the case, and I just never even consider that's a possibility.

[64] I would also say, today, we understand so much about the near cortex.

[65] We've made tremendous progress in the last few years.

[66] that I no longer think of it as an open question.

[67] The answers are very clear to me. The pieces we know, we don't know are clear to me, but the framework is all there.

[68] And it's like, oh, okay, we're going to be able to do this.

[69] This is not a problem anymore.

[70] It just takes time and effort.

[71] But there's no mystery, a big mystery anymore.

[72] So then let's get into it for people like myself who are not very well versed in the human brain, except my own.

[73] Can you describe to me at the highest level, what are the different parts of the human brain and then zooming in on the neocortex, the parts of the neocortex, and so on, a quick overview?

[74] Yeah, sure.

[75] A human brain, we can divide it roughly into two parts.

[76] There's the old parts, lots of pieces, and then there's a new part.

[77] The new part is the neocortex.

[78] It's new because it didn't exist before mammals.

[79] The only mammals have a neocortex, and in humans, it's primates.

[80] is very large.

[81] In a human brain, the neocortex occupies about 70 to 75 percent of the volume of the brain.

[82] It's huge.

[83] And the old parts of the brain are, there's lots of pieces there.

[84] There's a spinal cord, and there's the brain stem, and the cerebellum, and the different parts of the basal ganglia and so on.

[85] In the old parts of the brain, you have the autonomic regulation, like breathing and heart rate.

[86] You have basic behaviors.

[87] So like walking and running are controlled by the old parts of the brain.

[88] All the emotional centers of the brain are in the old part brain.

[89] So when you feel anger or hungry, lust, or things like that, those are all in the whole parts of the brain.

[90] And we associate with the neocortex, all the things we think about as sort of high -level perception and cognitive functions, anything from seeing and hearing and touching things to language, to mathematics and engineering and science, and so on.

[91] Those are all associated with the neocortex.

[92] And they're certainly correlated.

[93] Our abilities in those regards are correlated with the relative size of our neocortex compared to other mammals.

[94] So that's like the rough division, and you obviously can't understand the neocortex completely isolated, but you can understand a lot of it with just a few interfaces, so the all parts of the brain.

[95] And so it gives you a system to study.

[96] The other remarkable thing about the neocortex compared to the old parts of the brain is the neocortex is extremely uniform.

[97] it's not visibly or anatomically or it's very it's like a i always like to say it's like the size of a dinner napkin about two and a half millimeters thick and it looks remarkably the same everywhere everywhere you look in that two and a half millimeters is this detailed architecture and it looks remarkably the same everywhere and that's across species a mouse versus a cat and a dog and a human where if you look at the old parts of the brain there's lots of little pieces do specific things So it's like the old parts of a brain evolved, like, this is the part that controls heart rate, and this is the part that controls this, and this is this kind of thing, and that's this kind of thing.

[98] And these evolved for eons a long, long time, and they have their specific functions.

[99] And all of a sudden, mammals come along, and they got this thing called the near cortex, and it got large by just replicating the same thing over and over and over again.

[100] This is like, wow, this is incredible.

[101] So all the evidence we have, and this is an idea that was first articulated, in a very cogent and beautiful argument by a guy named Vernon Malkassel in 1978, I think it was, that the neocortex all works on the same principle.

[102] So language, hearing, touch, vision, engineering, all these things are basically underlying or all built in the same computational substrate.

[103] They're really all the same problem.

[104] So the low level of the building blocks all look similar.

[105] Yeah, and they're not even that low level.

[106] We're not talking about like neurons.

[107] We're talking about this very complex circuit that exists throughout the near cortex is remarkably similar.

[108] It is, it's like, yes, you see variations of it here and there, more of the cell, that's not all, and so on.

[109] But what Mountcast argued was, it says, you know, if you take a section of near cortex, why is one a visual area and one is a auditory area?

[110] And his answer was, it's because one is connected to eyes and one is connected to ears.

[111] Literally, you mean just it's most closest in terms of the number of connections to the sensor.

[112] Literally, if you took the optic nerve and attached it to a different part of the neocortex, that part would become a visual region.

[113] This actually, this experiment was actually done by Morgonka -sur.

[114] Oh, boy.

[115] In developing, I think it was lemurs, I can't remember what it was, or some animal.

[116] And there's a lot of evidence to this.

[117] You know, if you take a blind person, a person is born blind at birth, they're born with a visual neocortex.

[118] It doesn't, may not get any input from the eyes because of some congenital defect or something.

[119] And that region becomes, does something else.

[120] It picks up another task.

[121] So, and it's, it's, so it's this very complex thing.

[122] It's not like, oh, they're all built on neurons.

[123] No, they're all built in this very complex circuit.

[124] And, uh, and somehow that circuit underlies everything.

[125] And so this is, uh, the, it's called the common.

[126] cortical algorithm, if you will.

[127] Some scientists just find it hard to believe, and they just, I can't believe that's true, but the evidence is overwhelming in this case.

[128] And so a large part of what it means to figure out how the brain creates intelligence and what is intelligence in the brain is to understand what that circuit does.

[129] If you can figure out what that circuit does, as amazing as it is, then you can, then you understand what all these other cognitive functions are.

[130] So if you were to sort of put neurocortex outside of your book on intelligence, if you wrote a giant tome, a textbook on the neocortex, and you look maybe a couple centuries from now, how much of what we know now would still be accurate two centuries from now?

[131] So how close are we in terms of understanding?

[132] So I have to speak from my own particular experience here.

[133] So I run a small research lab here.

[134] It's like any other research lab.

[135] I'm sort of the principal investigator.

[136] There's actually two of us and there's a bunch of other people.

[137] And this is what we do.

[138] We started the near cortex and we published our results and so on.

[139] So about three years ago, we had a real breakthrough in this in this field.

[140] It's a tremendous breakthrough.

[141] We've now published, I think, three papers on it.

[142] And so I have a pretty good understanding of all the pieces and what we're missing.

[143] I would say that almost all the empirical data we've collected about the brain, which is enormous.

[144] If you don't know the neuroscience literature, it's just incredibly big.

[145] And it's, for the most part, all correct.

[146] It's facts and experimental results and measurements and all kinds of stuff.

[147] But none of that has been really assimilated into a theoretical framework.

[148] It's data without, in the language of Thomas Kuhn, the historian, it would be sort of a pre -paradigm science.

[149] Lots of data, but a way to fit it together.

[150] I think almost all of that's correct.

[151] There's going to be some mistakes in there.

[152] And for the most part, there aren't really good cogent theories about how to put it together.

[153] It's not like we have two or three competing good theories, which ones are right and which ones are wrong.

[154] It's like, yeah, people are just like scratching their heads, throwing things.

[155] You know, some people have given up on trying to like figure out what the whole thing does.

[156] In fact, there's very, very few labs that we do that focus really on theory and all this unassimilated data and trying to explain it.

[157] so it's not like we have we've got it wrong it's just that we haven't got it at all so it's really i would say pretty early days in terms of understanding the fundamental theories forces of the way our mind works i don't think so that i would have said that's true five years ago so as i said we had some really big breakthroughs on this recently and we started publishing papers on this so um uh so we'll get we'll get to there but so i don't think it's i you know i'm i'm an officer And from where I sit today, most people would disagree with this.

[158] But from where I sit today, from what I know, it's not super early days anymore.

[159] We are, you know, the way these things go is it's not a linear path, right?

[160] You don't just start accumulating and get better and better and better.

[161] No, you've got all this stuff you've collected.

[162] None of it makes sense.

[163] All these different things are just starting around.

[164] And then you're going to have some breaking points all of a sudden, oh my God, now we got it, right?

[165] That's how it goes in science.

[166] And I personally feel like we passed that little thing about a couple years ago.

[167] all that big thing a couple years ago.

[168] So we can talk about that.

[169] Time will tell if I'm right, but I feel very confident about it.

[170] That's what I'm willing to say it on tape like this.

[171] At least very optimistic.

[172] So let's, before those few years ago, let's take step back to H -TM, the hierarchical temporal memory theory, which you first proposed on intelligence and went through a few different generations.

[173] Can you describe what it is, how it evolved through the three generations since you first put it on paper?

[174] Yeah.

[175] So one of the things that neuroscientists just sort of missed for many, many years, and especially people who were thinking about theory, was the nature of time in the brain.

[176] Brains process information through time.

[177] The information coming into the brain is constantly changing.

[178] The patterns from my speech right now, if you're listening to it at normal speed, would be changing on your ears about every 10 milliseconds or so.

[179] You'd have a change.

[180] This constant flow.

[181] When you look at the world, your eyes are moving constantly, three to five times a second, and the input is completely.

[182] If I were to touch something like a coffee cup, as I move my fingers, the input changes.

[183] So this idea that the brain works on time -changing patterns is almost completely, or was almost completely missing from a lot of the basic theories like fears of vision and so on.

[184] It's like, oh, no, we're going to put this image in front of you and flash it and say, what is it?

[185] Convolutional neural networks work that way today, right?

[186] You know, classify this picture.

[187] But that's not what vision is like.

[188] Vision is this sort of crazy time -based pattern that's going all over the place, and so is touched, and so is hearing.

[189] So the first part of hierarchical temporal memory was the temporal part.

[190] It's to say you won't understand the brain, nor do you understand intelligent machines unless you're dealing with time -based patterns.

[191] The second thing was the memory component of it was, is to say that we aren't just processing input.

[192] We learn a model of the world, and the memory stands for that model.

[193] We have to, the point of the brain, the part of the neocortex, it learns a model of the world.

[194] We have to store things that our experiences in a form that leads to a model of the world.

[195] So we can move around the world, we can pick things up and do things and navigate and know how it's going on.

[196] So that's what the memory referred to.

[197] And many people just, they were thinking about like certain processes without memory at all.

[198] They're just like processing things.

[199] And then finally, the hierarchical component was a reflection to that the neocortex, although it's just uniformed sheet of cells, different parts of it project to other parts, which project to other parts.

[200] And there is a sort of a rough hierarchy in terms of that.

[201] So the hierarchical temporal memory is just saying, look, we should be thinking about the brain as time -based, you know, model memory -based, and hierarchical processing.

[202] And that was a placeholder for a bunch of components that we would then plug into that.

[203] We still believe all those things I just said, but we now know so much more.

[204] that I'm stopping to use the word hierarchal temper memory yet because it's insufficient to capture the stuff we know.

[205] So again, it's not incorrect, but it's, I now know more, and I would rather describe it more accurately.

[206] Yeah, so you're basically, we can think of H -TM as emphasizing that there's three aspects of intelligence that are important to think about whatever the eventual theory converges to.

[207] So in terms of time, how do you think of nature of time across different time scales?

[208] So you mentioned things changing, sensory input is changing every 10.

[209] What about it every few minutes, every few months and years?

[210] Well, if you think about a neuroscience problem, the brain problem, neurons themselves can stay active for certain periods of time.

[211] They're parts of the brain where they're active for minutes.

[212] You know, so you could hold a certain perception or an activity for a certain period of time.

[213] But most of them don't last that long.

[214] And so if you think about your thoughts are the activity neurons, if you're going to want to involve something that happened a long time ago, even just this morning, for example, the neurons haven't been active throughout that time.

[215] So you have to store that.

[216] So by I ask you, what did you have for breakfast today?

[217] That is memory.

[218] That is, you've built it into your model of the world now.

[219] You remember that.

[220] And that memory is in the synapses, is basically in the formation of synapses.

[221] And so it's, it's, you're still.

[222] sliding into what, you know, you said different timescales.

[223] There's timescales of which we are like understanding my language and moving about and seeing things rapidly over time.

[224] That's the time scales of activities and neurons.

[225] But if you want to get longer timescales, then it's more memory.

[226] And we have to invoke those memories to say, oh, yes, well, now I can remember what I had for breakfast because I stored that someplace.

[227] I may forget it tomorrow, but it's stored for now.

[228] So this memory also need to have, So the hierarchical aspect of reality is not just about concepts, it's also about time?

[229] Do you think of it that way?

[230] Yeah, time is infused in everything.

[231] It's like you really can't separate it out.

[232] If I ask you, what is your, you know, how's the brain learn a model of this coffee cup here?

[233] I have a coffee cup, then I have a coffee cup.

[234] I say, well, time is not an inherent property of this, of the model I have of this cup, whether it's a visual model or a tactile model.

[235] I can sense it through time, but the model self doesn't really have much time.

[236] If I asked you, if I said, well, what is the model of my cell phone?

[237] My brain has learned a model of the cell phone.

[238] So if you have a smartphone like this, and I said, well, this has time aspects to it.

[239] I have expectations when I turn it on, what's going to happen, what are how long it's going to take to do certain things.

[240] If I bring up an app, what sequences.

[241] And so I have, and it's like melodies in the world.

[242] Melody has a sense of time.

[243] So many things in the world move and act, and there's a sense of time related to them.

[244] Some don't, but most things do actually.

[245] So it's sort of infused throughout the models of the world.

[246] You build a model of the world, you're learning the structure of the objects in the world, and you're also learning how those things change through time.

[247] Okay, so it really is just a fourth dimension that's infused deeply.

[248] And you have to make sure that your models of intelligence, Inc. Incorporated.

[249] So, like you mentioned, the state of neuroscience is deeply empirical.

[250] A lot of data collection, it's, you know, that's where it is.

[251] You mentioned Thomas Coon, right?

[252] Yeah.

[253] And then you're proposing a theory of intelligence, which is really the next step, the really important step to take.

[254] But why is H .T .M. or what we'll talk.

[255] talk about soon the right theory.

[256] So is it more in this, is it backed by intuition?

[257] Is it backed by evidence?

[258] Is it backed by a mixture of both?

[259] Is it kind of closer to where string theory is in physics where there's mathematical components which show that, you know what, it seems that this fits together too well for not to be true, which is what, where string theory is.

[260] Is that where you're kind of sick of all those things although definitely where we are right now it's definitely much more on the empirical side than let's say string theory the way this goes about we're theorists right so we look at all this data and we're trying to come up with some sort of model that explains it basically and there's unlike string theory there's this vast more amounts of empirical data here that I think than most physicists deal with and so our challenge is to sort through that and figure out what kind of constructs would explain this.

[261] And when we have an idea, you come up with a theory of some sort, you have lots of ways of testing it.

[262] First of all, there are 100 years of assimilated, unassimulated empirical data from neuroscience.

[263] So we go back and read papers.

[264] And we say, oh, well, did someone find this already?

[265] We can predict X, Y, and Z. And maybe no one's even talked about it since 1972 or something, but we go.

[266] go back and find that.

[267] And we say, oh, either it can support the theory or it can invalidate the theory.

[268] And then we say, okay, we have to start over again.

[269] Oh, no, it's supported.

[270] Let's keep going with that one.

[271] So the way I kind of view it, when we do our work, we come up, we look at all this empirical data, and it's what I call it is a set of constraints.

[272] We're not interested in something that's biologically inspired.

[273] We're trying to figure out how the actual brain works.

[274] So every piece of empirical data is a constraint on a theory.

[275] If you have the correct theory, to explain every pin right so we have this huge number of constraints on the problem which initially makes it very very difficult if you don't have many constraints you can make up stuff all the day you can say oh here's an answer how you can do this you can do that you can do this but if you consider all biology as a set of constraints all neuroscience a set of constraints and even if you're working in one little part of the neocortex for example there are hundreds and hundreds of constraints these are empirical constraints that it's very very difficult initially to come up with a theoretical framework for that but But when you do, and it solves all those constraints at once, you have a high confidence that you got something close to correct.

[276] It's just mathematically almost impossible not to be.

[277] So that's the curse and the advantage of what we have.

[278] The curses, we have to meet all these constraints, which is really hard.

[279] But when you do meet them, then you have a great confidence that you've discovered something.

[280] In addition, then we work with scientific labs, so we'll say, oh, there's something we can't find, we can predict something, but we can't find it anywhere in the literature.

[281] So we will then, we have people we've collaborated with, we'll say, sometimes they'll say, you know what, I have some collected data, which I didn't publish, but we can go back and look in it and see if we can find that, which is much easier than designing a new experiment.

[282] You know, neuroscience experiments take a long time, years.

[283] So, although some people are doing that now, too.

[284] So, but between all of these things, I think it's a reasonable, it's actually a very, very good approach.

[285] We are blessed with the fact that we can test our theories out to yingyang here because there's so much unassimilar data.

[286] And we can also falsify our theories very usually, which we do often.

[287] So it's kind of reminiscent to whenever that was with Copernicus, you know, when you figure out that the sun is at the center of the solar system as opposed to Earth, the pieces just fall into place.

[288] Yeah, I think that's the general nature of aha moments is, and it's Copernicus, it could be, you could say the same thing about Darwin, you could say the same thing about, you know, about the double helix, that people have been working on a problem for so long and have all this data and they can't make sense of it, they can't make sense of it.

[289] But when the answer comes to you and everything falls into place, it's like, oh my gosh, that's it.

[290] that's got to be right.

[291] I asked both Jim Watson and Francis Crick about this.

[292] I asked them, you know, when you were working on trying to discover the structure of the double helix, and when you came up with the sort of the structure that ended up being correct, but it was sort of a guess, you know, it wasn't really verified yet, I said, did you know that it was right?

[293] And they both said, absolutely.

[294] that we absolutely knew it was right.

[295] And it doesn't matter if other people didn't believe it or not, we knew it was right, they'd get around to thinking it and agree with it eventually anyway.

[296] And that's the kind of thing you hear a lot with scientists who really are studying a difficult problem.

[297] And I feel that way too about our work.

[298] Have you talked to Crick or Watson about the problem you're trying to solve of finding the DNA of the brain?

[299] Yeah.

[300] In fact, Francis Crick was very interested in this.

[301] in the latter part of his life.

[302] And in fact, I got interested in brains by reading an essay he wrote in 1979 called Thinking About the Brain.

[303] And that was when I decided I'm going to leave my profession of computers and engineering and become a neuroscientist, just reading that one essay from Francis Craig.

[304] I got to meet him later in life.

[305] I spoke at the Salk Institute, and he was in the audience, and then I had a tea with him afterwards.

[306] You know, he was interested in a different problem.

[307] He was, he was focused on consciousness.

[308] The easy problem, right.

[309] Well, I think it's the red herring.

[310] And so we weren't really overlapping a lot there.

[311] Jim Watson, who's still alive, is also interested in this problem.

[312] And he was, when he was director of the Kulseman Harbor Labor laboratories, he was really sort of behind moving in the direction of neuroscience there.

[313] And so he had a personal interest in this field.

[314] And I have met with him numerous times.

[315] And in fact, the last time was a little bit over a year ago, I gave a talk at Cold Spring Harbor Labs about the progress we were making in our work.

[316] And it was a lot of fun because he said, well, you wouldn't be coming here unless you had something important to say.

[317] So I'm going to go attend your talk.

[318] So he sat in the very front row.

[319] Next him was the director of the lab, Bruce Stillman.

[320] So these guys were in the front row of this auditorium, right?

[321] So nobody else in the auditorium wants to sit in the front row because there's Jim Watson and there's the director.

[322] And I gave a talk and then I had dinner with Jim afterwards.

[323] But there's a great picture of my colleague, Subatio Modtuk, where I'm up there sort of screaming the basics of this new framework we have.

[324] And Jim Watson's on the edge of his chair.

[325] He's literally on the edge of his chair, like intently staring up at the screen.

[326] And when he discovered the structure of DNA, the first public talk he gave was at Cold Spring Harbor Labs.

[327] And there's a picture, there's a famous picture of Jim Watson standing at the whiteboard with an over -ed thing, pointing at something with holding a double.

[328] He looks at his pointer.

[329] And it actually looks a lot like the picture of me. So it was sort of funny.

[330] There's an AI up talking about the brain and there's Jim Watson staring at it, and of course, they were with, you know, whatever, 60 years earlier, he was standing, you know, pointing at the double helix.

[331] It's one of the great discoveries in all of, you know, whatever, biology, science, all of science is DNA.

[332] So it's funny that there's echoes of that in your presentation.

[333] Do you think in terms of evolutionary timeline and history, the development of the neocortex was a big leap?

[334] Or is it just a small step?

[335] So, like, if we ran the whole thing over again from the birth of life on Earth, how likely we would we develop the mechanism in the New York?

[336] Okay, well, those are two separate questions.

[337] One is, was it a big leap, and one was how like it is, okay?

[338] They're not necessarily related.

[339] Maybe correlated.

[340] Maybe correlated, maybe not, and we don't really have enough data to make a judgment about that.

[341] I would say definitely was a big leap, and I can tell you why.

[342] I don't think it was just another incremental step.

[343] I'll get that at the moment.

[344] I don't really have any idea how likely it is.

[345] If we look at evolution, we have one data point, which is Earth, right?

[346] Life formed on Earth billions of years ago, whether it was introduced here or it created it here or someone introduced it, we don't really know, but it was here early.

[347] It took a long, long time to get to multicellular life.

[348] And then for multi -cellular life, it took a long, long time to get the Xenotortex.

[349] And we've only had the New York Tractics for a few hundred thousand years.

[350] so that's like nothing okay so is it likely well certainly isn't something that happened right away on earth and there were multiple steps to get there so i would say it's probably not got something that would happen instantaneously on other planets that might have life it might take several billion years on average um is it likely i don't know but you'd have to survive for several billion years to find out probably um is it a big leap yeah i think it's um it is a quality difference in all other evolutionary steps.

[351] I can try to describe that if you're like.

[352] Sure, in which way?

[353] Yeah, I can tell you how.

[354] Pretty much, let's start with a little preface.

[355] Many of the things that humans are able to do do not have obvious survival advantages precedent.

[356] You know, we create music.

[357] Is there a really survival advantage to that?

[358] Maybe, maybe not.

[359] What about mathematics?

[360] Is there a real survival advantage to mathematics?

[361] You can stretch it.

[362] You can try to figure these things out, right?

[363] But most of evolutionary history, everything had immediate survival advantages to it.

[364] So I'll tell you a story, which I like, may not be true.

[365] But the story goes as follows.

[366] Organisms have been evolving since the beginning of life here on Earth, adding this sort of complexity onto that and this sort of complex on to that.

[367] And the brain itself is evolved this way.

[368] In fact, there's an old parts and older, older parts of the brain that kind of just keeps calming on new things and we keep adding capabilities.

[369] When we got to the neocortex, initially it had a very clear survival advantage in that it produced better vision and better hearing and better touch and maybe it would say so on.

[370] But what I think happens is that evolution discovery took a mechanism.

[371] And this is in our recent theories, but it took a mechanism that evolved a long time ago for navigating in the world, for knowing where you are.

[372] These are the so -called grid cells and place cells of an old part of the brain.

[373] And it took that mechanism for building maps of the world and knowing where you are on those maps and how to navigate those maps and turns it into a sort of a slim -down idealized version of it.

[374] And that idealist version could now apply to building maps.

[375] other things, maps of coffee cups, maps of phones, maps of mathematics, concepts, yes, and not just almost, exactly.

[376] And so you, and it just started replicating this stuff, right?

[377] You just think more and more and more.

[378] So we went from being sort of dedicated purpose neural hardware to solve certain problems that are important to survival, to a general purpose neural hardware that could be applied to all problems.

[379] And now it's escaped the orbit of survival.

[380] It's, we are now able to apply it to things which we find enjoyment, you know, but aren't really clearly survival characteristics.

[381] And that, it seems to only have happened in humans to the large extent.

[382] And so that's what's going on where we sort of have, we've sort of escaped the gravity of evolutionary pressure in some sense in the near cortex.

[383] and it now does things which are not that are really interesting discovering models of the universe which may not really help us.

[384] It doesn't matter.

[385] How is it help us surviving knowing that there might be multiverses or there might be, you know, the age of the universe or how do, you know, various stellar things occur?

[386] It doesn't really help us survive at all.

[387] But we enjoy it and that's what happened.

[388] Or at least not in the obvious way, perhaps.

[389] It is required.

[390] if you look at the entire universe in an evolutionary way, it's required for us to do interplanetary travel and therefore survive past our own son.

[391] But, you know, let's not get too quick.

[392] Yeah, but, you know, evolution works at one time frame.

[393] Yes.

[394] It's survival.

[395] If you think of survival of the phenotype, survival of the individual.

[396] Exactly.

[397] What you're talking about there is spans well beyond that.

[398] So there's no genetic, I'm not transferring any genetic traits to my children that are going to help them survive.

[399] I better on Mars.

[400] Right.

[401] Totally different mechanism.

[402] That's right.

[403] So let's get into the new, as you've mentioned, this idea.

[404] I don't know if you have a nice name, 1000.

[405] You call it the Thousand Brain Theory of Intellectuality.

[406] I like it.

[407] So can you talk about this idea of spatial view of concepts and so on?

[408] Yeah.

[409] So can I just describe sort of the, there's an underlying core discovery, which then everything comes from that.

[410] That's a very simple, this is really what happened.

[411] we were deep into problems about understanding how we build models of stuff in the world and how we make predictions about things.

[412] And I was holding a coffee cup just like this in my hand.

[413] And my finger was touching the side, my index finger.

[414] And then I moved it to the top, and I was going to feel the rim at the top of the cup.

[415] And I asked myself a very simple question.

[416] I said, well, first of all, let's say, I know that my brain predicts what it's going to feel before it touches it.

[417] You can just think about it and imagine it.

[418] And so we know that the brain's making predictions all the time.

[419] So the question is, what does it take to predict that?

[420] And there's a very interesting answer.

[421] First of all, it says the brain has to know it's touching a coffee cup.

[422] It has to have a model of a coffee cup.

[423] It needs to know where the finger currently is on the cup, relative to the cup.

[424] Because when I make a movement, it needs to know where it's going to be on the cup after the movement is completed relative to the cup.

[425] And then it can make a prediction about what it's going to sense.

[426] So this told me that the neocortex, which is making this prediction, needs to know that it's sensing, it's touching a cup, and it needs to know the location of my finger relative to that cup in a reference frame of the cup.

[427] It doesn't matter where the cup is relative to my body.

[428] It doesn't matter its orientation.

[429] None of that matters.

[430] It's where my finger is relative to the cup, which tells me then that the neocortex has a reference frame that's anchored to the cup, because otherwise I wouldn't be able to say the location, and I wouldn't be able to predict my new location.

[431] And then we quickly, very instantly, you can say, well, every part of my skin could touch this cup, and therefore every part of my skin is making predictions, and every part of my skin must have a reference frame that it's using to make predictions.

[432] So the big idea is that throughout the near cortex, there are, everything is being stored and referenced in reference frames.

[433] You can think of them like XYZ reference frames, but they're not like that.

[434] We know a lot about the neural mechanisms for this, but the brain thinks in reference frames.

[435] And as an engineer, if you're an engineer, this is not surprising.

[436] You'd say if I wanted to build a CAD model of the coffee cup, well, I would bring it up in some CAD software, and I would assign some reference fame and say this features at these locations and so on.

[437] But the fact that this, the idea that this is occurring throughout the near cortex everywhere, it was a novel idea.

[438] And then a zillion things fell into place after that, a zillion.

[439] So now we think about the near cortex as processing information quite differently than we used to do it.

[440] We used to think about the near cortex as processing sensory data and extracting features from that sensory data and then extracting features from the features, very much like a deep learning network does today.

[441] But that's not how the brain works at all.

[442] The brain works by assigning everything, every input, everything to reference frames.

[443] And there are thousands, hundreds, thousands of them active at once in your neocortex.

[444] It's a surprising thing to think about, but once you sort of internalize this, you understand that it explains almost every, almost all the mysteries we've had about this structure.

[445] So one of the consequences of that is that every small part of the near cortex, say a millimeter square, and there's 150 ,000 of those.

[446] So it's about 150 ,000 square millimeters.

[447] If you take every little square millimeter of the cortex, it's got some input coming into it, and it's going to have reference frames where it's assigning that input to.

[448] and each square millimeter can learn complete models of objects.

[449] So what do I mean by that?

[450] If I'm touching the coffee cup, well, if I just touch it in one place, I can't learn what this coffee cup is because I'm just feeling one part.

[451] But if I move it around the cup and touch it at different areas, I can build up a complete model of the cup because I'm now filling in that three -dimensional map, which is the coffee cup.

[452] I can say, oh, what are I feeling at all these different locations?

[453] That's the basic idea.

[454] It's more complicated than that.

[455] But so through time, and we talked about time earlier, through time, even a single column, which is only looking at, or a single part of the cortex, which is only looking at a small part of the world, can build up a complete model of an object.

[456] And so if you think about the part of the brain, which is getting input from all my fingers, so they're spread across the top of your head here.

[457] This is the somatosensory cortex.

[458] There's columns associated all the different areas of my skin.

[459] And what we believe is happening is that all of them are building models of this cup, every one of them, or things.

[460] They're not all building all, not every column or every part of the cortex builds models of everything, but they're all building models of something.

[461] And so you have, so when I, when I touch this cup with my hand, there are multiple models of the cup being invoked.

[462] If I look at it with my eyes, there are again many models of the cup being invoked because each part of the visual system, the brain doesn't process an image.

[463] That's a misleading idea.

[464] It's just like your fingers touching the cup, so different parts of my retina are looking at different parts of the cup.

[465] and thousands and thousands of models of the cup are being invoked at once and they're all voting with each other trying to figure out what's going on.

[466] So that's why we call it the Thousand Brain's Theory of Intelligence because there isn't one model of a cup.

[467] There are thousands of models of this cup.

[468] There are thousands of models of your cell phone and about cameras and microphones and so on.

[469] It's a distributed modeling system which is very different than what people have thought about it.

[470] So that's a really compelling and interesting idea.

[471] I have two first questions.

[472] So one, on the ensemble part of everything coming together, you have these thousand brains, how do you know which one has done the best job of forming the great question.

[473] Let me try explain.

[474] There's a problem that's known in neuroscience called the sensor fusion problem.

[475] Yes.

[476] And so the idea is there's something like, oh, the image comes from the eye.

[477] There's a picture on the retina.

[478] And it gets projected to the near cortex.

[479] Oh, by now, it's all sped out all over the place.

[480] And it's kind of squirrely and distorted.

[481] And pieces are all of it, you know, it doesn't look like a picture anymore.

[482] When does it all come back together again, right?

[483] Or you might say, well, yes, but I also, I also have sounds or touches associated with the cup.

[484] So I'm seeing the cup and touching the cup.

[485] How do they get combined together again?

[486] So this, it's called the sensor fusion problem, as if all these disparate parts have to be brought together into one model someplace.

[487] That's the wrong idea.

[488] The right idea is that you get all these guys voting.

[489] There's auditory models of the cup, there's visual models of the cup, those tactile models of the cup.

[490] In the vision system, there might be ones that are more focused on black and white, one's version of color.

[491] It doesn't really matter.

[492] There's just thousands and thousands of models of this cup.

[493] And they vote.

[494] They don't actually come together in one spot.

[495] Just literally think of it this way.

[496] Imagine you have these columns are like about the size of a little piece of spaghetti, okay, like a two and a half millimeters tall and about a millimeter in white.

[497] They're not physical like, but you can think of them that way.

[498] And each one's trying to guess what this thing is or touching.

[499] Now, they can do a pretty good job if they're allowed to move over touch.

[500] So I can reach my hand into a black box and move my finger around an object and if I touch enough spaces like, okay, I know I know what it is.

[501] But often we don't do that.

[502] Often I can just reach and grab something with my hand all at once and I get it.

[503] Or if I had to look through the world through a straw, so I'm only invoking one little column, I can only see a part of something because I have to move the straw around.

[504] But if I open my eyes, I see the whole thing at once.

[505] So what we think is going on is all these little pieces of spaghetti, if you will, all these little columns in the cortex or all trying to guess what it is that they're sensing.

[506] They'll do a better guess if they have time and can move over time, so if I move my eyes and move my fingers.

[507] But if they don't, they have a poor guess.

[508] It's a, it's a probabilistic guess of what they might be touching.

[509] Now, imagine they can post their probability at the top of little piece of spaghetti.

[510] Each one of them says, I think, and it's not really a probability distribution.

[511] It's more like a set of possibilities.

[512] In the brain, it doesn't work as a probability distribution works is more like what we call a union.

[513] You could say, and one column says, I think it could be a coffee cup, a soda can, or a water bottle.

[514] Another column says, I think it could be a coffee cup or, you know, telephone or camera or whatever, right?

[515] And all these guys are saying what they think it might be.

[516] And there's these long -range connections in certain layers in the cortex.

[517] So there's some layers in some cells types in each column send the projections across the brain.

[518] And that's the voting occurs.

[519] And so there's a simple associative memory mechanism.

[520] We've described this in a recent paper and we've modeled this that says they can all quickly settle on the only or the one best answer for all of them.

[521] If there is a single best answer, they all vote and say, yep, it's got to be the coffee cup.

[522] And at that point, they all know it's a coffee cup.

[523] And at that point, everyone acts as if it's the coffee cup.

[524] Yep, we know it's a coffee.

[525] Even though I've only seen one little piece of this world, I know it's a coffee cup I'm touching or I'm seeing or whatever.

[526] And so you can think all these columns are looking at different parts in different places, different sensory input, different locations.

[527] They're all different.

[528] But this layer that's doing the voting, it solidifies.

[529] It's just like it crystallizes and says, oh, we all know what we're doing.

[530] And so you don't bring these models together in one model.

[531] You just vote and there's a crystallization of the vote.

[532] Great.

[533] That's at least a compelling way to think about the way you form a model of the world.

[534] Now, you talk about a coffee cup.

[535] Do you see this, as far as I understand, that you are proposing this as well, that this extends to much more than coffee cups.

[536] Yeah, it does.

[537] Or at least the physical world.

[538] It expands to the world of concepts.

[539] Yeah, it does.

[540] And, well, the first, the prima facie evidence for that is that the regions of the neocortex that are associated with language or high -level thought or mathematics or things like that, they look like the regions of the neocortex that process vision and hearing and touch.

[541] They don't look any different, or they look only marginally different.

[542] And so one would say, well, if Vernon Mouncastle, who proposed it all the parts in the court are doing the same thing, if he's right, then the parts that are doing language or mathematics or physics are working on the same principle.

[543] They must be working on the principle of reference frames.

[544] So that's a little odd thought.

[545] So, hmm, but of course, we had no prior idea how these things happen.

[546] So let's go with that.

[547] and we in our recent paper we talked a little bit about that I've been working on it more since I have better ideas about it now I'm sitting here very confident that that's what's happening and I can give you some examples to help you think about that it's not we understand it completely but I understand it better than I've described it in any paper so far so but we did put that idea out there says okay this is it's it's a good place to start you know and the evidence would suggest that's how it's happening and then we can start tackling that problem one piece at a time like what does it mean to do high level thought what it means to do language how would that fit into a reference frame framework yeah so there's a there's a I don't know if you could tell me if there's a connection but there's an app called Anki that helps you remember different concepts and they they talk about like a memory palace that helps you remember completely random concepts by sort of trying to put them in a physical space in your mind yeah and putting them next to each other called the method of loci loci yeah for some reason that seems to work really well.

[548] Now that's a very narrow kind of application of just remembering some facts.

[549] But that's not, but that's a very, very telling one.

[550] Yes, exactly.

[551] So this seems like you're describing a mechanism why this seems to work.

[552] So, so basically the way, what we think is going on is all things you know, all concepts, all ideas, words, everything, you know, are stored in reference frames.

[553] And, and so if you want to remember something, you have to basically navigate through a reference frame the same way a rat navigates to a mave and the same way my finger rat navigates to this coffee cup you're moving through some space and so if you have a random list of things you ask to remember by assigning them to a reference frame you've already know very well to see your house right and the idea of the method of loci is you can say okay in my lobby I'm going to put this thing and then and then the bedroom I put this one I go down the hall I put this thing and then you want to recall those facts or recall those things you just walk mentally you walk through your house, you're mentally moving through a reference frame that you already had.

[554] And that tells you, there's two things that really important about that.

[555] It tells us the brain prefers to store things in reference frames and that the method of recalling things or thinking, if you will, is to move mentally through those reference frames.

[556] You could move physically through some reference frames, like I could physically move through the reference frame of this coffee cup.

[557] I can also mentally move through the reference frame of the coffee cup, imagining me touching it, but I can also mentally move my house.

[558] And so now we can ask yourself, or are all concepts stored this way?

[559] There was some recent research using human subjects in fMRI, and I'm going to apologize for not knowing the name of the scientists who did this.

[560] But what they did is they put humans in this fMRI machine, which was one of these imaging machines, and they gave the humans tasks to think about birds.

[561] So they had different types of birds and and the rich that look big and small and long necks and long legs, things like that.

[562] And what they could tell from the fMRI was a very clever experiment.

[563] You get to tell when humans were thinking about the birds, that the knowledge of birds was arranged in a reference frame similar to the ones that are used when you navigate in a room.

[564] These are called grid cells, and there are grid cell -like patterns of activity in the neocortex when they do this.

[565] So that, it's very clever.

[566] experiment, you know, and what it basically says is that even when you're thinking about something abstract and you're not really thinking about it as a reference frame, it tells us the brain is actually using a reference frame.

[567] And it's using the same neural mechanisms.

[568] These grid cells are the basic same neural mechanisms that we propose that grid cells, which exist in the old part of the brain, the entorionic cortex, that that mechanism is now similar mechanism is used throughout the near cortex.

[569] It's the same nature to preserve this interesting way of creating reference frames.

[570] And so now they have empirical evidence that when you, you you think about concepts like birds that you're using reference frames that are built on grid cells.

[571] So that's similar to the method of loci, but in this case, the birds are related so they create their own reference frame, which is consistent with bird space.

[572] And when you think about something, you go through that, you can make the same example.

[573] Let's take a mathematics.

[574] Let's say you want to prove a conjecture.

[575] Okay.

[576] What is a conjecture?

[577] Conjecture is a statement you believe to be true, but you haven't proven it.

[578] And so it might be an equation.

[579] I want to show that this is equal to that.

[580] And you have some places you start with.

[581] You say, well, I know this is true and I know this is true.

[582] And I think that maybe to get to the final proof, I need to go through some intermediate results.

[583] What I believe is happening is literally these equations or these points are assigned to a reference frame, a mathematical reference frame.

[584] And when you do mathematical operations, a simple one might be multiply or divide, but you might be a little plus transform or something else, that is like a movement in the reference frame of the math.

[585] And So you're literally trying to discover a path from one location to another location in a space of mathematics.

[586] And if you can get to these intermediate results, then you know your map is pretty good, and you know you're using the right operations.

[587] Much of what we think about is solving hard problems is designing the correct reference frame for that problem, figuring out how to organize the information and what behaviors I want to use in that space to get me there.

[588] Yeah, so if you dig in an idea of this reference frame, whether it's the math, you start a set of axioms to try to get to proving the conjecture, can you try to describe, maybe taking a step back, how you think of the reference frame in that context?

[589] Is it the reference frame that the axioms are happy in?

[590] Is it the reference frame that might contain everything?

[591] Is it a changing thing?

[592] So you have many, many reference frames.

[593] I mean, in fact, the way the theory, the 1 ,000 Brain Theory of Intelligence says that every single thing in the world has its own reference frames.

[594] So every word has its own reference names.

[595] And we can talk about this.

[596] The mathematics work out, this is no problem for neurons to do this.

[597] But how many reference frames does the coffee cup have?

[598] Well, it's on a table.

[599] Let's say you ask how many reference frames could a column in my finger that's touching the coffee cup half because there are many, many, many models of the coffee cup.

[600] So the coffee, there is no model of a coffee cup.

[601] There are many models of a coffee cup.

[602] And you can say, well, how many different things can my finger learn?

[603] Is this the question you want to ask?

[604] Imagine I say every concept, every idea, everything you've ever know about that you can say, I know that thing has a reference frame associated with it.

[605] And what we do when we build composite objects, we assign reference frames to point another reference frame.

[606] So my coffee cup has multiple components to it.

[607] It's got a limb, it's got a cylinder, it's got a handle.

[608] And those things have their own reference frames and they're assigned to a master reference frame, which is called this cup.

[609] And now I have this nomento logo on it.

[610] Well, that's something that exists elsewhere in the world.

[611] It's its own thing.

[612] So it has its own reference frame.

[613] So we now have to say, how can I assign the nmenta logo reference frame onto the cylinder or onto the coffee cup?

[614] So it's all, we talked about this in the paper that came out in December this last year, the idea of how you can assign reference frames to reference frames, how neurons could do this.

[615] So my question is, even though you mentioned reference frames a lot, I almost feel it's really useful to dig into how you think of what a reference frame is.

[616] I mean, it was already helpful for me to understand that you think of reference frames as something there is a lot of.

[617] Okay, so let's just say that we're going to have some neurons in the brain, not many actually, 10 ,000, 20 ,000 are going to create a whole bunch of reference frames.

[618] What does it mean?

[619] What is the reference frame in this case?

[620] First of all, these reference names are different than the ones you might be used to.

[621] We know lots of reference names.

[622] We, for example, example, we know the Cartesian coordinates, XYZ, that's a type of reference frame.

[623] We know longitude and latitude, that's a different type of reference frame.

[624] If I look at a printed map, it might have columns A through M and rows, you know, 1 through 20, that's a different type of reference frame.

[625] It's kind of a Cartesian coordinate reference frame.

[626] The interesting thing about the reference frames in the brain, and we know this because these have been established through neuroscience studying the entorionic cortex, so I'm not speculating here.

[627] This is known neuroscience in an old part of the brain.

[628] The way these cells create reference frames, they have no origin.

[629] So what it's more like you have a point, a point in some space, and you given a particular movement, you can then tell what the next point should be.

[630] And you can then tell what the next point would be and so on.

[631] You can use this to calculate how to get from one point to another.

[632] So how do I get from my house?

[633] to my home, or how do I get my finger from the side of my cup to the top of the cup?

[634] How do I get from the axioms to the conjecture?

[635] So it's a different type of reference frame, and if you want, I can describe in more detail, I can paint a picture how you might want to think about that.

[636] It's really helpful to think.

[637] It's something you can move through.

[638] Yeah.

[639] But is there, is it helpful to think of it as spatial in some sense, or is there something?

[640] No, it's definitely spatial.

[641] It's spatial in a mathematical sense.

[642] How many dimensions?

[643] Can it be a crazy number of dimensions?

[644] Well, that's an interesting question.

[645] In the old part of the brain, the antirinochortex, they studied rats.

[646] And initially it looks like, oh, this is just two -dimensional.

[647] It's like the rat is in some box in a maze or whatever, and they know whether the rat is using these two -dimensional reference frames and know where it is in the maze.

[648] We say, well, okay, but what about what about bats?

[649] That's a mammal, and they fly in three -dimensional space.

[650] How do they do that?

[651] They seem to know where they are, right?

[652] So this is a current area of active research, and it seems like somehow the neurons in the interonacortex can learn three -dimensional space.

[653] We just, two members of our team, along with Ila Fett from MIT, just released a paper, this literally last week, it's on bio -archive, where they show that you can, if you, the way these things work, and I won't get, unless you want to, I won't get into the detail, but grid cells can represent.

[654] any n -dimensional space.

[655] It's not inherently limited.

[656] You can think of it this way.

[657] If you had two -dimensional, the way it works is you had a bunch of two -dimensional slices.

[658] That's the way these things work.

[659] There's a whole bunch of two -dimensional models.

[660] And you can just, you can slice up any n -dimensional space with two -dimensional projections.

[661] So, and you could have one -dimensional models.

[662] So there's, there's nothing inherent about the mathematics about the way the neurons do this, which, which constraining the dimensionality of the space, which I think was important.

[663] So obviously, I have a three -dimensional map of this cup.

[664] Maybe it's even more than that.

[665] I don't know.

[666] But it's clearly a three -dimensional map of the cup.

[667] I don't just have a projection of the cup.

[668] But when I think about birds or when I think about mathematics, perhaps it's more than three dimensions.

[669] Who knows?

[670] So in terms of each individual column building up more and more information over time, do you think that mechanism is well understood in your mind?

[671] You've proposed a lot of architectures there.

[672] Is that a key piece?

[673] or is it, is the big piece, the thousand brain theory of intelligence, the ensemble of it all?

[674] Well, I think they're both big.

[675] I mean, clearly, the concept, as a theorist, the concept is most exciting, right?

[676] We want a high level concept.

[677] This is a totally new way of thinking about how the near cortex works.

[678] So that is appealing.

[679] It has all these ramifications.

[680] And with that, as a framework for how the brain works, you can make all kinds of predictions and solve all kinds of problems.

[681] Now, we're trying to work through many of these details right now.

[682] Okay, how do the neurons actually do this.

[683] Well, it turns out, if you think about grid cells and play cells in the old parts of the brain, there's a lot that's known about them, but there's still some mysteries.

[684] There's a lot of debate about exactly the details, how these work, and what are the songs.

[685] And we have that still that same level of detail, the same level of concern.

[686] What we spend here most of our time doing is trying to make a very good list of the things we don't understand yet.

[687] That's the key part here.

[688] What are the constraints?

[689] It's not like, oh, this thing seems to work.

[690] We're done.

[691] No, it's like, okay, it kind of works, but These are other things we know it has to do, and it's not doing those yet.

[692] I would say we're well on the way here.

[693] We're not done yet.

[694] There's a lot of trickiness to this system, but the basic principles about how different layers in the near cortex are doing much of this, we understand.

[695] But there's some fundamental parts that we don't understand as well.

[696] So what would you say is one of the harder open problems or one of the ones that have been bothering you keeping up at night the most?

[697] Oh, well, right now, this is a detailed thing that wouldn't apply to most people, okay.

[698] But you want me to answer that question?

[699] Yeah, please.

[700] We've talked about as if, oh, to predict what you're going to sense on this coffee cup, I need to know where my finger is going to be on the coffee cup.

[701] That is true, but it's insufficient.

[702] Think about my finger touches the edge of the coffee cup.

[703] My finger can touch it at different orientations.

[704] I can rotate my finger around here.

[705] And that doesn't change.

[706] I can make that prediction and somehow, so it's not just the location.

[707] There's an orientation component of this as well.

[708] This is known in the old part of the brain too.

[709] There's things called head direction cells, which way the rat is facing.

[710] It's the same kind of basic idea.

[711] So if my finger were a rat, you know, in three dimensions, I have a three -dimensional orientation and I have a three -dimensional location.

[712] If I was a rat, I might think of it as a two -dimensional location, a two -dimensional orientation, a one -dimensional orientation, like just which way is it facing.

[713] So how the two components work together, how it is that I combine orientation, the orientation of my sensor, as well as the location, is a tricky problem.

[714] And I think I've made progress on it.

[715] So at a bigger version of that, so perspective is super interesting, but super specific.

[716] Yeah, I warned you.

[717] No, no, that's really good, but there's a more general version of that.

[718] Do you think context matters, the fact that we are in a building in North America, that we, in the day and age where we have mugs?

[719] I mean, there's all this extra information that you bring to the table about everything else in the room that's outside of just the coffee cup.

[720] How does it get connected, do you think?

[721] Yeah, and that is another really interesting question.

[722] I'm going to throw that under the rubric or the name of attentional problems.

[723] First of all, we have this model.

[724] I have many, many models.

[725] And also the question, does it matter?

[726] Well, it matters for certain things.

[727] Of course it does.

[728] Maybe what we think of that as a coffee cup in another part of the world is viewed as something completely different.

[729] Or maybe our logo, which is very benign in this part of the world, it means something very different than another part of the world.

[730] So those things do matter.

[731] I think the way to think about it is the following, one way to think about it, is we have all these models of the world, okay.

[732] And we model everything.

[733] And as I said earlier, I kind of snuck it in there.

[734] Our models are actually, we build composite structure.

[735] So every object is composed of other objects, which are composed of other objects, and they become members of other objects.

[736] So this room is chairs and a table and a room and walls and so on.

[737] Now we can just arrange these things in a certain way.

[738] You go, oh, that's something I meant the conference room.

[739] So, and what we do is when we go around the world and we experience the world, by walk into a room, for example, the first thing I say, oh, I'm in this room.

[740] Do I recognize the room?

[741] Then I can say, oh, look, there's a table here.

[742] And by attending to the table, I'm then assigning this table in a context of the room.

[743] Then I say, oh, on the table, there's a coffee cup.

[744] Oh, and on the table, there's a logo.

[745] And in the logo, there's the word nomenta.

[746] Oh, look, in the logo, there's the letter E. Oh, look, it has an unusual seraph.

[747] And it doesn't actually, but pretend it does.

[748] So the point is your attention is kind of, drilling deep in and out of these nested structures.

[749] And I can pop back up and I can pop back down.

[750] I can pop back up and I can pop back down.

[751] So when I attend to the coffee cup, I haven't lost the context of everything else, but it's sort of nested structure.

[752] So the attention filters the reference frame formation for that particular period of time?

[753] Yes, it basically, moment to moment you attend the subcomponents and then you can attend the subcomponents to subcomponents.

[754] And you can move up and down that.

[755] You can move up and down that.

[756] We do that all the time.

[757] not even, now that I'm aware of it, I'm very conscious of it.

[758] But most people don't even think about this.

[759] You know, you just walk in a room and you don't say, oh, I look at the chair and I looked to the board and I looked at that word on the board and I looked over here.

[760] What's going on?

[761] Right.

[762] So what percentage of your day?

[763] Are you deeply aware of this?

[764] And what part can you actually relax and just be, Jeff?

[765] Me personally, like my personal day.

[766] Yeah.

[767] Unfortunately, I'm afflicted with too much of the former.

[768] I think, unfortunately are unfortunate.

[769] Yeah.

[770] Yeah.

[771] You don't think it's useful?

[772] Oh, it is useful.

[773] Totally useful.

[774] I think about this stuff almost all the time.

[775] And one of my primary ways of thinking is when I'm in sleep at night, I always wake up in the middle of night.

[776] And then I stay awake for at least an hour with my eyes shut in sort of a half -sleep state thinking about these things.

[777] I come up with answers to problems very often in that sort of half -sleeping state.

[778] I think about it on my bike ride.

[779] I think about it walks.

[780] I'm just constantly thinking about this.

[781] I have to almost schedule time to not.

[782] not think about the stuff because it's very, it's mentally taxing.

[783] Are you, when you're thinking about the stuff, are you thinking introspectively, like almost taking a step outside of yourself and trying to figure out what is your mind doing right now?

[784] I do that all the time, but that's not all I do.

[785] I'm constantly observing myself.

[786] So as soon as I started thinking about grid cells, for example, and getting into that, I started saying, oh, well, grid cells have my place of sense in the world.

[787] You know, that's where you know where you are.

[788] And it's interesting, you know, we always have a sense of where we are, unless we're lost.

[789] And so I started at night when I got up to go to the bathroom, I would start trying to do it completely with my eyes closed all the time.

[790] And I would test my sense of grid cells.

[791] I would walk, you know, five feet and say, okay, I think I'm here.

[792] Am I really there?

[793] What's my error?

[794] And then I would calculate my error again and see how the errors can accumulate.

[795] So even something as simple as getting up in the middle of night to go to the bathroom, I'm testing these theories out.

[796] It's kind of fun.

[797] I mean, the coffee cup is an example of that too.

[798] So I think I find that these sort of everyday introspection are actually quite helpful.

[799] It doesn't mean you can ignore the science.

[800] I mean, I spend hours every day reading ridiculously complex papers.

[801] That's not nearly as much fun, but you have to sort of build up those constraints and the knowledge about the field and who's doing what and what exactly they think is happening here.

[802] And then you can sit back and say, okay, let's try to piece this all together.

[803] Let's come up with some, you know, I'm very, in this group, hearing people, they know they do this, I do this all those time.

[804] I come in with these introspective ideas and say, well, do you ever thought about this?

[805] watch well let's all do this together and it's helpful it's not if as long as you do you don't all you did was that then you're just making up stuff right but if you're constraining it by the reality of the neuroscience then it's really helpful so let's talk a little bit about deep learning and the successes in the applied space of neural networks the ideas of training model on data and these these simple computational units, neuron, artificial neurons that with bad propagation, the statistical ways of being able to generalize from the training set onto data that's similar to that training set.

[806] So where do you think are the limitations of those approaches, what do you think are strengths relative to your major efforts of constructing a theory of human intelligence?

[807] Yeah.

[808] Well, I'm not an expert in this field.

[809] I'm somewhat knowledgeable.

[810] Some of it is just your intuition.

[811] Well, I have a little bit more than intuition, but I just want to say, like, you know, one of the things that you ask me, do I spend all my time thinking about neuroscience I do?

[812] That's to the exclusion of thinking about things like convolutional neural networks.

[813] But I try to stay current.

[814] So, look, I think it's great the progress they've made.

[815] It's fantastic.

[816] And as I mentioned earlier, it's very highly useful for many things.

[817] The models that we have today are actually derived from a lot of neuroscience principles.

[818] There are distributed processing systems and distributed memory systems, and that's how the brain works.

[819] They use things that we might call them neurons, but they're really not neurons at all.

[820] So we can just, they're not really neurons.

[821] They're not really distributed processing systems.

[822] And the nature of hierarchy that came also from neuroscience.

[823] And so there's a lot of things, the learning rules, basically, not back prop, but other, you know, sort of heavy and type learning.

[824] I'll be curious to say they're not neurons at all.

[825] Can you describe in which way?

[826] I mean, some of it is obvious, but.

[827] I'd be curious if you have specific ways in which you think are the biggest differences.

[828] Yeah, we had a paper in 2016 called Why Neurons of Thousands of Synapses, and if you read that paper, you'll know what I'm talking about here.

[829] A real neuron in the brain is a complex thing.

[830] Let's just start with the synapses on it, which is a connection between neurons.

[831] Real neurons can everywhere from 5 to 30 ,000 synapses on them.

[832] The ones near the cell body, the ones are to close.

[833] to the soma, the cell body, those are like the ones who people model in artificial neurons.

[834] There is a few hundred of those, maybe they can affect the cell, they can make the cell become active.

[835] 95 % of the synapses can't do that.

[836] They're too far away.

[837] So if you activate one of those synapses, it just doesn't affect the cell body enough to make any difference.

[838] Any one of them individually?

[839] Any one of them individually?

[840] Or even if you do a mass of them.

[841] What real neurons do is the following.

[842] If you activate, or you know, you get 10 to 20 of them active at the same time, meaning they're all receiving an input at the same time.

[843] And those 10 to 20 synapses or 40 synapses within a very short distance on the dendron, like 40 microns, a very small area.

[844] So if you activate a bunch of these right next to each other at some distant place, what happens is it creates what's called the dendritic spike.

[845] And dendritic spike travels through the dendrites and can reach the soma or the cell body.

[846] Now, when it gets there, it changes the voltage, which is sort of like going to make the cell fire, but never enough to make the cell fire.

[847] It's sort of what we call it, says we depolarize the cell, you raise the voltage a little bit, but not enough to do anything.

[848] It's like, well, what good is that?

[849] And then it goes back down again.

[850] So we propose a theory, which I'm very confident in basics are, is that what's happening there is those 95 % of those synapses are recognizing dozens to hundreds of unique patterns.

[851] They can, you know, about 10, 20 synapses at a time.

[852] And they're acting like predictions.

[853] So the neuron actually is a predictive engine on its own.

[854] It can fire when it gets enough what they call approximately input from those ones near the cell fire, but it can get ready to fire from dozens to hundreds of patterns that are recognized this from the other guys.

[855] And the advantage of this to the neuron is that when it actually does produce a spike in action potential, it does so slightly sooner than it would have otherwise.

[856] And so what could is slightly sooner?

[857] Well, the slightly sooner part is it, there's all the neurons in the, the excitatory neurons in the brain are surrounded by these inhibitory neurons, and they're very fast, the inhibit neurons, these baskets all.

[858] And if I get my spike out a little bit sooner than someone else, I inhibit all my neighbors around me, right?

[859] And what you end up with is a different representation.

[860] You end up with a representation that matches your prediction.

[861] It's a sparser representation, meaning it's fewer neurons are active, but it's much more specific.

[862] And so we showed how networks of these neurons can do very sophisticated temporal prediction, basically.

[863] So, summarize this, real neurons in the brain are time -based prediction engines, and there's no concept of this at all in artificial, what would be called point neurons.

[864] I don't think you can mail the brain without them.

[865] I don't think you can build intelligence about it because it's where large, part of the time comes from.

[866] These are predictive models, and the time is, there's a prior and a, you know, a prediction and an action, and it's inherent through every neuron in the near cortex.

[867] So, so I would say that point neurons sort of model a piece of that, and not very well at that either, but, you know, like, for example, synapses are very unreliable, and you cannot assign any precision to them.

[868] So even one digit of precision is not possible.

[869] them.

[870] So the way real neurons work is they don't add these, they don't change these weights accurately like artificial neural networks do.

[871] They basically form new synapses.

[872] And so what you're trying to always do is detect the presence of some 10 to 20 active synapses at the same time as opposed, and they're almost binary.

[873] It's like, because you can't really represent anything much finer than that.

[874] So these are the kind of dish.

[875] And I think that's actually another essential component because the brain works on sparse patterns and all that all that mechanism is based on sparse patterns and i don't actually think you could build our real brains or machine intelligence without incorporating some of those ideas it's hard to even think about the complexity that emerges from the fact that the timing of the firing matters in the brain the fact that you form new uh new synapses and i mean everything you just mentioned in the past you can trust me if you spend time on it you can get your mind around it.

[876] It's not like, it's no longer a mystery to me. No, but sorry, as a function in a mathematical way, it's, can you get it, start getting an intuition about what gets it excited, what not?

[877] Yeah, it's not as easy as, there's many other types of neural networks that are more amenable to pure analysis, you know, especially very simple networks.

[878] You know, oh, I have four neurons and they're doing this.

[879] Can we, you know, describe them mathematically and what they're doing, type of thing.

[880] even the complexity of convolutional neural networks today it's sort of a mystery that can't really describe the whole system and so it's different my colleague Subitaine Ahmad he did a nice paper on this you can get all the stuff on our website if you're interested talking about sort of mathematical properties of sparse representations and so we can't what we can do is we can show mathematically for example why 10 to 20 synapses to recognize a pattern is the correct number, is the right number you'd want to use.

[881] And by the way, that matches biology.

[882] We can show mathematically some of these concepts about the show why the brain is so robust to noise and error and fallout and so on.

[883] We can show that mathematically as well as empirically in simulations.

[884] But the system can't be analyzed completely.

[885] Any complex system can't.

[886] So that's out of the realm.

[887] But there is mathematical benefits and intuitions that can be derived from mathematics.

[888] And we try to do that as well.

[889] Most of our papers have a section about that.

[890] So I think it's refreshing and useful for me to be talking to you about deep neural networks because your intuition basically says that we can't achieve anything like intelligence with artificial neural networks.

[891] Well, not in the current form.

[892] Not in the current form.

[893] Sure, we can do it in the ultimate form, sure.

[894] So let me dig into it and see what your thoughts are there a little bit.

[895] So I'm not sure if you read this little blog post called Bitter Lesson by Rich Sutton recently.

[896] He's a reinforcement learning pioneer.

[897] I'm not sure if you're familiar with him.

[898] His basic idea is that all the stuff we've done in AI in the past 70 years, he's one of the old school guys.

[899] The biggest lesson learned is that all the tricky things we've done don't, you know, they benefit.

[900] in the short term, but in the long term, what wins out is a simple general method that just relies on Moore's law, on computation getting faster and faster.

[901] This is what he's saying, this is what has worked up to now.

[902] This is what has worked up to now that if you're trying to build a system, if we're talking about, he's not concerned about intelligence, he's concerned about a system that works in terms of making predictions and applied narrow AI problems, right?

[903] that's what the discussion is about that you just try to go as general as possible and wait years or decades for the computation to make it actually is he saying that as a criticism or is he saying this is a prescription of what we ought to be doing well it's very difficult he's saying this is what has worked and yes a prescription but it's a difficult prescription because it says all the fun things you guys are trying to do we are trying to do he's part of the community is saying it's only going to be short -term gains.

[904] So this all leads up to a question, I guess, on artificial neural networks and maybe our own biological neural networks is, do you think if we just scale things up significantly?

[905] So take these dumb artificial neurons, the point neurons, I like that term, if we just have a lot more of them, do you think some of the elements that we see in the brain may start emerging.

[906] No, I don't think so.

[907] We can do bigger problems and of the same type.

[908] I mean, it's been pointing out by many people that today's convolutional networks aren't really much different than the ones we had quite a while ago.

[909] We just, they're bigger and trained more and we have more labeled data and so on.

[910] But I don't think you can get to the kind of things I know the brain can do and that we think about as intelligence by just scaling it up.

[911] So that may be, it's a good description of what's happened in the past, what's happened recently with the re -emergence of artificial neural networks.

[912] It may be a good prescription for what's going to happen in the short term, but I don't think that's the path.

[913] I've said that earlier.

[914] There's an alternate path.

[915] I should mention to you, by the way, that we've made sufficient progress on the whole cortical theory in the last few years, that last year we decided to start actively pursuing How do we get these ideas embedded into machine learning?

[916] That's, again, being led by my colleague, Supertai Maude.

[917] And he's more of a machine learning guy.

[918] I'm more of a neuroscience guy.

[919] So this is now our new, this is, I wouldn't say our focus, but it is now an equal focus here.

[920] Because we need to proselytize what we've learned, and we need to show how it's beneficial to the machine world.

[921] So we're putting, we have a plan in place right now.

[922] In fact, we just did our first paper on this.

[923] I can tell you about that.

[924] But one of the reasons I want to talk to you is because I'm trying to get more people in the machine learning community to say, I need to learn about this stuff.

[925] And maybe we should just think about this a bit more about what we've learned about the brain.

[926] And what are those team, they meant to have, what have they done?

[927] Is that useful for us?

[928] Yeah, so is there elements of all the cortical theory that things we've been talking about that may be useful in the short term?

[929] Yes, in the short term, yes.

[930] This is the, sorry to interrupt, but the open question is it certainly feels from my perspective that in the long term, some of the ideas we're talking about will be extremely useful.

[931] The question is whether in the short term.

[932] Well, this is always what I would call the entrepreneur's dilemma.

[933] So you have this long -term vision, oh, we're going to all be driving electric cars or we're all going to have computers or we're all going to, whatever.

[934] And you're at some point in time and you say, I can see that long -term vision.

[935] it's going to happen.

[936] How do I get there without killing myself, you know, without going out of business, right?

[937] That's the challenge.

[938] That's the dilemma.

[939] That's the really difficult thing to do.

[940] So we're facing that right now.

[941] So ideally what you'd want to do is find some steps along the way that you can get there incrementally.

[942] You don't have to like throw it all out and start over again.

[943] The first thing that we've done is we focus on the sparse representations.

[944] So just in case you don't know what that means or some of the listeners don't know what that in the brain, if I have like 10 ,000 neurons, what you would see is maybe 2 % of them active at a time.

[945] You don't see 50%, you don't think 30%, you might see 2%.

[946] And it's always like that.

[947] For any set of sensory inputs?

[948] It doesn't matter of anything.

[949] It doesn't matter any part of the brain.

[950] But which neurons differs?

[951] Which neurons are active?

[952] Yeah, so let's say I take 10 ,000 neurons that are representing something.

[953] They're sitting there in a block together.

[954] It's a teeny little block of neurons, 10 ,000 neurons.

[955] And they're representing a location, they're representing a cup, they're representing the input from my sensors, I don't know, it doesn't matter.

[956] It's representing something.

[957] The way the representations occur, it's always a sparse representation, meaning it's a population code.

[958] So which 200 cells are active tells me what's going on.

[959] It's not individual cells aren't that important at all.

[960] It's the population code that matters.

[961] And when you have sparse population codes, then all kinds of beautiful properties come out of them.

[962] So the brain uses sparse population codes that we've written and described.

[963] these benefits in some of our papers.

[964] So they give this tremendous robustness to the systems.

[965] You know, brains are incredibly robust.

[966] Neurons are dying all the time and spasoming and synapses falling apart and, you know, all the time and it keeps working.

[967] So what Subitai and Louise, one of our other engineers here have done, have shown they're introducing sparseness into convolution neural networks.

[968] Now other people are thinking along these lines, but we're going about it in a more principled way, I think.

[969] And we're showing that if you enforce sparseness throughout these convolutional neural networks, in both the, which neurons are active and the connections between them, that you get some very desirable properties.

[970] So one of the current hot topics in deep learning right now are these adversarial examples.

[971] So you give me any deep learning network, and I can give you a picture that looks perfect, and you're going to call it, you know, you're going to say the monkey is, you know, an airplane.

[972] So that's a problem.

[973] And DARPA just announced some big thing.

[974] We're trying to, you know, have some contest for this.

[975] But if you enforce sparse representations here, many of these problems go away.

[976] They're much more robust and they're not easy to fool.

[977] So we've already shown some of those results.

[978] Just literally in January or February, just like last month we did that.

[979] And you can, I think it's on bioarchive right now or on archive.

[980] You can read about it but um so that's like a baby step okay that's a take something from the brain we know we know about sparseness we know why it's important we know what it gives the brain so let's try to enforce that on to this what's your intuition why sparsity leads to robustness because it feels like it would be less robust so why why would you feel the rest robust to you so it uh it just feels like if the fewer neurons are involved the uh more fragile the representation but but I didn't say there was lots of few I said let's say 200.

[981] That's a lot.

[982] There's still a lot.

[983] Yeah.

[984] So here's an intuition for it.

[985] This is a bit technical.

[986] So for, you know, for engineers, machine learning people, this be easy, but all the listeners, maybe not.

[987] If you're trying to classify something, you're trying to divide some very high dimensional space into different pieces, A and B. And you're trying to create some point where you say, all these points in this high dimensional space are A and all these points inside dimensional space are B. And If you have points that are close to that line, it's not very robust.

[988] It works for all the points you know about, but it's not very robust, because you can just move a little bit and you've crossed over the line.

[989] When you have sparse representations, imagine I'm going to pick 200 cells active out of 10 ,000, okay?

[990] So I have 200 cells active.

[991] Now let's say I pick randomly a different representation, 200.

[992] The overlap between those is going to be very small, just a few.

[993] I can pick millions millions of samples randomly of 200 neurons and not one of them will overlap more than just a few.

[994] So one way to think about is if I want to fool one of these representations to look like one of those other representations, I can't move just one cell or two cells or three cells or four cells.

[995] I have to move a hundred cells.

[996] And that makes them robust.

[997] In terms of further, so you mentioned sparsity.

[998] Will we do the next thing?

[999] Yeah.

[1000] Okay.

[1001] So we're we have, we picked one.

[1002] We don't know if it's going to work well yet.

[1003] So again, we're trying to come up incremental ways to moving from brain theory to add pieces to machine learning, current machine learning world and one step at a time.

[1004] So the next thing we're going to try to do is sort of incorporate some of the ideas of the thousand brains theory, that you have many, many models that are voting.

[1005] Now, that idea is not new.

[1006] There's a mixture of models that's been around for a long time.

[1007] But the way the brain does is a little different.

[1008] And the way it votes is different.

[1009] And the kind of way it represents uncertainty is different.

[1010] So we're just starting this work, but we're going to try to see if we can sort of incorporate some of the principles of voting or principles of a thousand brain theory, like lots of simple models that talk to each other in a very certain way.

[1011] And can we build more of machines, the systems that learn faster and also, well, mostly are multimodal and robust to multimodal type of issues.

[1012] So one of the challenges there is, you know, the machine learning, computer vision community has certain sets of benchmarks, sets of tests based on which they compete.

[1013] And I would argue, especially from your perspective, that those benchmarks aren't that useful for testing the aspects that the brain is good at or intelligent.

[1014] They're not really testing intelligence.

[1015] It's a very fine, and it's been extremely useful for developing specific mathematical models, but it's not useful in the long term for creating intelligence.

[1016] So you think you also have a role in proposing better tests?

[1017] Yeah, this is a very, you've identified a very, serious problem.

[1018] First of all, the tests that they have are the tests that they want, not the test of the other things that we're trying to do, right?

[1019] You know, what are the, so on.

[1020] The second thing is sometimes to be competitive in these tests, you have to have huge data sets and huge computing power.

[1021] And so, you know, and we don't have that here.

[1022] We don't have it as well as other big teams that big companies do.

[1023] So, the...

[1024] There's numerous issues there.

[1025] You know, we come out, you know, our approach to this is all based on, in some sense, you might argue elegance.

[1026] We're coming out of from like a theoretical base that we think, oh my God, this is so clearly elegant.

[1027] This is how brains work.

[1028] This is one intelligence is.

[1029] But the machine learning world has gotten in this phase where they think of it doesn't matter.

[1030] It doesn't matter what you think.

[1031] As long as you do, you know, 0 .1 % better on this benchmark.

[1032] That's what that's all that matters.

[1033] And that's a problem.

[1034] You know, we have to figure out how to get around that.

[1035] That's a challenge for us.

[1036] That's one of the challenges that we have to do.

[1037] deal with.

[1038] So I agree, you've identified a big issue.

[1039] It's difficult for those reasons.

[1040] But, you know, part of the reasons I'm talking to you here today is I hope I'm going to get some machine learning people to say, I'm going to read those papers.

[1041] Those might be some interesting ideas.

[1042] I'm tired.

[1043] I'm tired of doing this 0 .1 % improvement stuff, you know.

[1044] Well, that's why I'm here as well because I think machine learning now as a community is at a place where the next step needs to be orthogonal to what, has received success in the past.

[1045] You see other leaders saying this, machine learning leaders, you know, Jeff Hinton with this capsule's idea, many people have gotten up to say, you know, we're going to hit road, maybe we should look at the brain, you know, things like that.

[1046] So hopefully that thinking will occur organically, and then we're in a nice position for people to come and look at our work and say, well, what can we learn from these guys?

[1047] Yeah, MIT's just launching a billion -dollar computing college that's centered around this idea.

[1048] on this idea of what of uh well the idea that uh you know the humanity psychology and neuroscience have to work all together to get to build the ask yeah i mean Stanford just did this human centered a i think i'm a little disappointed in these initiatives because yeah you know they're they're focusing as sort of the human side of it and it can very easily slip into how humans interact with um intelligent machines which is nothing wrong with that but that's not that is is orthogonal to what we're trying to do.

[1049] We're trying to say, like, what is the essence of intelligence?

[1050] I don't care.

[1051] In fact, I want to build intelligent machines that aren't emotional, that don't smile at you, that, you know, that aren't trying to tuck you in at night.

[1052] Yeah, there is that pattern that you talk about understanding humans is important for understanding intelligence, that you start slipping into topics of ethics or, yeah, like you said, the interactive elements are supposed to, no, no, no, we have to zoom in on the brain, and study what the human brain, the baby.

[1053] Let's study what a brain does.

[1054] It does.

[1055] And then we can decide which parts of that we want to recreate in some system.

[1056] But until you have that theory about what the brain does, what's the point?

[1057] You know, you're going to be wasting time, I think.

[1058] Just to break it down on the artificial neural network side, maybe you can speak to this on the biological neural network side, the process of learning versus the process of inference.

[1059] Maybe you can explain to me, is there a different?

[1060] between, you know, in artificial neural networks, there's a difference between the learning stage and the inference stage.

[1061] Do you see the brain is something different?

[1062] One of the, one of the big distinctions that people often say, I don't know how correct it is, is artificial neural networks need a lot of data.

[1063] They're very inefficient learning.

[1064] Do you see that as a correct distinction from the biology of the human brain, that the human brain is very efficient?

[1065] Or is that just something we deceive ourselves?

[1066] No, it is efficient.

[1067] Obviously, we can learn new things almost instantly.

[1068] And so what elements do you think?

[1069] Yeah, I can talk about that.

[1070] You brought up two issues there.

[1071] So remember I talked early about the constraints.

[1072] We always feel, well, one of those constraints is the fact that brains are continually learning.

[1073] That's not something we said, oh, we can add that later.

[1074] That's something that was up front, had to be there from the start, made our problems harder.

[1075] But we showed, going back to the 2016 paper on sequence memory, we showed how that happens.

[1076] How do brains infer and learn at the same time?

[1077] And our models do that.

[1078] They're not two separate phases or two separate sets of time.

[1079] I think that's a big, big problem in AI, at least for many applications, not for all.

[1080] So I can talk about that.

[1081] There are some, it gets detailed.

[1082] There are some parts of the neocortex in the brain where actually what's going on, there's these cycles, they're like cycles of activity in the brain.

[1083] And there's very strong evidence that you're doing more of inference on one part of the phase and more of learning on the other part of the phase.

[1084] So the brain can actually sort of step with different populations of cells are going back and forth like this.

[1085] But in general, I would say that's an important problem.

[1086] We have, you know, all of our networks that we've come up with do both.

[1087] And they're learning, continuous learning networks.

[1088] And you mentioned benchmarks earlier.

[1089] Well, there are no benchmarks about that.

[1090] Exactly.

[1091] So, so we, you know, we have to like, you know, we get our little.

[1092] soapbox and hey by the way we you know this is important you know and here's a mechanism for doing that but then you know but until you can prove it to someone in some you know commercial system or something it's a little harder so yeah one of the things I to linger on that is in some ways to learn the concept of a coffee cup you only need this one coffee cup and maybe some time alone in a room with it well the first thing is I when I imagine I reach my hand into a black box and I'm reaching I'm trying to touch something yeah I don't know up front if it's something I already know or if it's a new thing.

[1093] And I have to, I'm doing both at the same time.

[1094] I don't say, oh, let's see if it's a new thing.

[1095] Oh, let's see if it's an old thing.

[1096] I don't do that.

[1097] As I go, my brain says, oh, it's new or it's not new.

[1098] And if it's new, I start learning what it is.

[1099] And by the way, it starts learning from the get -go, even if it's going to recognize it.

[1100] So they're not separate problems.

[1101] And so that's the thing.

[1102] The other thing you mentioned was the fast learning.

[1103] So I was just talking about continuous learning, but there's also fast learning.

[1104] Literally, I can show you this coffee cup and I say, here's a new coffee cup.

[1105] It's got the logo on it.

[1106] Take a look at it.

[1107] Done.

[1108] You're done.

[1109] You can predict what it's going to look like.

[1110] You know in different positions.

[1111] So I can talk about that too.

[1112] In the brain, the way learning occurs, I mentioned this earlier, but I mentioned again, the way learning occurs.

[1113] Imagine I have a section of a dendrite of a neuron.

[1114] And I want to learn, I'm going to learn something new.

[1115] It doesn't matter what it is.

[1116] I'm just going to learn something new.

[1117] I need to recognize a new pattern.

[1118] So what I'm going to do is I'm going to form new synapses.

[1119] New synapses, we're going to rewire the brain onto that section of the dendrite.

[1120] Once I've done that, everything else that neuron has learned is not affected by it.

[1121] That's because it's isolated to that small section of the dendrites.

[1122] They're not all being added together like a point neuron.

[1123] So if I learn something new on this segment here, it doesn't change any of the learning that occur anywhere else in that neuron.

[1124] So I can add something without affecting previous learning.

[1125] And I can do it quickly.

[1126] Now, we can talk about the quickness, how it's done in real neurons.

[1127] You might say, well, doesn't it take time to form synapses?

[1128] Yes, it can take maybe an hour to form a new synapse.

[1129] We can form memories quicker than that.

[1130] And I can explain that happens too, if you want.

[1131] But it's getting a bit neuroscienty.

[1132] That's great.

[1133] But is there an understanding of these mechanisms at every level?

[1134] Yeah.

[1135] from the short -term memories and the forming new connections.

[1136] So there's idea of synaptogenesis, the growth of new synapses.

[1137] That's well described, as well understood.

[1138] And that's an essential part of learning.

[1139] That is learning.

[1140] That is learning.

[1141] Okay.

[1142] You know, back, you know, going back many, many years, people, you know, was, what just named, the psychologist proposed, Hebb, Donald Heb, he proposed that learning was the modification of the strength of connection between two neurons.

[1143] People interpreted that as the modification of the strength of a synapse.

[1144] He didn't say that.

[1145] He just said there's a modification between the effect of one neuron and another.

[1146] So synaptogenesis is totally consistent with Donald Heb said.

[1147] But anyway, there's these mechanisms that growth in new synaps, you can go online, you can watch a video of a synapse growing in real time.

[1148] It's literally, you can see this little thing going boop, right?

[1149] It's pretty impressive.

[1150] So those mechanisms are known.

[1151] Now, there's another thing that we've speculated and we've written about, which is consistent with no neuroscience, but it's less proven.

[1152] And this is the idea, how do I form a memory really, really quickly?

[1153] Like, instantaneous.

[1154] If it takes an hour to grow synapse, like, that's not instantaneous.

[1155] So there are types of synapses called silent synapses.

[1156] They look like a synapse, but they don't do anything.

[1157] They're just sitting there.

[1158] It's like if an action potential that comes in, it doesn't release any neurotransmitter.

[1159] Some parts of the brain have more.

[1160] these and others.

[1161] For example, the hippocampus has a lot of them, which is where we associate most short -term memory with.

[1162] So what we speculated, again, in that 2016 paper, we proposed that the way we form very quick memories, very short -term memories, or quick memories, is that we convert silent synapses into active synapses.

[1163] It's like saying a synapse has a zero -weight and a one -weight, but the long -term memory has to be formed by synaptogenesis.

[1164] So you can remember something really quickly by just flipping a bunch of these guys from silent to active.

[1165] It's not like it's not from 0 .1 to 0 .15.

[1166] It's like doesn't do anything to it releases transmitter.

[1167] And if I do that over a bunch of these, I've got a very quick short -term memory.

[1168] So I guess the lesson behind this is that most neural networks today are fully connected.

[1169] Every neuron connects every other neuron from layer to layer.

[1170] That's not correct in the brain.

[1171] We don't want that.

[1172] We actually don't want that.

[1173] It's bad.

[1174] You want a very sparse connectivity so that any neuron connects to some subset of the neurons in the other layer.

[1175] And it does so on a dendrite -by -dendrite segment basis.

[1176] So it's a very parcelated out type of thing.

[1177] And that then learning is not adjusting all these weights, but learning is just saying, okay, connect to these 10 cells here right now.

[1178] In that process, you know, with artificial neural networks, it's a very simple process of back propagation that adjust the weights.

[1179] It's the process of synaptiogenesis.

[1180] Synaptogenesis.

[1181] It's even easier.

[1182] It's even easier.

[1183] Bad propagation requires something we, it really can't happen in brains.

[1184] This back propagation of this aerosignal.

[1185] It really can't happen.

[1186] People are trying to make it happen in brains, but it doesn't happen in brain.

[1187] This is pure heavy in learning.

[1188] What synaptogenesis is pure heavy in learning.

[1189] It's basically saying there's a population of cells over here that are active right now.

[1190] And there's a population of cells over here active right now.

[1191] How do I form connections between those?

[1192] active cells.

[1193] And it's literally saying this guy became active, these 100 neurons here became active before this neuron became active.

[1194] So form connections to those ones.

[1195] That's it.

[1196] There's no propagation of error, nothing.

[1197] All the networks we do, all the models we have work on almost completely on Hebbian learning, but on dendritic segments and multiple synapses at the same time.

[1198] So now let's have turned the question that you already answered, and maybe you can answer it again.

[1199] If you look at the history of artificial intelligence, where do you think we stand?

[1200] How far are we from solving intelligence?

[1201] You said you were very optimistic.

[1202] Can you elaborate on that?

[1203] Yeah, you know, it's always the crazy question to ask.

[1204] Because, you know, no one can predict the future.

[1205] Absolutely.

[1206] So I'll tell you a story.

[1207] I used to run a different neuroscience institute called the Redburn Neuroscience Institute.

[1208] And we would hold these symposiums and we'd get like 35 scientists from around the world to come together.

[1209] And I used to ask them all the same question.

[1210] I would say, well, how long do you think it'll be before we understand how the New York cortex works?

[1211] And everyone went around and they had introduced a name and they have to answer that question.

[1212] So I got, the typical answer was 50 to 100 years.

[1213] Some people would say 500 years.

[1214] Some people said, never.

[1215] I said, why are you, why are you a neuroscientist?

[1216] It's never going to be.

[1217] It's a good pay.

[1218] It's interesting.

[1219] So, you know, but, It doesn't work like that.

[1220] As I mentioned earlier, these are step functions.

[1221] Things happen and then bingo that happen.

[1222] You can't predict that.

[1223] I feel I've already passed a step function.

[1224] So if I can do my job correctly over the next five years, then meaning I can proselytize these ideas, I can convince other people they're right, we can show that other people or machine learning people should pay attention to these ideas, then we're definitely in an under 20 -year time frame.

[1225] If I can do those things, if I'm not successful in that, and this is the last time anyone talks to me and no one reads our papers and, you know, I'm wrong or something like that, then I don't know.

[1226] But it's not 50 years.

[1227] It's, you know, it'll, it'll, you know, the same thing about electric cars.

[1228] How quickly are they going to populate the world?

[1229] It probably takes about a 20 year span.

[1230] It'll be something like that, but I think if I could do what I said, we're starting it.

[1231] And of course there could be other you said step functions.

[1232] It could be everybody gives up on your ideas for 20 years and then all of a sudden somebody picks it up again.

[1233] Wait, that guy was onto something.

[1234] Yeah.

[1235] So that would be a failure on my part, right?

[1236] You know?

[1237] Yeah.

[1238] Think about Charles Babbage.

[1239] You know, Charles Babbage.

[1240] He's a guy who invented the computer back in 18 -something.

[1241] Yeah.

[1242] And everyone forgot about it until, you know, 100 years later and say, hey, this guy figured this stuff out a long time ago, but he was ahead of his time.

[1243] I don't think, you know, like, as I said, I recognize this is part of any entrepreneur's challenge.

[1244] I use an entrepreneur broadly in this case.

[1245] I'm not meaning like I'm building a business trying to sell something.

[1246] I mean, like I'm trying to sell ideas.

[1247] And this is the challenge is to how you get people to pay attention to you.

[1248] How do you get them to give you positive or negative feedback?

[1249] How do you get to people act differently based on your ideas?

[1250] So, you know, we'll see how well we do on that.

[1251] you know that there's a lot of hype behind artificial intelligence currently do you uh as as you look to spread the ideas that are of uh in your cortical theory of the things you're working on do you think there's some possibility will hit an AI winter once again yeah it's certainly certainly a possibility no question about yeah well I guess do I worry about it um I haven't decided yet if that's good or bad for my mission that's true that's very true because uh It's almost like you need the winter to refresh the palate.

[1252] Yeah, it's like I want, here's what you want to have it is.

[1253] You want, like, to the extent that everyone is so thrilled about the current state of machine learning and AI and they don't imagine they need anything else, it makes my job harder.

[1254] Right.

[1255] If everything crashed completely and every student left the field and there was no money for anybody to do anything and it became an embarrassment to talk about machine intelligence and AI, that wouldn't be good for us either.

[1256] You want sort of the soft landing approach, right?

[1257] You want enough people, the senior people in AI and machine learning and say, you know, we need other approaches.

[1258] We really need other approaches.

[1259] Damn, we need other approaches.

[1260] Maybe we should look to the brain.

[1261] Okay, let's look to the brain.

[1262] Who's got some brain ideas?

[1263] Okay, let's start a little project on the side here, trying to do brain idea -related stuff.

[1264] That's the ideal outcome we would want.

[1265] So I don't want a total winter, and yet I don't want it to be sunny all the time either.

[1266] So what do you think it takes to build a system with?

[1267] human level intelligence where once demonstrated you would be very impressed.

[1268] So does it have to have a body?

[1269] Does it have to have the C word we used before, consciousness, as an entirety, in a holistic sense?

[1270] First of all, I don't think the goal is to create a machine that is human level intelligence.

[1271] I think it's a false goal.

[1272] Back to Turing, I think it was a false statement.

[1273] We want to understand what intelligence is, and then we can build intelligent machines of all different scales, all different capabilities.

[1274] You know, a dog is intelligent.

[1275] I don't need, you know, that'd be pretty good to have a dog, you know, but what about something that doesn't look like an animal at all?

[1276] In different spaces.

[1277] So my thinking about this is that we want to define what intelligence is, agree upon what makes an intelligence system.

[1278] We can then say, okay, we're now going to build systems that work on those principles or some subset of them, and we can apply them to all different types of problems.

[1279] And the kind, the idea, it's like computing.

[1280] We don't ask, if I take a little, you know, little one -chip computer, I don't say, well, that's not a computer because it's not as powerful is this, you know, a big server over here.

[1281] No, because we know that what the principles are computing are, and I can apply those principles to a small problem or into a big problem.

[1282] And same, intelligence needs to get there.

[1283] We have to say, these are the principles.

[1284] I can make a small one, a big one, I can make them distributed, I can put them on different sensors.

[1285] They don't have to be human like at all.

[1286] Now, you did bring up a very interesting question about embodied, Does it have to have a body?

[1287] It has to have some concept of movement.

[1288] It has to be able to move through these reference frames I talked about earlier.

[1289] Whether it's physically moving, like I need, if I'm going to have an AI that understands coffee cups, it's going to have to pick up the coffee cup and touch it and look at it with its eyes and hands or something equivalent to that.

[1290] If I have a mathematical AI, maybe it needs to move through mathematical spaces.

[1291] I could have a virtual AI that lives in the internet.

[1292] And its movements are traversing links and digging into files.

[1293] But it's got a location that it's traveling through some space.

[1294] You can't have an AI that just take some flash thing input.

[1295] We call a flash inference.

[1296] Here's a pattern.

[1297] No, it's movement.

[1298] Movement pattern.

[1299] Movement pattern.

[1300] Attention.

[1301] Digging, building structure.

[1302] Just figuring out the model of the world.

[1303] So some sort of embodiment, whether it's physical or not, has to be part of.

[1304] of it.

[1305] So self -awareness in the way to be able to answer, where am I?

[1306] You're bringing up self -well, that's a different topic, self -awareness.

[1307] No, no, the very narrow definition of self, meaning knowing a sense of self -enough to know where am I in the space where it's interesting.

[1308] Yeah, basically the system, the system needs to know its location, or each component of the system needs to know where it is in the world at that point in time.

[1309] So self -awareness and consciousness, do you think one, from the perspective of neuroscience, and your cortex, these are interesting topics, solvable topics.

[1310] Do you have any ideas of why the heck it is that we have a subjective experience at all?

[1311] Yeah, I have a lot of questions.

[1312] And is it useful or is it just a side effect of us?

[1313] It's interesting to think about.

[1314] I don't think it's useful as a means to figure out how to build intelligent machines.

[1315] It's something that systems do, and we can talk about what it is, that are like, well, if I build a system like this, then it would be self -aware.

[1316] And if I built it like this, it wouldn't be self -aware.

[1317] So that's a choice I can have.

[1318] It's not like, oh, my God, it's self -aware.

[1319] You know, I can't turn.

[1320] I heard an interview recently with this philosopher from Yale.

[1321] I can't remember his name.

[1322] I apologize for that.

[1323] But he was talking about, well, if these computers are self -aware, then it would be a crime to unplug them.

[1324] And I'm like, oh, come on.

[1325] You know, that's not, you know, I can plug myself every night.

[1326] I go to sleep.

[1327] Is that a crime?

[1328] You know, I plugged myself in again in the morning.

[1329] I'm there I am.

[1330] So, you know, people get kind of bent out of shape about this.

[1331] I have very different, very detailed understanding or opinions about what it means to be conscious and what it means to be self -aware.

[1332] I don't think it's that interesting a problem.

[1333] You've talked to Christopher Cox, you know, he thinks that's the only problem.

[1334] I didn't actually listen to your interview with him.

[1335] But I know him, and I know that's the thing he says about that.

[1336] He also thinks intelligence and consciousness and have disjoint.

[1337] So, I mean, it's not, you don't have to have one or the other.

[1338] So he is, I disagree with that.

[1339] I just totally disagree with that.

[1340] So where is your thoughts of consciousness?

[1341] Where does it emerge from?

[1342] Because it is.

[1343] So then we have to break it down to the two parts, okay?

[1344] Because consciousness isn't one thing.

[1345] That's part of the problem with that term is it means different things to different people.

[1346] And there's different components of it.

[1347] There is a concept of self -awareness.

[1348] Okay.

[1349] That can be very easily explained.

[1350] You have a model of your own body.

[1351] The neocortex models things in the world.

[1352] And it also models your own body.

[1353] and then it has a memory.

[1354] It can remember what you've done.

[1355] Okay, so it can remember what you did this morning.

[1356] Can remember what you had for breakfast and so on.

[1357] And so I can say to you, okay, Lex, were you conscious this morning when you, you know, I had your, you know, bagel?

[1358] And you'd say, yes, I was conscious.

[1359] Now, what if I could take your brain and revert all the synapses back to the state they were this morning?

[1360] And then I said to you, Lex, were you conscious when you ate the bagel?

[1361] And you said, no, I wasn't conscious.

[1362] I said, here's a video of eating the bagel.

[1363] saying, I wasn't there.

[1364] That's not possible because I must have been unconscious at that time.

[1365] So we can just make this one -to -one correlation between memory of your body's trajectory through the world over some period of time, a memory of it, and the ability to recall that memory is what you would call conscious.

[1366] I was conscious of that.

[1367] It's a self -awareness.

[1368] And any system that can recall, memorize what it's done recently, and bring that back and invoke it again, would say, yeah, I'm aware.

[1369] I remember what I did.

[1370] Yep.

[1371] All right.

[1372] I got it.

[1373] That's an easy one, although some people think that's a hard run.

[1374] The more challenging part of consciousness is this is one that sometimes used going by the word qualia, which is, you know, why does an object seem red?

[1375] Or what is pain?

[1376] And why does pain feel like something?

[1377] Why do I feel redness?

[1378] Or what do I feel painous?

[1379] And then I could say, well, why the sight seems different than hearing?

[1380] You know, that's the same problem.

[1381] It's really, these are all just neurons and so how is it that why does looking at you feel different than you know i'm hearing you it feels different but there's just neurons in my head they're all doing the same thing so that's an interesting question um the best treatise i've read about this is by guy named oregon or reagan he wrote a book called um why red doesn't sound like a bell um it's a little um it's not it's not a trade book easy read but it and and it's an interesting question um take something like color, color really doesn't exist in the world.

[1382] It's not a property of the world.

[1383] Property of the world that exists is light frequency, and that gets turned into, we have certain cells in the retina that respond to different frequencies, different than others.

[1384] And so when they enter the brain, you just have a bunch of axons that are firing at different rates.

[1385] And from that, we perceive color.

[1386] But there is no color in the brain.

[1387] I mean, there's no color coming in on those synapses.

[1388] It's just a correlation between some axons and some property of frequency.

[1389] And that isn't even color itself.

[1390] Frequency doesn't have a color.

[1391] It's just what it is.

[1392] So then the question is, well, why does it even appear to have a color at all?

[1393] Just as you're describing it, there seems to be a connection to those ideas of reference frames.

[1394] I mean, it just feels like consciousness having the subject, assigning the feeling of red to the actual color or to the wavelength is used to.

[1395] for intelligence.

[1396] Yeah, I think that's a good way of putting it.

[1397] It's useful as a predictive mechanism or useful as a generalization idea.

[1398] It's a way of grouping things together to say it's useful to have a model like this.

[1399] Think about the, there's a well -known syndrome that people who've lost a limb experience called phantom limbs.

[1400] And what they claim is they can have their arm as removed, but they feel their arm.

[1401] That not only feel it, they know it's there.

[1402] It's there.

[1403] I know it's there.

[1404] They'll swear to you that it's there.

[1405] And then they can feel pain in their arm.

[1406] And they'll feel their finger.

[1407] And if they move their non -existent arm behind your back, then they feel the pain behind their back.

[1408] So this whole idea that your arm exists is a model of your brain.

[1409] It may or may not really exist.

[1410] And just like, but it's useful to have a model of something that sort of correlates to things in the world so you can make predictions about what would happen when those things occur.

[1411] It's a little bit of a fuzzy, but I think you're getting right towards the answer there.

[1412] It's useful for the model to express things certain ways that we can then map them into these reference frames and make predictions about them.

[1413] I need to spend more time on this topic.

[1414] It doesn't bother me. Do you really need to spend more time?

[1415] Yeah.

[1416] It does feel special that we have subjective experience, but I'm yet to know why.

[1417] I'm just personally curious.

[1418] It's not necessary for the work we're doing here.

[1419] I don't think I need to solve that problem to build intelligent machines at all, not at all.

[1420] But there is sort of the silly notion that you described briefly that doesn't seem so silly to us humans is, you know, if you're successful building intelligent machines, it feels wrong to then turn them off.

[1421] Because if you're able to build a lot of them, it feels wrong to then be able to, you know, to turn off the...

[1422] Well, why?

[1423] But just, let's break that down a bit.

[1424] As humans, why do we fear death?

[1425] There's two reasons we fear death.

[1426] Well, first of all, I'll say when you're dead, it doesn't matter.

[1427] Oh, who cares?

[1428] You're dead.

[1429] So why do we fear death?

[1430] We fear death for two reasons.

[1431] One is because we are programmed genetically to fear death.

[1432] That's a survival and propaganda of the genes thing.

[1433] And we also are program to feel sad when people we know die.

[1434] We don't feel sad for someone we don't know dies.

[1435] People dying right now, they're only going to say it.

[1436] I don't feel bad about them because I don't know them.

[1437] But by I knew them, I'd feel really bad.

[1438] So again, these are old brain genetically embedded things that we fear death.

[1439] Outside of those uncomfortable feelings, there's nothing else to worry about it.

[1440] Well, wait, hold on a second.

[1441] Do you know the denial of death by Becker, Denna?

[1442] You know, there's a thought that death is.

[1443] you know our whole conception of our world model kind of assumes immortality and then death is this terror that underlies it all so like well some people's world model not mine but okay so what what becker would say is that you're just living in an illusion you've constructed illusion for yourself because it's such a terrible terror the fact that what's the illusion the illusion that that doesn't matter you're still not coming to grips with the illusion of what that death is uh going to happen oh like it's not going to happen you're a met you're actually operating you haven't even though you said you've accepted it you haven't really accepted the notion of what you say so it sounds like it sounds like you disagree with that notion i mean yeah yeah totally i like i literally every night every night i go to bed it's like dying with little deaths it's it's it's a little death and if i didn't wake up it wouldn't matter to me only if i knew that was going to happen wouldn't be bothersome.

[1444] If I didn't know, how would I know?

[1445] Then I would worry about my wife.

[1446] So imagine, imagine I was a loner and I lived in Alaska and I lived out there and there's no animals.

[1447] Nobody knew I existed.

[1448] I was just eating these roots all the time.

[1449] And nobody knew I was there.

[1450] And one day I didn't wake up.

[1451] What pain in the world would there exist?

[1452] Well, so most people that think about this problem would say that you're just deeply enlightened or are completely delusional.

[1453] one of the two, but I would say, I would say that's a very enlightened way to see the world is, that, that's the rational one as well.

[1454] Well, I think it's rational, that's right.

[1455] But the fact is we don't, I mean, we really don't have an understanding of why the heck it is we're born and why we die and what happens after we die.

[1456] Well, maybe there isn't a reason, maybe there is.

[1457] So I'm interested in those big problems too, right?

[1458] You know, you interviewed Max Tagmark, you know, and there's people like that, right?

[1459] I mean, some of those big problems as well.

[1460] And in fact, when I was young, I made a list of the biggest problems I could think of.

[1461] First, why does anything exist?

[1462] Second, why did we have the laws of physics that we have?

[1463] Third, is life inevitable and why is it here?

[1464] Fourth, is intelligence inevitable and why is it here?

[1465] I stopped there because I figured if you can make a truly intelligent system, that would be the quickest way to answer the first three questions.

[1466] I'm serious.

[1467] Yeah.

[1468] And so I said, my mission, you know, you asked me earlier, my first mission is to understand the brain, but I felt that is the shortest way to get to true machine intelligence.

[1469] And I want to get to true machine intelligence because even if it doesn't occur in my lifetime, other people will benefit from it because I think it will occur in my lifetime, but, you know, 20 years, you never know.

[1470] And, but that will be the quickest way for us to, you know, we can make super mathematicians, we can make super space explorers.

[1471] We can make super physicists brains that do.

[1472] these things and that can run experiments that we can't run.

[1473] We don't have the abilities to manipulate things and so on.

[1474] But we can build and tell the machines that do all those things with the ultimate goal of finding out the answers to the other questions.

[1475] Let me ask you another depressing and difficult question, which is once we achieve that goal, do you, of creating it, no, of understanding intelligence.

[1476] Do you think we would be happier, more fulfilled as a species.

[1477] Understanding intelligence or understanding the answers to the big questions?

[1478] Understanding intelligence.

[1479] Oh, totally.

[1480] Totally.

[1481] It would be a far more fun place to live.

[1482] You think so?

[1483] Oh, yeah.

[1484] Why not?

[1485] I mean, you know, just put aside this, you know, Terminator nonsense and just think about, you can think about the, we can talk about the risk of AI if you want.

[1486] I'd love to, so let's talk about it.

[1487] But I think the world is far better knowing things.

[1488] We're always better than know things.

[1489] Do you think it's better, is it a better place to live in that I know that our planet is one of many in the solar system and the solar system is one of many in the galaxy?

[1490] I think it's more, I dread, I sometimes think like, God, what would be like to live 300 years ago?

[1491] I'd be looking up the sky.

[1492] I can't understand anything.

[1493] Oh my God.

[1494] I'd be like going to bed of me like going, what's going on here?

[1495] Well, I mean, in some sense I agree with you, but I'm not exactly sure.

[1496] So I'm also a scientist, so I have, I share your views, but I'm not, we're like rolling down the hill together.

[1497] of what's down the hill i feel if we're climbing a hill whatever we're getting closer to enlightenment and you're going down we're climbing we're getting pulled up a hill by our curiosity we are putting our poliosity is pulling ourselves up the hill by our curiosity yeah cispos we're doing the same thing with the rock yeah yeah uh but okay our happiness aside do you have concerns about you know you talk about sam harris Elon Musk of existential threats of intelligence systems.

[1498] No, I'm not worried about existential threats at all.

[1499] There are some things we really do need to worry about.

[1500] Even today's AI, we have things we have to worry about.

[1501] We have to worry about privacy and about how impacts false beliefs in the world.

[1502] And we have real problems and things to worry about with today's AI.

[1503] And that will continue as we create more intelligent systems.

[1504] There's no question, you know, the whole issue about, you know, making intelligent armament and weapons.

[1505] something that really we have to think about carefully.

[1506] I don't think of those as existential threats.

[1507] I think those are the kind of threats we always face and we'll have to face them here and we'll have to deal with them.

[1508] We could talk about what people think are the existential threats, but when I hear people talking about them, they all sound hollow to me. They're based on ideas.

[1509] They're based on people who really have no idea what intelligence is.

[1510] And if they knew what intelligence was they wouldn't say those things so those are not experts in the field you know so yeah so there's two right there's so one is like super intelligence so a system that becomes far far superior in reasoning ability uh than us humans how is that an existential threat uh then so there's a lot of ways in which it could be one way is us humans are actually irrational inefficient and get in the way of not happiness, but whatever the objective function is, of maximizing that objective function.

[1511] It's super intelligent.

[1512] The paper clip problem and things like that.

[1513] So the paper clip problem, but with a super intelligent.

[1514] Yeah, yeah, yeah, yeah.

[1515] So we already face this threat in some sense.

[1516] They're called bacteria.

[1517] These are organisms in the world that would like to turn everything into bacteria.

[1518] And they're constantly morphing, they're constantly changing, to evade our protections.

[1519] And in the past, they have killed huge swaths of populations of humans on this planet.

[1520] So if you want to worry about something that's gonna multiply endlessly, we have it.

[1521] And I'm far more worried in that regard, I'm far more worried that some scientists in a laboratory will create a super virus or a super bacteria that we cannot control.

[1522] That is a more of existential strep.

[1523] Putting an intelligence thing on top of it actually seems to make it less existential to me. me. It's like it limits its power.

[1524] It's limits where it can go.

[1525] It limits the number of things it can do in many ways.

[1526] A bacteria is something you can't, you can't even see.

[1527] So that's only one of those problems.

[1528] Yes, exactly.

[1529] So the other one, just in your intuition about intelligence, when you think about intelligence of us humans, do you think of that as something, if you look at intelligence on a spectrum from zero to us humans, do you think you can scale that to something far superior, all the mechanisms we've been talking about.

[1530] Let me, I want to make another point here, Lex, before I get there.

[1531] Intelligence is the near cortex.

[1532] It is not the entire brain.

[1533] If I, the goal is not to be, make a human.

[1534] The goal is not to make an emotional system.

[1535] The goal is not to make a system that wants to have sex and reproduce.

[1536] Why would I build that?

[1537] If I want to have a system that wants to reproduce and have sex, make bacteria, make computer viruses.

[1538] Those are bad things.

[1539] Don't do that.

[1540] Those are really bad.

[1541] Don't do those things.

[1542] regulate those.

[1543] But if I just say I want an intelligent system, why does it have to have any of the human -like emotions?

[1544] Why could it, why does it even care if it lives?

[1545] Why does it even care if it has food?

[1546] It doesn't care about those things.

[1547] It's just, you know, it's just in a trance thinking about mathematics or it's out there just trying to build the space plant, you know, for it on Mars.

[1548] It's a, we don't, that's a choice we make.

[1549] Don't make human -like things.

[1550] Don't make replicating things.

[1551] Don't make things that have emotions.

[1552] Just stick to the near cortex.

[1553] That's a view, actually, that I share, but not everybody shares, in the sense that you have faith and optimism about us as engineers as systems, humans as builders of systems to not put in stupid things.

[1554] So this is why I mentioned the bacteria one, because you might say, well, some person's going to do that.

[1555] Well, some person today could create a bacteria that's resistant to all the non -antibacterial agents.

[1556] So we already have that threat.

[1557] We already know this is going on.

[1558] It's not a new threat.

[1559] So just accept that and then we have to deal with it, right?

[1560] My point has nothing to do with intelligence.

[1561] Intelligence is a separate component that you might apply to a system that wants to reproduce and do stupid things.

[1562] Let's not do that.

[1563] Yeah.

[1564] In fact, it is a mystery why people haven't done that yet.

[1565] My dad as a physicist believes that the reason, for example, nuclear weapons haven't proliferated amongst evil people.

[1566] So one belief that I share is that there's not that many evil people in the world that would use, whether there's bacteria or nuclear weapons or maybe the future AI systems to do bad.

[1567] So the fraction is small.

[1568] And the second is that it's actually really hard technically.

[1569] So the intersection between evil and competent is small in terms of...

[1570] And by the way, to really, annihilate humanity, you'd have to have, you know, sort of the nuclear winter phenomenon, which is not one person shooting, you know, or even 10 bombs.

[1571] You'd have to have some automated system that, you know, detonates a million bombs or whatever many thousands we have.

[1572] So extreme evil combined with extreme competence.

[1573] And it's just by with building some stupid system that would automatically, you know, Dr. Strangelove type of thing, you know, detonate it all.

[1574] I mean, look, we could have some nuclear bomb go off in some major city in the world.

[1575] I think that's actually quite likely even in my lifetime i don't think that's unlike the thing and it'll be a tragedy um but it won't be an existential threat um and it's the same as you know the virus of 1917 whatever it was you know the influenza um these bad things can happen and the plague and so on it we can't always prevent we always to always try but we can't but they're not existential threats until we combine all those crazy things together in one so on the on the spectrum of intelligence from zero to human Do you have a sense of whether it's possible to create several orders of magnitude or at least double that of human intelligence?

[1576] I think it's the wrong thing to say double the intelligence.

[1577] Break it down into different components.

[1578] Can I make something that's a million times fast in a human brain?

[1579] Yes, I can do that.

[1580] Could I make something that has a lot more storage than the human brain?

[1581] Yes, I can do that.

[1582] More copies of comp.

[1583] Can I make something that attaches a different sense?

[1584] than human brain?

[1585] Yes, I can do that.

[1586] Could I make something that's distributed?

[1587] So these people, yeah, we talked earlier about the importance of your cortex voting.

[1588] They don't have to be co -located.

[1589] Like, you know, they can be all around the places.

[1590] I could do that too.

[1591] Those are the levers I have, but is it more intelligent?

[1592] What depends what I train in on?

[1593] What is it doing?

[1594] Well, so here's the thing.

[1595] So let's say larger in your cortex and or whatever size that allows for higher and higher, higher, hierarchies to form, we're talking about reference frames and context.

[1596] So could, could I have something that's a super physicist or a super mathematician?

[1597] Yes.

[1598] And the question is, once you have a super physicist, will they be able to understand something?

[1599] Do you have a sense that it will be orders to match, like us compared to ants?

[1600] Could we ever understand it?

[1601] Yeah.

[1602] Most people cannot understand general relativity.

[1603] It's a really hard thing to get.

[1604] I mean, you know, you can paint it in a fuzzy picture.

[1605] stretchy space you know yeah but the the field equations to do that and the deep intuitions are really really hard and um i've tried i'm not able to do it i it's easy to get you know it's easier to get special relative but general it's man that's too much um and so we already live with this to some extent the vast majority people can't understand actually what the vast majority of other people actually know uh we're just either we don't have the effort to or we can't or we don't time or just not smart enough, whatever.

[1606] So, but we have ways of communicating.

[1607] Einstein has spoken in a way that I can understand.

[1608] He's given me analogies that are useful.

[1609] I can use those analogies for my own work and think about, you know, concepts that are similar.

[1610] It's not stupid.

[1611] It's not like he's existed some other plane and there's no connection to with my plane in the world here.

[1612] So that will occur.

[1613] It already has occurred.

[1614] That's my point at this story is it already has occurred.

[1615] We live it every day.

[1616] One could argue that when we create machine intelligence that think a million times faster than us that it'll be so far, we can't make the connections.

[1617] But, you know, at the moment, everything that seems really, really hard to figure out in the world when you actually figure it out, it's not that hard.

[1618] You know, almost everyone can understand the multiverses.

[1619] Almost everyone can understand quantum physics.

[1620] Almost everyone can understand these basic things, even though hardly any people could figure those things out.

[1621] Yeah, but really understand.

[1622] But you don't need to really, only a few people really don't understand.

[1623] You need to only understand the projections, the sprinkles of the useful insights from that.

[1624] That was my example of Einstein, right?

[1625] His general theory of relativity is one thing that very, very, very few people can get.

[1626] And what if we just said those other few people are also artificial intelligences?

[1627] How bad is that?

[1628] In some sense, they are, right?

[1629] Yeah, right.

[1630] So they say already, I mean, Einstein wasn't a really normal person.

[1631] He had a lot of weirdly quirks.

[1632] And so the other people who work with them.

[1633] So, you know, maybe they already were sort of this astral plane of intelligence that we live with it already.

[1634] It's not a problem.

[1635] It's still useful and, you know.

[1636] So do you think we are the only intelligent life out there in the universe?

[1637] I would say that intelligent life has and will exist elsewhere in the universe.

[1638] I'll say that.

[1639] There is a question about contemporaneous intelligence life, which is hard to even answer when we think about relativity in the nature of space time.

[1640] You can't say what exactly is this time someplace else in the world.

[1641] But I think it's, you know, I do worry a lot about the filter idea, which is that perhaps intelligent species don't last very long.

[1642] And so we haven't been around very long.

[1643] And as a technological species, we've been around for almost nothing, you know, what, 200 years, something like that.

[1644] And we don't have any data, a good data point on whether it's likely that we'll survive or not.

[1645] So do I think that there have been intelligent life elsewhere in the universe?

[1646] Almost certain, of course.

[1647] In the past, in the future, yes.

[1648] Does it survive for a long time?

[1649] I don't know.

[1650] This is another reason I'm excited about our work, is our work meaning the general world of AI.

[1651] I think we can build intelligent machines that outlast us.

[1652] They don't have to be tied to earth.

[1653] They don't have to, you know, I'm not saying they're recreating.

[1654] you know, aliens, I'm just saying, if I ask myself, and this might be a good point to end on here, if I ask myself, you know, what's special about our species?

[1655] We're not particularly interesting physically.

[1656] We're not, we don't fly, we're not good swimmers, we're not very fast, we're not very strong, you know.

[1657] It's our brain.

[1658] That's the only thing.

[1659] And we are the only species on this planet that's built the model of the world that extends beyond what we can actually sense.

[1660] We're the only people who know about the far side of the moon and the other universes, I mean, other galaxies and other stars and what happens in the atom.

[1661] That knowledge doesn't exist anywhere else.

[1662] It's only in our heads.

[1663] Cats don't do it.

[1664] Dogs and do, monkeys don't do it.

[1665] It's this on.

[1666] And that is what we've created.

[1667] That's unique.

[1668] Not our genes.

[1669] It's knowledge.

[1670] And if I ask me, what is the legacy of humanity?

[1671] What should our legacy be?

[1672] It should be knowledge.

[1673] We should preserve our knowledge in a way that it can exist beyond us.

[1674] And I think the best way of doing that, in fact, you have to do it, is it has to go along with intelligent machines to understand that knowledge.

[1675] It's a very broad idea, but we should be thinking, I call it a state planning for humanity.

[1676] We should be thinking about what we want to leave behind when as a species we're no longer here.

[1677] And that will happen sometime.

[1678] Sooner or later, it's going to happen.

[1679] And understanding intelligence and creating intelligence gives us a better chance to prolong...

[1680] It does give us a better chance to prolong life.

[1681] Yes.

[1682] It gives us a chance to live on other planets.

[1683] but even beyond that, I mean, our solar system will disappear one day.

[1684] Just give it enough time.

[1685] So I don't know.

[1686] I doubt we'll ever be able to travel to other things.

[1687] But we could tell the stars, but we could send intelligent machines to do that.

[1688] So you have an optimistic, a hopeful view of our knowledge of the echoes of human civilization living through the intelligence systems we create?

[1689] Oh, totally.

[1690] Well, I think the intelligence systems are greater in some sense.

[1691] the vessel for bringing him beyond earth, or making them last beyond humans themselves.

[1692] How do you feel about that?

[1693] That they won't be human, quote -unquote.

[1694] Who can't?

[1695] It's not, but human, what is human?

[1696] Our species are changing all the time.

[1697] Human today is not the same as human just 50 years ago.

[1698] It's, what is human?

[1699] Do we care about our genetics?

[1700] Why is that important?

[1701] As I point out, our genetics are no more interesting than a bacterium's genetics.

[1702] It's no more interesting than, you know, monkeys' genetics.

[1703] What we have, what's unique and what's valuable is our knowledge, what we've learned about the world.

[1704] And that is the rare thing.

[1705] That's the thing we want to preserve.

[1706] Who cares about our genes?

[1707] It's the knowledge.

[1708] It's the knowledge.

[1709] That's a really good place to end.

[1710] Thank you so much for talking to me. It was fun.