WEBVTT

00:00:03.517 --> 00:00:10.497
This is the Convergent Science Network podcast. Leading researchers in the domain

00:00:10.497 --> 00:00:16.717
of neuroscience, brain theory and technology are interviewed by Paul Verschure and Tony Prescott.

00:00:20.997 --> 00:00:25.957
Okay, this is Paul Verschure for the Convergent Science Network podcast.

00:00:25.957 --> 00:00:31.217
Podcast together with colleague tony prescott and we welcome joshua who is a

00:00:31.217 --> 00:00:36.897
speaker at the bcd summer school in 2018 where you have two lectures actually

00:00:36.897 --> 00:00:38.577
and in some sense they're all.

00:00:39.777 --> 00:00:45.437
Turn around i think the general question of how we can build an artificial general

00:00:45.437 --> 00:00:48.997
intelligence i feel this is really i think that the core concept that you're

00:00:48.997 --> 00:00:51.997
or the challenge that you're that you're dealing with, right?

00:00:52.057 --> 00:00:57.017
So what's an artificial general intelligence in your mind?

00:00:58.137 --> 00:01:02.477
It's a system that has a mind similar to our own and beyond.

00:01:02.657 --> 00:01:06.057
I suspect that our minds have a particular kind of generality.

00:01:06.197 --> 00:01:10.937
And of course, you cannot really know this, whether we are universal functional approximators.

00:01:11.157 --> 00:01:16.257
But if it turns out that we can build AI similar to us, that's a strong indication

00:01:16.257 --> 00:01:18.297
that we are in that class of systems, right?

00:01:18.297 --> 00:01:22.617
Right. So the whole enterprise is spawned by the question of who we are,

00:01:22.717 --> 00:01:26.857
what is our relationship to reality, what's our nature, what is consciousness,

00:01:27.037 --> 00:01:30.637
why are things happening to us, what does it mean that something is happening to us.

00:01:30.757 --> 00:01:36.457
And we realize that psychology has become somewhat unproductive and philosophy

00:01:36.457 --> 00:01:40.337
has become even more unproductive as a field because you need to perform experiments

00:01:40.337 --> 00:01:44.857
that you cannot actually perform in psychology and that you never performed in philosophy.

00:01:44.857 --> 00:01:51.337
Philosophy so we now can build computer models and implement our ideas and usually

00:01:51.337 --> 00:01:55.477
our ideas will not work which will force us to update our theories and it's

00:01:55.477 --> 00:01:56.357
the way we can make progress.

00:01:57.554 --> 00:02:04.314
So, okay, so this is a bit the outlook, but to get there, you saw different

00:02:04.314 --> 00:02:08.274
stages in which then this search for AI has evolved, right?

00:02:08.274 --> 00:02:12.234
And so we go from our symbolic AI,

00:02:12.554 --> 00:02:19.954
this is sort of version one or first order, to an AI that's more like the current

00:02:19.954 --> 00:02:25.674
forms of AI as we know, where we use different machine learning methods to actually

00:02:25.674 --> 00:02:27.394
that can learn functions, right?

00:02:27.454 --> 00:02:31.714
This was the whole point as far as humans writing them out, we can try to learn them.

00:02:32.474 --> 00:02:39.034
And then we would jump into meta learning and meta search. And so what's the logic of these steps?

00:02:39.154 --> 00:02:43.874
What's sort of the driving principles that you would see sort of meta-search

00:02:43.874 --> 00:02:45.314
as a tip of that pinnacle?

00:02:45.554 --> 00:02:48.694
There are many ways of looking at this whole issue. I mean, there's a big rift

00:02:48.694 --> 00:02:52.674
in AI traditionally between symbolic AI and everybody else's AI.

00:02:52.974 --> 00:02:58.034
And before I came to Cambridge, I thought this was because of everybody else

00:02:58.034 --> 00:02:59.354
not listening to Marvin Minsky.

00:03:00.154 --> 00:03:05.894
And after I got to Cambridge and met Marvin Minsky, I felt it was basically the opposite.

00:03:06.694 --> 00:03:11.914
That Marvin Minsky, who co-founded the field with John McCarthy and a few others,

00:03:12.094 --> 00:03:16.294
realized that in order to understand cognition, we need to teach computers how to think.

00:03:16.994 --> 00:03:21.734
So we need to build cognitive AI in a way. And then he made the decision that

00:03:21.734 --> 00:03:22.514
this needs to be symbolic.

00:03:23.688 --> 00:03:26.528
And this was the wrong horse, but Minsky was too influential.

00:03:26.588 --> 00:03:30.208
He started yelling at people that did low-level cybernetics and neural learning

00:03:30.208 --> 00:03:32.388
because he felt what they were doing was too simplistic.

00:03:32.708 --> 00:03:36.008
And in a way, he was correct. What they were doing was very simplistic.

00:03:36.148 --> 00:03:40.668
But it turns out the way we can get to simple grounding and to the kind of operation

00:03:40.668 --> 00:03:43.568
that our minds are doing, they start from there.

00:03:43.648 --> 00:03:48.108
They don't start from symbolic representations, which are so highly abstract

00:03:48.108 --> 00:03:53.368
that we can only form them after generalizing over our perceptual representations and operations.

00:03:53.688 --> 00:03:57.628
And so on, right? These symbolic operations, we get them by making our representations

00:03:57.628 --> 00:04:03.128
so discrete and low-dimensional that we can apply analytic operators on them.

00:04:03.248 --> 00:04:06.628
It doesn't work the other way. You cannot start with these analytic operators,

00:04:06.728 --> 00:04:08.828
or at least this project has not worked out.

00:04:08.988 --> 00:04:12.908
But what happened is basically this big rift because Minsky basically started

00:04:12.908 --> 00:04:16.728
yelling at people for doing other things than he did, and this created an area

00:04:16.728 --> 00:04:18.508
of burned ground around cognitive AI,

00:04:18.728 --> 00:04:23.328
which was also symbolic AI, and the AI of other people, which never bothered

00:04:23.328 --> 00:04:27.248
to read Piaget anymore and instead mostly did neural learning and so on.

00:04:27.408 --> 00:04:30.668
But what happened in the first half of AI,

00:04:30.848 --> 00:04:36.828
basically since 1950 until quite recently, was mostly that people thought about,

00:04:36.928 --> 00:04:40.948
well, when we want to perform the following tasks that requires human intelligence,

00:04:40.968 --> 00:04:43.848
like playing chess or parsing natural language.

00:04:44.908 --> 00:04:47.088
What do we need to do to make that happen?

00:04:47.188 --> 00:04:50.368
And they came up with a clever algorithm and then they implemented this algorithm

00:04:50.368 --> 00:04:54.308
at some point because our computers are very deterministic and scale very well,

00:04:54.428 --> 00:04:58.288
these algorithms became smarter than people. So I'm born in 1973.

00:04:58.888 --> 00:05:03.548
There was never a point in my life where I had a chance to play better chess than a computer.

00:05:04.594 --> 00:05:08.074
In a way, depressing, right? Even though these algorithms were so simplistic

00:05:08.074 --> 00:05:12.154
that I was able to write an algorithm that would play better chess than myself, right?

00:05:12.354 --> 00:05:18.254
Even though I'm not that good of a programmer compared to the people that put all this effort in this.

00:05:18.394 --> 00:05:22.474
But we know that our strategy of playing chess is very different than the ones

00:05:22.474 --> 00:05:23.534
of this simplistic algorithm.

00:05:24.054 --> 00:05:28.574
We are able to cope with much more uncertainty in our search space than our

00:05:28.574 --> 00:05:33.014
programs, so we can have much more fuzzy representations that are more perceptual

00:05:33.014 --> 00:05:34.994
and difficult to grasp what they are about.

00:05:35.254 --> 00:05:41.134
And to find such ways of navigating very complex, hard to define problem spaces,

00:05:41.534 --> 00:05:46.394
we cannot enumerate these problem spaces directly, but instead we go to programs

00:05:46.394 --> 00:05:47.734
that learn how to do this.

00:05:47.914 --> 00:05:52.734
So this new wave of deep learning is mostly about building systems that learn

00:05:52.734 --> 00:05:54.974
how to solve a particular kind of problem.

00:05:55.234 --> 00:05:59.594
And they're not general in the sense that we now have systems that can do this

00:05:59.594 --> 00:06:02.954
for any class of problems, but you basically set up a learning algorithm that

00:06:02.954 --> 00:06:06.254
learns how to play chess, or another learning algorithm that learns how to play Go.

00:06:06.514 --> 00:06:10.814
And we can now use very similar learning algorithms for the same thing.

00:06:10.954 --> 00:06:15.414
And then this thing just plays against itself and abstracts over what it sees,

00:06:15.554 --> 00:06:21.054
and then it outperforms everything humanity has done so far in this particular domain.

00:06:21.354 --> 00:06:25.514
And the next step might be that we go one step above this, which means we don't

00:06:25.514 --> 00:06:28.634
write this learning algorithm by hand for this particular task,

00:06:28.734 --> 00:06:32.594
but we write another algorithm which learns how to find that learning algorithm,

00:06:32.874 --> 00:06:34.574
which means we go to meta-learning, right?

00:06:34.694 --> 00:06:38.614
And it is tempting to think of our brain not as a learning system,

00:06:38.774 --> 00:06:43.174
but as a meta-learning system because many domains require very different learning

00:06:43.174 --> 00:06:44.954
strategies when we want to conquer them, right?

00:06:45.114 --> 00:06:50.334
And our brain is, in some sense, a machine that figures out how to learn a new type of thing.

00:06:50.874 --> 00:06:56.094
And from this perspective, evolution could also be seen as a very slow and unprincipled

00:06:56.094 --> 00:07:00.134
search for meta-learning systems to which we are the first solution that has

00:07:00.134 --> 00:07:01.034
managed to build computers.

00:07:02.920 --> 00:07:07.920
But now, okay, this is fine, right? Okay, fine, I take it. But in some sense,

00:07:08.040 --> 00:07:09.160
I could also say, well, there's no...

00:07:10.740 --> 00:07:15.940
I understood the words. I understood the logic. That gives me hope.

00:07:16.160 --> 00:07:21.080
But now I could also say, well, that's great, because that's the scheme that

00:07:21.080 --> 00:07:26.020
just shows that the author of the scheme understands recursion, right?

00:07:26.140 --> 00:07:30.820
And then you just follow the logic of recursion to say, ah, so this is now the

00:07:30.820 --> 00:07:34.940
logic I should follow, right? because they're unknowns. I have algorithms, oh, they're unknowns.

00:07:35.180 --> 00:07:37.820
Okay. Then I have to cover the space of puzzle algorithms.

00:07:38.340 --> 00:07:41.500
Ah, but when I have a trick to cover on the space of algorithms,

00:07:42.180 --> 00:07:43.940
there might be other spaces I have to cover.

00:07:44.080 --> 00:07:49.900
So the process seems to have a meta level that's actually not so complicated.

00:07:50.200 --> 00:07:54.140
And indeed, when I was growing up in sort of this transition from symbolic AI

00:07:54.140 --> 00:07:58.940
to what then called was new AI or embodied AI and new artificial life,

00:07:59.020 --> 00:08:01.480
whatever, ever, that was exactly really the dream.

00:08:01.640 --> 00:08:07.620
We want to use Jack algorithms to find algorithms that can define other algorithms, right?

00:08:07.700 --> 00:08:11.200
So we're in a loop because this is late eighties, right?

00:08:11.360 --> 00:08:16.600
Yeah. So, so what should give us any hope that also as a research program,

00:08:16.940 --> 00:08:20.280
this has any leverage and any traction?

00:08:20.560 --> 00:08:23.820
Because in some sense, also, when I see the proponents of this approach.

00:08:24.640 --> 00:08:28.700
Uh, going around all for many years, telling me these kinds of stories,

00:08:29.200 --> 00:08:32.560
It's not necessarily you, right? It's just a general reflection on that field.

00:08:33.180 --> 00:08:36.640
If we then pose a simple question, okay, but then show me an example that I

00:08:36.640 --> 00:08:40.220
find impressive or that can convince me also from a scientific perspective.

00:08:41.320 --> 00:08:45.880
Examples are maybe a bit poor, right? So in that sense, for me,

00:08:45.900 --> 00:08:48.700
things like AlphaGo, you can also say, okay, it just means it's another game

00:08:48.700 --> 00:08:49.760
that's not interesting, right?

00:08:49.840 --> 00:08:58.620
So why should this program or this logic actually get us to this promised land, right?

00:08:58.620 --> 00:09:04.340
So why should I have any kind of confidence in that track and trajectory that

00:09:04.340 --> 00:09:08.500
people have already announced for the last 30 years, along which we have not

00:09:08.500 --> 00:09:10.860
found much, right? You're still in the desert.

00:09:11.080 --> 00:09:15.600
So you're going all to grumpy AI programmer that has gone bitter on me now.

00:09:15.740 --> 00:09:18.820
Okay. No, I understand this. I can relate to this.

00:09:18.820 --> 00:09:25.860
A grumpy scientist who tries to understand the reality he finds himself in,

00:09:26.020 --> 00:09:31.480
and he does that by trying to get traction on issues like explanation and prediction.

00:09:33.280 --> 00:09:39.200
But grumpy nevertheless. Yes. I'm old enough to remember that we thought for

00:09:39.200 --> 00:09:43.800
a long time Go is going to be for a long time outside of our reach because the

00:09:43.800 --> 00:09:46.400
search space is so fuzzy that we cannot really enumerate it.

00:09:46.480 --> 00:09:50.100
And the situations cannot easily be ordered like chess situation where you say,

00:09:50.100 --> 00:09:52.200
my queen is gone. This is probably very bad.

00:09:52.300 --> 00:09:56.460
And there should be something making up for this if I go down that part of the search track.

00:09:57.294 --> 00:09:59.574
And in gold, all these things are much, much harder to define.

00:09:59.694 --> 00:10:04.034
So we thought for a long time this might be out of our reach for maybe even

00:10:04.034 --> 00:10:06.094
our generation or the foreseeable future.

00:10:06.314 --> 00:10:09.994
And a system that combines human intelligence with machine intelligence is going

00:10:09.994 --> 00:10:12.094
to dramatically outperform the whole thing.

00:10:12.134 --> 00:10:15.514
And now we know if we add human intelligence to this, it's going to get worse, right?

00:10:15.574 --> 00:10:19.834
There is nothing that a human player right now can add to AlphaGo to make it better.

00:10:21.494 --> 00:10:26.194
And so I think this is a reason to have a slight bit hope, but it's a very unprincipled reason.

00:10:26.714 --> 00:10:30.314
To get to a more principled perspective, I think we need to take a step back

00:10:30.314 --> 00:10:33.254
and wonder what is it that a mind is doing, right?

00:10:33.334 --> 00:10:38.694
A mind is some system to make models in the service of regulation for an organism.

00:10:39.374 --> 00:10:43.654
And to make a model, you need to identify the meaning behind information.

00:10:43.954 --> 00:10:46.514
But what we get from the world is information, certain patterns.

00:10:46.634 --> 00:10:48.834
We try to predict other patterns with it.

00:10:48.914 --> 00:10:53.514
And the meaning of information is its relationship to change and other information.

00:10:54.274 --> 00:10:56.974
And that is what we call a model so when you

00:10:56.974 --> 00:10:59.854
have a blip on your retina the meaning of that blip

00:10:59.854 --> 00:11:02.514
is the relationship that you discover to other blips on

00:11:02.514 --> 00:11:05.154
your retina at the same point or at different points in

00:11:05.154 --> 00:11:07.994
time right and these models that you can come up

00:11:07.994 --> 00:11:11.514
with describe for instance the movement of people in a three-dimensional space

00:11:11.514 --> 00:11:14.874
while they exchange ideas and have an economy and all these wonderful things

00:11:14.874 --> 00:11:19.054
and there are a simulation generated in your brain to explain and predict the

00:11:19.054 --> 00:11:22.574
patterns on your retina you will not be able to do this for all the patterns

00:11:22.574 --> 00:11:26.354
some Some of them are just going to remain blips because you don't find a relationship.

00:11:26.654 --> 00:11:30.534
So we call those noise. And it's a lot of noise in our retina, right?

00:11:30.654 --> 00:11:35.514
So when we try to make sense of our universe, we get a few samples of bits at

00:11:35.514 --> 00:11:36.794
different points in time.

00:11:36.874 --> 00:11:40.174
We try to relate them by making these models. This is the task of our mind.

00:11:40.314 --> 00:11:45.094
So we already know our mind is a function approximator. And we also roughly

00:11:45.094 --> 00:11:48.574
know over how many data our mind is going to approximate.

00:11:48.614 --> 00:11:52.654
We know how many seconds we are alive. We know roughly the frequency at which neurons operate.

00:11:52.854 --> 00:11:58.854
We know our surface that basically is our systemic interface to the environment

00:11:58.854 --> 00:12:00.334
that we operate in and so on.

00:12:00.434 --> 00:12:05.494
So we have a rough idea of the ballpark in which we operate and over which we make models.

00:12:05.734 --> 00:12:09.334
Are you describing just one side of the elephant? Because now you're pulling

00:12:09.334 --> 00:12:13.614
at the tail and you say, okay, we need to compress information, right?

00:12:13.654 --> 00:12:18.294
So that's why we build models. But on the other hand, on the other side of the

00:12:18.294 --> 00:12:23.994
elephant, we have also the need to generate action, to control physical systems

00:12:23.994 --> 00:12:26.954
in real time. Yeah, in the service of regulation.

00:12:28.474 --> 00:12:30.754
That's fine. Yeah. But that doesn't

00:12:30.754 --> 00:12:34.834
necessarily satisfy this whole idea of compression and model building.

00:12:35.294 --> 00:12:39.054
Because now I have to sort of flip it around and sort of again decompress into

00:12:39.054 --> 00:12:42.154
a real world that is open. open, right?

00:12:42.214 --> 00:12:46.754
You cannot necessarily capture all of that with the idea of model building and data compression.

00:12:48.868 --> 00:12:51.848
I'm not quite sure in which sense you would ever decompress.

00:12:52.208 --> 00:12:55.948
Because I don't interact with some decompressed world.

00:12:56.008 --> 00:12:59.008
I'm interacting with the world of my model, with the simulation,

00:12:59.128 --> 00:13:01.028
with the dream generated in my neocortex.

00:13:01.268 --> 00:13:03.348
This is subjectively the world that I inhabit.

00:13:03.628 --> 00:13:08.528
I'm never going to decompress you back into a part of the evolving state vector

00:13:08.528 --> 00:13:10.588
of the universe, into some weird quantum graph.

00:13:10.808 --> 00:13:14.268
Instead, I always interact with you as a person that has intentions,

00:13:14.568 --> 00:13:18.568
that has a certain geometry and so on. And so I'm always interacting with this

00:13:18.568 --> 00:13:19.648
data compression model.

00:13:19.828 --> 00:13:22.988
Yes, but with decompression, what I mean, you interact with me,

00:13:23.128 --> 00:13:27.108
I'm going to tell you things that you cannot fully predict, right?

00:13:27.248 --> 00:13:28.988
Yeah. I might say, right or wrong.

00:13:29.988 --> 00:13:34.808
So now there's a decompression sense of the way you interact with the world. The world is open.

00:13:35.268 --> 00:13:39.148
So the world now again generates highly variable feedback to you.

00:13:39.428 --> 00:13:42.588
Right? So you're compressing. the compressed states

00:13:42.588 --> 00:13:45.768
in the end get mapped to an action that you

00:13:45.768 --> 00:13:49.448
bring back to the world now the world doesn't get open right

00:13:49.448 --> 00:13:52.288
so now again you face a set of

00:13:52.288 --> 00:13:55.248
possibilities that collapse into one that's all you're exposed to

00:13:55.248 --> 00:13:58.148
yeah right and then again you go back

00:13:58.148 --> 00:14:01.968
into your compression mode and you generate your next action yeah right so that's

00:14:01.968 --> 00:14:05.748
more what I meant with the decompression part of it that your action is like

00:14:05.748 --> 00:14:10.228
this finger in this massive state space that generates yeah but basically what

00:14:10.228 --> 00:14:15.488
I do is I try to filter as many invariances from the world as I can and put them into a model.

00:14:15.668 --> 00:14:19.808
And everything else is the state of that model in as far as it is relevant, right?

00:14:19.868 --> 00:14:24.628
So I have a space of relatively few latent variables, just a few million of

00:14:24.628 --> 00:14:27.148
them, right? Or a few billion of them in which I operate.

00:14:27.408 --> 00:14:33.328
And I have a much, much larger static set of constraints that captures the laws

00:14:33.328 --> 00:14:38.908
of perspective and biology and intuitive physics and so on to make the world interpretable.

00:14:38.908 --> 00:14:43.048
So I can actually regulate my needs in a better way than a frog can.

00:14:43.228 --> 00:14:48.208
But with that, you're not necessarily saying anything that should knock me off

00:14:48.208 --> 00:14:50.768
my chair, right? Oh, that would surprise me.

00:14:53.088 --> 00:14:59.648
But this is a very reasonable way to summarize what a mind or an embodied mind would do.

00:15:00.968 --> 00:15:04.468
Should I apologize for suggesting a reasonable way?

00:15:04.508 --> 00:15:09.488
I suspect that we have even converged in many of our perspectives on AI have

00:15:09.488 --> 00:15:11.948
similar intuitions on how to get there.

00:15:12.068 --> 00:15:16.328
I don't know if our intuitions disagree on how long this is going to take and

00:15:16.328 --> 00:15:18.228
what is going to be necessary to get there.

00:15:18.468 --> 00:15:23.848
But it would be very surprising if we wouldn't be getting there even within our lifetimes.

00:15:24.128 --> 00:15:29.008
I think, David, I can understand where you're coming from in this sort of first,

00:15:29.048 --> 00:15:30.748
second, third-order, fourth-order systems.

00:15:30.988 --> 00:15:34.708
But people have been thinking about this for a long time. And...

00:15:36.715 --> 00:15:40.775
Whether it could be so neat is, I mean, it's kind of like a physicist's view

00:15:40.775 --> 00:15:43.235
of how things could possibly be.

00:15:43.395 --> 00:15:48.215
And you're a psychologist, so it surprises me you take such a sort of innate

00:15:48.215 --> 00:15:51.475
taxonomy of intelligent systems.

00:15:51.715 --> 00:15:54.495
Oh, no, I'm just trying to tell a story that fits into a podcast.

00:15:54.975 --> 00:15:59.635
It's brain, more complex than body. I mean, to say that intelligence is function

00:15:59.635 --> 00:16:02.655
approximation, which is a huge statement in itself.

00:16:02.755 --> 00:16:05.795
I mean, that's one of the things that is. What else is it?

00:16:07.055 --> 00:16:10.975
It's the ability to act appropriately in real time, and that may not require

00:16:10.975 --> 00:16:11.915
function approximation.

00:16:12.135 --> 00:16:16.515
The reason why I don't do this, why I don't lump this in this regulation,

00:16:16.575 --> 00:16:22.315
is because we observe that there are people that are very bad at acting appropriately.

00:16:23.235 --> 00:16:25.775
And they might nevertheless be extremely intelligent.

00:16:26.355 --> 00:16:30.515
So I would distinguish the ability from reaching your goals,

00:16:30.695 --> 00:16:35.195
which we could call smartness, from intelligence, which is the ability to make

00:16:35.195 --> 00:16:37.875
models, or wisdom, which is the ability to pick the right goals.

00:16:38.255 --> 00:16:42.355
But we find very often when we observe people is that excessive intelligence

00:16:42.355 --> 00:16:48.415
is often a prosthesis that our brain is using to compensate for the lack of

00:16:48.415 --> 00:16:50.155
being smart or being wise.

00:16:50.575 --> 00:16:56.095
Yeah. Okay. So we can have this particular definition of intelligence which

00:16:56.095 --> 00:16:58.395
maybe captures one aspect of humanity.

00:16:58.595 --> 00:17:03.035
Yes, of what minds are doing. Whether it can capture general intelligence you

00:17:03.035 --> 00:17:07.635
know, in terms of all the different kinds of intelligences that humans have,

00:17:07.695 --> 00:17:09.755
you know, sort of in a Howard Gardner sense.

00:17:09.995 --> 00:17:13.935
There might be multiple kinds of intelligence, maybe they're not all function approximation.

00:17:14.555 --> 00:17:19.175
But then, you know, if we look at what learning is, what learning to learn is,

00:17:19.335 --> 00:17:22.375
what searching for the optimal theory of function discovery is,

00:17:22.955 --> 00:17:28.055
these aren't necessarily, as you say, a recursive thing of applying the same

00:17:28.055 --> 00:17:30.015
idea multiple times to itself.

00:17:30.395 --> 00:17:35.135
They could be completely different. Oh, I fully agree. And so what we have in

00:17:35.135 --> 00:17:41.915
the current wave of AI is basically the optimization of inductive learning systems.

00:17:42.515 --> 00:17:48.915
But we still need deductive systems, and we need to discover abductive learning

00:17:48.915 --> 00:17:53.435
systems, which reason beyond the data, which no current deep learning system

00:17:53.435 --> 00:17:55.535
can do, but humans can do.

00:17:55.695 --> 00:17:57.855
And we don't know how humans do that.

00:17:58.055 --> 00:18:02.655
And we need to also automate the process of scientific discovery,

00:18:02.815 --> 00:18:05.355
and we have no idea how to do that either.

00:18:05.735 --> 00:18:09.815
Wait a moment. You might be too bold in your statements here.

00:18:10.755 --> 00:18:15.915
Me? Yes. I do remember when deep learning started to be… I think he's trying to be nice.

00:18:17.355 --> 00:18:22.515
When deep learning came up, basically everybody around me, including myself,

00:18:22.815 --> 00:18:25.815
started to make statements about what deep learning cannot do.

00:18:26.444 --> 00:18:31.724
And time and again, we had to revise our statements about what it can and cannot do.

00:18:31.824 --> 00:18:35.404
And it's a surprising thing to me. Still, it's surprising because you have these

00:18:35.404 --> 00:18:39.124
essentially mostly feed-forward network with very limited recurrence because

00:18:39.124 --> 00:18:40.464
we cannot train this very well.

00:18:40.524 --> 00:18:43.524
And very few lateral links because we cannot deal with those very well.

00:18:43.524 --> 00:18:49.644
And still, at some point, Google threw away everything that computer linguists

00:18:49.644 --> 00:18:53.124
had been done over decades in terms of machine translation and replaced the

00:18:53.124 --> 00:18:56.884
whole thing with an end-to-end trained, relatively dumb feed-forward network.

00:18:56.884 --> 00:19:02.604
And there is this thing, you know, if you think about your neural network,

00:19:02.744 --> 00:19:06.384
your feedforward network, not as representing a brain, something like a neural

00:19:06.384 --> 00:19:08.884
architecture with all these occurring connections,

00:19:09.084 --> 00:19:12.784
but with a function that correlates adjacent brain states,

00:19:13.004 --> 00:19:15.484
it's not quite clear what it cannot do, right?

00:19:15.484 --> 00:19:21.024
The question is, do we find a sufficiently defined loss function and an algorithm

00:19:21.024 --> 00:19:25.144
that converges quickly enough on a solution to that particular function?

00:19:25.444 --> 00:19:31.144
So when you want to find an algorithm that discovers truths in a particular

00:19:31.144 --> 00:19:34.284
mathematical domain, for instance, by discovering new theorems,

00:19:34.384 --> 00:19:38.524
it's not entirely clear to me that we can prove that our current deep learning

00:19:38.524 --> 00:19:40.664
approaches cannot get there. No way.

00:19:40.804 --> 00:19:47.104
You're going too quickly here, because if you take this example of machine translation by Google.

00:19:47.844 --> 00:19:52.704
Right, we have to, I think, really think also about the epistemic autonomy of

00:19:52.704 --> 00:19:55.804
these systems, right? Right, the old symbol rounding problem.

00:19:56.204 --> 00:20:02.424
Why could Google get away with throwing away a lot of the traditional methods,

00:20:03.104 --> 00:20:06.984
is they could also capitalize on the collective knowledge of a huge amount of

00:20:06.984 --> 00:20:12.424
humans who would just add exceptions and new information to the lookup tables they can build.

00:20:13.364 --> 00:20:18.024
So by virtue of that, the system actually worked. So that means epistemically

00:20:18.024 --> 00:20:19.164
it was not a common message.

00:20:19.924 --> 00:20:26.924
So in that sense, it didn't solve the problem of translation as biological systems

00:20:26.924 --> 00:20:28.784
as ourselves would form that.

00:20:28.944 --> 00:20:33.124
Of course. They found a workaround. Yes. Which has been really shrewd,

00:20:33.124 --> 00:20:37.284
if you want, and smart, and the way they capitalized on human knowledge is fantastic, great.

00:20:37.684 --> 00:20:41.784
But it doesn't mean that they really solved the actual challenge.

00:20:41.984 --> 00:20:43.164
They found a workaround.

00:20:43.824 --> 00:20:47.944
Of course, the question is, what is the challenge in terms of providing machine translation?

00:20:49.064 --> 00:20:52.184
If you... But we're talking about it in the account of mind,

00:20:52.344 --> 00:20:56.924
right? Of course, this is not a mind. Nobody says this. We all agree.

00:20:57.484 --> 00:21:01.384
So, yeah, it keeps surprising me. It keeps surprising me in the sense that I

00:21:01.384 --> 00:21:05.364
thought, oh, it does more than I thought it would do with these relatively simplistic methods.

00:21:05.604 --> 00:21:09.464
Because most of them we discovered in the 1980s already. Yeah, and I think...

00:21:09.984 --> 00:21:15.184
In the 1980s, certainly people like me, I had the faith.

00:21:15.284 --> 00:21:19.524
I thought these systems would be super powerful once we scaled computing power.

00:21:19.884 --> 00:21:24.784
I mean, I did my PhD in reinforcement learning, and it took me a week to train

00:21:24.784 --> 00:21:31.104
a little bug to move around and avoid obstacles, and now it would take half a minute or less, maybe.

00:21:31.484 --> 00:21:36.164
So you can look at the algorithm and know that it's a super powerful algorithm.

00:21:36.164 --> 00:21:41.604
In fact, given the right, you know, sort of, and AlphaGo is probably an amazing

00:21:41.604 --> 00:21:46.924
design of network, but you can see that reinforcement learning in itself is

00:21:46.924 --> 00:21:47.804
such a powerful algorithm.

00:21:48.244 --> 00:21:55.064
Given a state space that's so easily defined as Go, it ought to be able to do really well in that.

00:21:55.064 --> 00:21:59.384
So the things where I disagree with you is, I do think it's possible,

00:21:59.564 --> 00:22:02.984
or it's actually very likely, that the current class of algorithms is not going

00:22:02.984 --> 00:22:06.364
to carry us all the way. And we need to invent new classes of algorithms.

00:22:06.604 --> 00:22:11.304
We agree. Yes. But the thing where we disagree is, I cannot even prove that

00:22:11.304 --> 00:22:14.444
one. And I haven't seen a proof that they cannot do this.

00:22:14.784 --> 00:22:19.824
So as long as there is no proof for this, we don't know the limitations of these

00:22:19.824 --> 00:22:20.964
things. You're arguing against yourself, then.

00:22:21.004 --> 00:22:25.044
You're saying that we don't need to go to a possible discovery model.

00:22:25.044 --> 00:22:31.284
You put an uncertainty to the negative, where it's not even that one is justified.

00:22:31.424 --> 00:22:35.724
We don't know that. I think we don't know how far the current approaches can carry us.

00:22:36.004 --> 00:22:40.744
Well, it's very difficult to find a proof for the absence of something.

00:22:40.884 --> 00:22:44.884
So you're asking the impossible. So, no, well, you could say there are certain

00:22:44.884 --> 00:22:47.564
classes of algorithms that you can, in principle, not learn.

00:22:47.644 --> 00:22:52.064
If you could find a proof to show, okay, when Tony says something,

00:22:52.344 --> 00:22:55.784
we cannot get this thing to discover new theorems with this.

00:22:56.264 --> 00:22:59.324
This is a pretty bold statement, and you could probably formalize this.

00:22:59.524 --> 00:23:05.824
And then I could ask Tony, what is the evidence for that claim,

00:23:05.964 --> 00:23:08.124
except vague intuition? Well,

00:23:08.164 --> 00:23:12.924
no, I think scientific discovery requires interaction with the world.

00:23:13.689 --> 00:23:16.469
And that's what makes scientific discovery different. You know,

00:23:16.509 --> 00:23:21.729
you can't just, you need to go out there and you need to make observations and

00:23:21.729 --> 00:23:23.609
you don't a priori what to observe.

00:23:23.829 --> 00:23:26.449
None of these things in neural networks can do at the moment.

00:23:26.689 --> 00:23:31.609
And people have tried to automate that. But I think you guys are much too cozy there together.

00:23:32.329 --> 00:23:36.009
Because I don't understand why you guys are not in a state of panic over this

00:23:36.009 --> 00:23:37.469
challenge of understanding autonomy.

00:23:38.189 --> 00:23:42.329
Because this is actually, I wrote a few papers in the 90s already on that, analyzing that talk.

00:23:42.949 --> 00:23:47.289
NetTalk was one of the big breakthrough moments in connectionism because,

00:23:47.449 --> 00:23:50.629
oh, wow, we can map words to pronunciation features, right?

00:23:50.689 --> 00:23:55.369
And oh, wow, we have emergent vowels and consonants in terms of the pronunciation features.

00:23:55.929 --> 00:24:02.549
But the thing is, if you analyze the data, the patterns that were presented

00:24:02.549 --> 00:24:08.989
to NetTalk already encoded this distinct separation between vowels and consonants.

00:24:09.469 --> 00:24:13.589
So there was no magic there. It had everything to do with how humans had pre-coded

00:24:13.589 --> 00:24:14.649
the data, something that again

00:24:14.649 --> 00:24:18.809
means it was not epistemic, it was not epistemically autonomous, right?

00:24:19.029 --> 00:24:22.129
And this is the problem that Bumper has again all the time. These systems learn

00:24:22.129 --> 00:24:24.189
fantastically well on human-edited data.

00:24:24.709 --> 00:24:28.789
That's not helping us. So as long as we're not really solving this simple grounding

00:24:28.789 --> 00:24:30.609
problem without making progress, there's one issue.

00:24:30.929 --> 00:24:35.009
And I haven't heard an answer for that yet from your side of the table.

00:24:35.189 --> 00:24:40.529
You know, it's a round table, but okay. And secondly, you both seem to be very

00:24:40.529 --> 00:24:45.609
comfortable with this notion of intelligence, which worries me.

00:24:45.769 --> 00:24:51.329
Because intelligence is actually a very circumspect construct that McCarthy

00:24:51.329 --> 00:24:55.669
just pulled out of his head in some sense in the 50s to just give the field a name.

00:24:56.149 --> 00:24:59.489
Because they had computers, they had computers that would sort of imitate things

00:24:59.489 --> 00:25:02.089
that humans could do, like problem solving, and they had to give it a name.

00:25:02.089 --> 00:25:08.729
So he goes back to constructed go back to Galton late 19th century and Binet that were actually used.

00:25:09.389 --> 00:25:13.849
To make distinguish to distinguish between classes and races but for the rest

00:25:13.849 --> 00:25:18.409
it had no ontology to it and which was also exemplified by Binet he said well

00:25:18.409 --> 00:25:22.549
just what my test measures and even when Adrian Owen a few years ago put humans

00:25:22.549 --> 00:25:25.709
in the scanner and have them do the Raven progressive matrix test,

00:25:26.229 --> 00:25:30.389
for intelligence he saw that actually depending on the specific item you work

00:25:30.389 --> 00:25:31.889
on different parts of the brain and get involved.

00:25:32.069 --> 00:25:36.389
So there's only one underlying process, right? So how can you guys even in this

00:25:36.389 --> 00:25:40.749
search for a synthetic mind be so comfortable with this knowledge of intelligence?

00:25:41.049 --> 00:25:43.229
I think it's leading us astray. It's not helping.

00:25:43.749 --> 00:25:46.929
And the second thing is, if you want to insist on that, how are you going to

00:25:46.929 --> 00:25:48.529
solve this epistemic autonomy problem?

00:25:50.355 --> 00:25:54.115
So first of all, I don't think that intelligence is such a problematic concept.

00:25:54.315 --> 00:25:58.915
I do think that the notion of being able to make models is a very useful notion

00:25:58.915 --> 00:26:00.555
to understand what our minds are doing.

00:26:00.715 --> 00:26:03.995
And it's also clear that this is sometimes domain dependent.

00:26:04.075 --> 00:26:08.655
So we have certain priors in which areas we pay attention, which means where

00:26:08.655 --> 00:26:11.735
we put more resources into learning new models.

00:26:11.875 --> 00:26:16.215
And this leads to different performance. And so differences in people's innate

00:26:16.215 --> 00:26:21.175
or differences in talent is probably differences in innate attention to some degree.

00:26:21.995 --> 00:26:27.435
There are certain things that we as organisms are paid for to pay attention to, right?

00:26:27.535 --> 00:26:32.735
Paying attention, for instance, to facial expressions or to the emotional states of others and so on.

00:26:32.775 --> 00:26:36.335
And this will create different degrees of resolution in our mind with respect

00:26:36.335 --> 00:26:40.015
to these domains and bias us in a certain way.

00:26:40.695 --> 00:26:43.575
There's this other thing what animates us to do certain things

00:26:43.575 --> 00:26:47.235
are the needs of the organism that are implementing some

00:26:47.235 --> 00:26:50.215
kind of reward function that gives us pleasure and pain

00:26:50.215 --> 00:26:53.095
to reward us for doing certain things or punish

00:26:53.095 --> 00:26:58.015
us for doing other things and then anticipated pleasure and pain which make

00:26:58.015 --> 00:27:02.495
us go into certain areas of the behavior space or avoid certain areas of the

00:27:02.495 --> 00:27:07.675
behavior space in advance and eventually lead us to generate models about the

00:27:07.675 --> 00:27:11.415
structure of that that behavioral space and the policy space in which we operate.

00:27:11.795 --> 00:27:16.315
And so if you ever build a robot that plays soccer, it's not very difficult

00:27:16.315 --> 00:27:19.635
to make a robot want to go to the ball in some emergent sense, right?

00:27:19.835 --> 00:27:23.895
And if you have imagined that your robot has many more needs that it needs to satisfy,

00:27:23.995 --> 00:27:28.535
like recharging its battery from time to time, avoiding injuries,

00:27:28.635 --> 00:27:32.895
and maybe getting repairs from time to time when the injuries have accumulated

00:27:32.895 --> 00:27:35.315
to such a point that they impair its function.

00:27:35.835 --> 00:27:39.115
So you could imagine that this thing needs to satisfy a number of needs and

00:27:39.115 --> 00:27:42.295
achieve some kind of dynamic homeostasis about them.

00:27:42.395 --> 00:27:47.835
And suddenly you have a system with some motivational system that gives relevance to what it needs to do.

00:27:48.035 --> 00:27:52.175
And then when you think about your mind as some general function approximator

00:27:52.175 --> 00:27:57.435
that has, so to speak, a bit of cortical real estate that it can use to model

00:27:57.435 --> 00:28:00.415
certain domains of its environment with more or less resolution,

00:28:00.715 --> 00:28:04.375
then what you should be doing is whenever Whenever you have a problem that you

00:28:04.375 --> 00:28:07.875
cannot solve with your existing feedback loops, you need to create secondary,

00:28:07.895 --> 00:28:10.335
tertiary feedback loops. And at some point, this is not enough.

00:28:10.495 --> 00:28:15.455
So you need to spawn a completely computational model of that domain.

00:28:15.635 --> 00:28:18.935
So this is probably what our neocortex is doing. It creates a computational

00:28:18.935 --> 00:28:22.915
model of a domain so it can regulate this domain because you cannot regulate

00:28:22.915 --> 00:28:24.135
what you don't understand.

00:28:25.140 --> 00:28:29.180
Ashby's good regulator theory. If you build a regulator, then the structure

00:28:29.180 --> 00:28:33.160
that it implements needs to be isomorphic to the system that it regulates.

00:28:33.280 --> 00:28:37.060
And the world that we are in is a very complex, three-dimensional,

00:28:37.120 --> 00:28:38.200
dynamic, physical world.

00:28:38.300 --> 00:28:42.720
And then it's overlaid with a very complex, even more complex social world and

00:28:42.720 --> 00:28:46.920
economical world and so on that are in dynamic interactions with each other.

00:28:47.080 --> 00:28:49.400
And this is what our cortex, in some sense, is modeling.

00:28:49.640 --> 00:28:53.080
But not in an isomorphic way, I hope. You do abstract, right?

00:28:53.180 --> 00:28:54.920
And compress, as you said earlier.

00:28:55.140 --> 00:28:57.980
Yeah. So isomorphic to what, right?

00:28:58.020 --> 00:29:03.100
There is a certain world that happens there physically, and we already know

00:29:03.100 --> 00:29:06.360
that what we perceive as being the world is very, very different from this.

00:29:06.460 --> 00:29:08.340
So, for instance, there is no sound in the world.

00:29:08.520 --> 00:29:14.200
There are air molecules, at least been seen from some perspective, right?

00:29:14.320 --> 00:29:17.780
They are actually also not air molecules, but from some frame of reference,

00:29:17.920 --> 00:29:21.340
we can describe this stuff as air molecules and their electromagnetic forces

00:29:21.340 --> 00:29:23.720
between them that give waves of traveling pressure.

00:29:24.020 --> 00:29:27.100
And then there's some Fourier transform happening in our cochlea that measures

00:29:27.100 --> 00:29:29.940
the energy of these waves in different frequency ranges.

00:29:30.200 --> 00:29:33.840
And this gets sent in the neocortex. And now the neocortex does its very best

00:29:33.840 --> 00:29:35.980
to predict the structure of these patterns.

00:29:36.200 --> 00:29:42.340
The result is some interesting geometric calculation that our neocortex approximates.

00:29:42.380 --> 00:29:45.600
And these geometric patterns, we interpret as sound, right?

00:29:45.720 --> 00:29:51.020
And then we organize the sound into things like phonemes or into musical patterns and so on.

00:29:51.800 --> 00:29:57.440
That's fine but none of what you said necessitates the use of the word intelligence

00:29:57.440 --> 00:30:01.380
right because now you're talking about induction but the ability to build models.

00:30:02.617 --> 00:30:08.397
The ability to estimate the sources of the sensory states that you're exposed to.

00:30:09.557 --> 00:30:13.737
I think that before you get to the sources, a lot of other things are happening.

00:30:13.897 --> 00:30:18.557
This idea that all the patterns that you perceive go to sources is already a

00:30:18.557 --> 00:30:20.297
major inference that you're making, right?

00:30:20.417 --> 00:30:23.757
So you make these inferences because they give good predictions, right?

00:30:23.877 --> 00:30:28.217
But what you eventually try to do is you try to predict patterns.

00:30:28.217 --> 00:30:34.277
Patterns and maybe a good metaphor for this is to think of a synthesizer right a synthesizer is a.

00:30:35.337 --> 00:30:38.517
Haverset arrangement of electronic components that have

00:30:38.517 --> 00:30:42.117
the property if you connect them they go into weird oscillations

00:30:42.117 --> 00:30:46.057
and produce interesting dynamics and you have a few knobs that you can use to

00:30:46.057 --> 00:30:51.317
fiddle until it makes the sound that you want and imagine you have something

00:30:51.317 --> 00:30:55.657
like this built from neurons that the brain fiddles at the knobs until it's

00:30:55.657 --> 00:30:59.137
able to predict the patterns coming in from the cochlea in a few dimensions.

00:30:59.497 --> 00:31:03.257
And when it has done this exhaustively, it moves to the next layer and then

00:31:03.257 --> 00:31:06.877
tries to identify the covariance between the knob states, between the latent

00:31:06.877 --> 00:31:08.137
variables that it has discovered.

00:31:08.597 --> 00:31:12.077
And now uses this to generalize over the previous patterns.

00:31:12.297 --> 00:31:16.417
And it does this until it's able to fuse the different sensory modalities into

00:31:16.417 --> 00:31:19.557
a simulation of a dynamic free space with moving objects.

00:31:19.897 --> 00:31:24.557
It does that in the context of a superbly tuned body, a sort of physical apparatus,

00:31:24.557 --> 00:31:27.257
which provides all the transducers,

00:31:27.377 --> 00:31:32.257
which these, you know, sort of pattern recognisers, function approximators can

00:31:32.257 --> 00:31:38.257
work on, and already decomposes the world in a way that is well set up in order

00:31:38.257 --> 00:31:39.597
to be able to make sense of it.

00:31:39.937 --> 00:31:44.697
So, I mean, in your notion of function approximation doesn't really,

00:31:44.777 --> 00:31:48.157
can't really stretch to include this aspect of the.

00:31:48.780 --> 00:31:53.520
What some people call morphological intelligence. Yeah, I'm totally biased against

00:31:53.520 --> 00:31:55.440
that. I don't believe in morphological intelligence.

00:31:55.680 --> 00:31:58.560
It must be true that the body makes the challenge of the brain.

00:31:58.620 --> 00:31:59.520
What do you call this? Intelligence.

00:31:59.800 --> 00:32:02.780
Well, I don't like the word intelligence or morphological intelligence,

00:32:02.980 --> 00:32:04.240
but other people use the phrase.

00:32:04.480 --> 00:32:08.340
But it's certainly true that the body makes the challenge.

00:32:08.500 --> 00:32:12.800
The brain has much simpler. It makes it easier to control walking because your

00:32:12.800 --> 00:32:14.220
legs are in the pendulums.

00:32:14.360 --> 00:32:18.480
It makes it easier to do hearing because you have ears. You're talking as if

00:32:18.480 --> 00:32:22.640
you have never seen a toddler walking with legs that are absolutely not fine-tuned to walking.

00:32:23.100 --> 00:32:28.020
Your movement of your geometry of your body is optimized for your adult state.

00:32:28.180 --> 00:32:32.120
If you are a toddler, it's actually much harder to walk because all the centers

00:32:32.120 --> 00:32:34.680
of gravity are off, and you still manage to do this.

00:32:34.800 --> 00:32:39.220
In the same sense, you have all these nerves scattered over your skin.

00:32:39.500 --> 00:32:42.640
This is not done in a very particular order.

00:32:42.800 --> 00:32:45.640
Your brain is doing the co-occurring statistics over these nerves.

00:32:45.640 --> 00:32:49.600
That's not a bad way to go in because there are always examples where just the

00:32:49.600 --> 00:32:51.400
biomechanics just helps you.

00:32:51.580 --> 00:32:54.380
Think grasping, right? You just push the center of the hand and grasp.

00:32:54.620 --> 00:32:58.440
And this is true for any age or development, right?

00:32:58.860 --> 00:33:03.320
So if you push hard enough in the hand, it closes like this so you grasp something.

00:33:04.220 --> 00:33:08.140
So this helps you. You don't have to plan this. Oh, yeah. I'm not against the

00:33:08.140 --> 00:33:12.960
idea that your body is optimizing because it's subject to the same evolutionary

00:33:12.960 --> 00:33:15.220
constraints. But I wouldn't call this intelligence.

00:33:16.000 --> 00:33:19.600
Because your body is not making interesting models. But look,

00:33:19.700 --> 00:33:22.820
you and Tony both want the same intelligence. No, we're not on the same side

00:33:22.820 --> 00:33:27.840
here because you're saying that these models aren't grounded epistemically.

00:33:27.960 --> 00:33:31.220
Paul said this, and I'm agreeing with him. And I'm saying that what grounds

00:33:31.220 --> 00:33:35.420
those models is that they interact with the world through a body which is optimized

00:33:35.420 --> 00:33:37.160
through evolution to do that.

00:33:37.300 --> 00:33:41.900
So to think about intelligence in this entirely abstract way may not lead you

00:33:41.900 --> 00:33:45.200
very far because the embodiment is critical.

00:33:45.640 --> 00:33:49.360
To the way that human intelligence operates and has evolved.

00:33:50.120 --> 00:33:52.720
And we may not be able to build disembodied

00:33:52.720 --> 00:33:57.740
AIs just by having recursively bootstrapping and learning systems.

00:33:58.100 --> 00:34:00.980
And people are trying to do that and have been trying to do that for decades.

00:34:01.200 --> 00:34:04.560
And what I don't see is any breakthroughs in that or imminent breakthroughs.

00:34:04.700 --> 00:34:08.980
So maybe we need to follow this more biologically grounded approach,

00:34:09.260 --> 00:34:13.200
which is to say, let's build embodied AIs.

00:34:13.340 --> 00:34:16.200
That would be my approach. I think it's probably your approach,

00:34:16.280 --> 00:34:20.120
certainly false. I think it's a super popular approach. People like that approach, right?

00:34:20.380 --> 00:34:25.600
Well, because there's a lot of work. But if you go to NIPS, that's not what they're doing.

00:34:26.060 --> 00:34:28.180
Yeah, they do the stuff that actually works.

00:34:29.300 --> 00:34:31.660
Abortimentalism didn't work. Historically, it's been a failure.

00:34:31.820 --> 00:34:34.960
When this new AI stuff came up, people have been building lots of robots.

00:34:35.960 --> 00:34:40.140
Like Rolf Pfeiffer's Klapperaturen in Zurich developed.

00:34:41.588 --> 00:34:45.988
They're very impressive. It's very aesthetic, but it didn't get us intelligent systems.

00:34:46.448 --> 00:34:49.488
And so I'm of two minds with respect to this.

00:34:49.588 --> 00:34:52.988
I think that for our human intelligence, there's something that's quite different

00:34:52.988 --> 00:34:55.468
from what a lot of machine learning is right now doing.

00:34:55.568 --> 00:34:57.868
For instance, when we train a vision system in machine learning,

00:34:58.008 --> 00:35:01.648
the typical approach is we give it an annotated database with a few million

00:35:01.648 --> 00:35:04.388
images and ask it to generalize over this.

00:35:04.568 --> 00:35:08.988
Or we might even give it a non-annotated database of video streams and ask it

00:35:08.988 --> 00:35:11.608
to generalize over this. it's going to figure something out.

00:35:11.708 --> 00:35:15.288
But it's figuring something out that is slightly different from what we figure

00:35:15.288 --> 00:35:17.648
out, because, for instance, it's prone to adversarial examples.

00:35:17.848 --> 00:35:21.288
So you change a few pixels strategically in the input image,

00:35:21.408 --> 00:35:25.048
and suddenly instead of a cat, it sees an ostrich. How is that possible?

00:35:25.208 --> 00:35:29.248
It's probably because it's mapping this onto a space with a very different structure

00:35:29.248 --> 00:35:33.828
than the space that our minds is mapping these things on, which means it gives

00:35:33.828 --> 00:35:39.048
you a similar classification under most circumstances, but not under all circumstances.

00:35:39.048 --> 00:35:42.468
Especially it's not good at dealing with novelty so if

00:35:42.468 --> 00:35:45.228
you throw a end-to-end train system a picture for

00:35:45.228 --> 00:35:48.408
instance of a plane that is crashing into the tarmac it might

00:35:48.408 --> 00:35:52.368
recognize okay there is a plane and there is a tarmac and it might figure out

00:35:52.368 --> 00:35:55.888
oh there is a plane on the airport and this is what it's going to tell you it's

00:35:55.888 --> 00:35:59.348
going to miss out on the crucial interesting fact that this plane is about to

00:35:59.348 --> 00:36:04.168
crash and that thing is going to happen so how can we do this and i think it

00:36:04.168 --> 00:36:07.308
starts In this sense, I agree with this embodiment thesis.

00:36:07.588 --> 00:36:10.768
We start with our proprioception when we are born. It's already set up.

00:36:10.908 --> 00:36:14.528
And we already have made a map of our body surface before we are born.

00:36:14.748 --> 00:36:19.788
And then we add stuff to that map. We probably add stuff to that local perceptual space.

00:36:20.268 --> 00:36:23.888
And whenever we see something, we can work based on the assumption,

00:36:24.128 --> 00:36:28.368
what I'm currently looking at is a window into the same dynamic free space.

00:36:28.368 --> 00:36:35.648
So everything that I perceptually sense is a part of the same unified world.

00:36:36.412 --> 00:36:39.272
This is a very interesting thing. So when somebody shows me a picture of a cat,

00:36:39.392 --> 00:36:42.352
I don't see an isolated set of pixels that it now has to craftify.

00:36:42.572 --> 00:36:46.812
I know it's a person giving a picture to me of a cat that has been produced

00:36:46.812 --> 00:36:50.652
by a machine and at some point was a window in a different part of the same reality.

00:36:50.992 --> 00:36:55.532
This is the way I make sense of it. And it's very hard to fake such a universe, right?

00:36:55.752 --> 00:37:00.632
Wait, just before we go there, so I heard me challenge you that,

00:37:00.632 --> 00:37:05.652
so you were saying, look, a lot of people believe in this sort of embodied view of mind.

00:37:06.492 --> 00:37:08.252
I'm explicitly not saying intelligence.

00:37:10.232 --> 00:37:13.412
And then I challenged you and said, look, but if you go to NIPS,

00:37:13.532 --> 00:37:15.412
no one is doing it. And you said, it doesn't work.

00:37:17.332 --> 00:37:23.432
But the things that the people who go to NIPS believe work might work for the

00:37:23.432 --> 00:37:28.932
wrong reasons because the tools they build work because they rely on massive

00:37:28.932 --> 00:37:30.672
amounts of human annotated data,

00:37:31.672 --> 00:37:32.992
which is not the way forward.

00:37:33.032 --> 00:37:37.372
We know that will hit the wall somewhere, right? Well, then you said,

00:37:37.432 --> 00:37:38.612
well, I live in World War V.

00:37:38.812 --> 00:37:44.272
Well, I brought robots to World War V's lab in 1991, so I know wolves work pretty well, right?

00:37:45.495 --> 00:37:49.495
Then we have to be clear what does it mean to work, right? In our own trajectory,

00:37:49.815 --> 00:37:54.355
this whole embodied approach has worked extremely well because we have really

00:37:54.355 --> 00:37:59.675
explained aspects of the brain to the extent that we could make testable predictions

00:37:59.675 --> 00:38:02.735
that were confirmed to some extent, but were falsified.

00:38:02.915 --> 00:38:07.755
So from the perspective of a scientific program, this has been extremely successful

00:38:07.755 --> 00:38:12.255
to the extent that now we are repairing the brains of strong patients with, right?

00:38:12.255 --> 00:38:18.175
So it also depends very much on how you define it worked. So I think we should be very careful there.

00:38:18.775 --> 00:38:23.115
And so did we scale to anything you might want to call artificial general intelligence?

00:38:23.415 --> 00:38:28.835
No, that's the future. That's a challenge. But I think the situation frame-bodied

00:38:28.835 --> 00:38:33.055
models of mind are not that bleak. It just depends what you're looking for.

00:38:33.175 --> 00:38:38.615
And if your benchmark is what the computer scientists and the big tech companies

00:38:38.615 --> 00:38:41.655
are looking for, sure, then we might not be performing really well.

00:38:42.255 --> 00:38:46.435
But if you're interested in a more scientific program, I think we really have made huge progress.

00:38:46.735 --> 00:38:51.475
I think there is value there. And this might be an alternative approach to these

00:38:51.475 --> 00:38:54.515
questions than the more computer engineering-informed one.

00:38:54.675 --> 00:38:59.055
I think we should also take care to distinguish the practical work from the

00:38:59.055 --> 00:39:02.935
industries of AI hype and anti-AI hype.

00:39:03.155 --> 00:39:08.315
These are two variable industries that feed a number of people to produce AI

00:39:08.315 --> 00:39:13.555
hype or anti-AI hype. and there is actual research work and there's also scientific work.

00:39:13.695 --> 00:39:16.275
And these are slightly different things. And if you look at NIPS,

00:39:16.435 --> 00:39:17.815
it's a very diverse thing.

00:39:17.875 --> 00:39:21.775
I mean, it's about as large as Burning Man, but sells out in a quarter of the time by now.

00:39:22.155 --> 00:39:27.755
And there are a very large number of people that do very constrained vision

00:39:27.755 --> 00:39:29.235
staff or convolution network things.

00:39:29.455 --> 00:39:33.235
And there's also many, many other people. And the absolute number of people

00:39:33.235 --> 00:39:37.595
that wanted to do a science of intelligence under the umbrella of AI was always

00:39:37.595 --> 00:39:43.735
a pretty small minority. minority and I think the ratio of people that do this is almost unchanged.

00:39:44.175 --> 00:39:47.875
So the vast majority of people in AI do engineering.

00:39:48.315 --> 00:39:51.355
They basically build useful applications for data

00:39:51.355 --> 00:39:54.095
processing systems at the forefront of what we knew

00:39:54.095 --> 00:39:57.135
about how to do clever data processing and a small subset

00:39:57.135 --> 00:40:00.155
a few percent of these people are interested in actually

00:40:00.155 --> 00:40:03.295
how the mind works and dedicate their careers to that question and just

00:40:03.295 --> 00:40:06.035
because so many people now do AI doesn't mean

00:40:06.035 --> 00:40:09.775
that these others are not doing it is just it's the same fraction the absolute

00:40:09.775 --> 00:40:13.595
numbers are probably larger than ever in the history of AI and still it's a

00:40:13.595 --> 00:40:17.995
small subset and the absolute number of people that are interested in building

00:40:17.995 --> 00:40:21.875
models of minds and have very similar ideas to you and me with respect to this

00:40:21.875 --> 00:40:25.515
that are at NIPS are probably still a few hundred to a few thousand people.

00:40:26.075 --> 00:40:31.215
No sure I think this is a really important distinction it's okay so now we got that out of the way,

00:40:32.935 --> 00:40:37.615
everyone said the second slide exactly so,

00:40:39.491 --> 00:40:46.671
You did emphasize that consciousness has a role to play in reaching this synthetic mind, right?

00:40:47.171 --> 00:40:49.591
So what's your definition of consciousness?

00:40:50.651 --> 00:40:55.871
What's its function? And why is it such an important ingredient to synthesize mind?

00:40:56.451 --> 00:41:00.391
So there are several aspects to consciousness. And when we use this word,

00:41:00.511 --> 00:41:03.491
we often don't make clear what we mean by it, right?

00:41:03.551 --> 00:41:07.271
It also doesn't help. It's mostly a thing that we know by pointing and not by

00:41:07.271 --> 00:41:10.371
agreement about what we mean by its particular functionality.

00:41:11.231 --> 00:41:16.271
So there is something like alertness, the ability to react to stimuli in real time and so on.

00:41:16.351 --> 00:41:21.771
Then there is the ability to learn and update and the ability to reason about

00:41:21.771 --> 00:41:23.711
your behavior and reflect on it.

00:41:23.791 --> 00:41:28.431
There is the ability to attend to stimuli and high-level stimuli that we abstract

00:41:28.431 --> 00:41:32.531
in your brain into mental simulations and to remember that you attended to them

00:41:32.531 --> 00:41:36.691
and to to remember that the second order attention to this,

00:41:36.691 --> 00:41:39.811
so this reflexive consciousness above this excess consciousness.

00:41:40.071 --> 00:41:44.431
And then last but not least, there is this phenomenal experience,

00:41:44.691 --> 00:41:49.891
what it feels like to be in a particular situation, right? So maybe we start with this one.

00:41:50.151 --> 00:41:54.311
At the core of what it feels like, if you really meditate very,

00:41:54.411 --> 00:41:59.251
very deeply, you feel that there is, at least this is what appears to me like,

00:41:59.351 --> 00:42:02.911
a fundamental kind of disagreement with the universe at some level.

00:42:03.511 --> 00:42:07.751
And whenever you perceive a feature, it has a particular meaning with respect

00:42:07.751 --> 00:42:11.291
to this kind of disagreement. It gives you a particular kind of relevance.

00:42:11.551 --> 00:42:13.951
It's like zero-order consciousness.

00:42:14.331 --> 00:42:19.211
And then as soon as you start conceptualizing objects, before you think about

00:42:19.211 --> 00:42:23.011
them in terms of reflexive language and so on, you get something like first-order

00:42:23.011 --> 00:42:26.131
consciousness, where you see a particular thing and you feel,

00:42:26.211 --> 00:42:28.251
I am perceiving a particular thing.

00:42:28.251 --> 00:42:34.591
And when you go above this thing then you are in the conceptual level already

00:42:34.591 --> 00:42:37.891
that basically you now start carrying that thing around.

00:42:38.931 --> 00:42:43.171
Second order like in a glass bowl and you see this glow of consciousness with

00:42:43.171 --> 00:42:46.891
every step it gets a little bit darker but it's a symbolic manipulation that

00:42:46.891 --> 00:42:50.691
you perform to shift that thing around in your mind and do things with it and

00:42:50.691 --> 00:42:54.231
when I give a linguistic report about the whole affair to you then consciousness

00:42:54.231 --> 00:42:58.191
is a far distant memory to the processes that produce this or it's like a myth.

00:42:58.311 --> 00:43:02.391
It's something that my linguistic apparatus is not actually participating in very much.

00:43:02.691 --> 00:43:08.371
So there is something like a subjective feeling of being conscious in actuality.

00:43:08.871 --> 00:43:13.151
And the weird thing is that we probably cannot be conscious in actuality in

00:43:13.151 --> 00:43:15.711
the sense that a physical system cannot do that.

00:43:15.831 --> 00:43:18.551
The brain takes a long time to process stimuli.

00:43:18.791 --> 00:43:22.171
For instance, when I stop my tool, probably takes something in order of 700

00:43:22.171 --> 00:43:26.991
milliseconds for that to register in my neocortex in my integrated world model.

00:43:27.571 --> 00:43:30.991
But I have the impression not only do I perceive this in real time,

00:43:31.091 --> 00:43:33.331
but I also act on it in real time.

00:43:34.297 --> 00:43:38.077
Subjectively, I'm not just experiencing this with a nice time delay that can

00:43:38.077 --> 00:43:42.277
be filtered away with some editing trick in my mind, but I also have the impression

00:43:42.277 --> 00:43:46.317
it is me that acts on that stimulus which happens right now, right here.

00:43:46.497 --> 00:43:49.557
And that is not possible because the initiation of that action is going probably

00:43:49.557 --> 00:43:54.337
to take another second until I'm able to catch myself stumbling and react to

00:43:54.337 --> 00:43:58.337
that pain and integrate this all into a cohesive story, which means the story

00:43:58.337 --> 00:44:01.037
must be instantiated with hindsight. side.

00:44:01.737 --> 00:44:05.657
At least if physics is correct and we live in a mechanical universe.

00:44:06.717 --> 00:44:11.497
Subjectively, I don't live in a mechanical universe. So how do I what sense

00:44:11.497 --> 00:44:12.917
can I make of this dualism?

00:44:13.597 --> 00:44:18.917
Subjectively, I live in another universe in which I can react to me stopping

00:44:18.917 --> 00:44:22.037
my tour in real time and experience that, right?

00:44:22.177 --> 00:44:26.437
The thing is that you do but your conscious self doesn't.

00:44:26.457 --> 00:44:29.377
But your body does. and so this is the

00:44:29.377 --> 00:44:32.037
other thing with the conscious self which is

00:44:32.037 --> 00:44:35.017
problematic and your body is reacting

00:44:35.017 --> 00:44:37.917
as quickly as it can yes so in a way it reacts

00:44:37.917 --> 00:44:42.177
as quickly as it can but also not in real time but the problem is that you're

00:44:42.177 --> 00:44:46.577
identifying with your subjective consciousness not with the rest of your body

00:44:46.577 --> 00:44:51.177
so when you say I can't respond that fast yes you can if you identify with your

00:44:51.177 --> 00:44:56.757
toe and with the reflex systems in your spine that's as fast as you can respond,

00:44:57.784 --> 00:45:02.584
And then those systems report up to your stream of consciousness that this has happened.

00:45:03.044 --> 00:45:06.704
But I don't understand why that's a problem. I mean, scientifically,

00:45:06.884 --> 00:45:09.164
we know that that's what must be happening. Exactly.

00:45:10.604 --> 00:45:15.344
But this is what gives rise to this confusion that people, including many philosophers,

00:45:15.444 --> 00:45:17.584
have about that. Personally, I don't think it's a problem.

00:45:17.784 --> 00:45:21.064
So there is an answer to that conundrum, right? Before you agree.

00:45:22.284 --> 00:45:25.524
Joshua is making an assumption, which is not declared,

00:45:25.764 --> 00:45:30.584
which is relevant for this argument, because you seem to be saying there's something

00:45:30.584 --> 00:45:36.524
like behavioral time constants, and I hit my toe, and now within a certain time

00:45:36.524 --> 00:45:40.104
frame, I will have to respond to that or I'll fall on my nose. right?

00:45:40.724 --> 00:45:46.764
And the time constant of that process is so short that it's not reasonable to

00:45:46.764 --> 00:45:52.944
expect that the system that would support consciousness can operate at that same time constant.

00:45:53.364 --> 00:45:56.384
That seems to be the argument you're making. Yes, but it's not an assumption.

00:45:56.584 --> 00:45:58.664
It's something that we have established by measurements.

00:45:58.904 --> 00:46:03.864
We can put electrodes to somebody's skull and we can see how long it takes until

00:46:03.864 --> 00:46:05.244
the stimulus arrives there, right?

00:46:05.304 --> 00:46:08.144
So we know about this delay. laser we can do the table now

00:46:08.144 --> 00:46:11.524
yeah but you didn't declare that before okay that

00:46:11.524 --> 00:46:14.944
that's the argument hinges on that you say i have two processes one

00:46:14.944 --> 00:46:22.064
fast and slow and they cannot be running they cannot coalesce so therefore the

00:46:22.064 --> 00:46:26.444
slow conscious process by necessity is the story you tell yourself afterwards

00:46:26.444 --> 00:46:32.484
so no it's it's even worse the problem is that yeah that this This conscious process,

00:46:33.084 --> 00:46:37.184
that thing that I perceive as happening, is happening instantaneously.

00:46:37.184 --> 00:46:40.664
It's happening in actuality. It's happening here and now without any delay.

00:46:40.984 --> 00:46:44.824
It's not even happening fast. It's happening instantaneously.

00:46:45.124 --> 00:46:50.564
So how is that possible that I act instantaneously on me stabbing my toe when

00:46:50.564 --> 00:46:54.404
physics does not permit my neurons to transmit signals instantaneously,

00:46:54.504 --> 00:46:57.984
but does so in time frames that I should be able to perceive?

00:46:59.004 --> 00:47:01.984
Right because i uh i can distinguish time frames that

00:47:01.984 --> 00:47:04.944
are uh happening at something like half a

00:47:04.944 --> 00:47:08.604
second or a third of a second i don't get the instantaneous i

00:47:08.604 --> 00:47:13.484
don't get the instantaneous can you explain that so uh when i act based on when

00:47:13.484 --> 00:47:17.584
i act on my perceptions subjectively and acting on my perception in real time

00:47:17.584 --> 00:47:24.064
really yeah i see you moving your head and i'm reacting subjectively instantaneously

00:47:24.064 --> 00:47:25.784
on movement of your head.

00:47:26.944 --> 00:47:31.864
And of course the solution to this is this is a story that's been created after

00:47:31.864 --> 00:47:36.324
the fact to basically make it possible for me to learn from the past interactions

00:47:36.324 --> 00:47:39.964
that my nervous system has orchestrated with the physical world.

00:47:40.804 --> 00:47:44.264
Well maybe that's also a little bit tricky bit of subjectivity.

00:47:44.384 --> 00:47:45.964
I don't experience it like that.

00:47:46.564 --> 00:47:52.944
Because I live according to my own theory which is real time is subconscious.

00:47:53.244 --> 00:47:56.024
And God's experience is the way.

00:47:57.577 --> 00:48:01.477
I don't experience stomach-mite tongue in real time. I don't,

00:48:01.497 --> 00:48:03.497
not at all. Maybe you're not conscious.

00:48:04.257 --> 00:48:06.817
That's definitely possible. We can't exclude that option.

00:48:07.697 --> 00:48:12.697
But it seems… Why are we so surprised that folk psychology is wrong?

00:48:13.437 --> 00:48:17.117
And we grow up being… Well, this is not folk psychology. I'm talking about my

00:48:17.117 --> 00:48:18.057
subjective experience.

00:48:18.497 --> 00:48:22.717
Your subjective experience is folk psychology because that's what you've been

00:48:22.717 --> 00:48:26.957
told as soon as you were able to listen. and these were explanations.

00:48:27.317 --> 00:48:30.017
Nobody talked to me about consciousness when I was old enough.

00:48:30.037 --> 00:48:32.557
They talked to you about... Where did you grow up?

00:48:32.777 --> 00:48:36.377
They talked to you about the fact that you could control your toe and that if

00:48:36.377 --> 00:48:39.577
you stubbed your toe... No, my parents mostly stayed out of my way.

00:48:39.637 --> 00:48:41.257
I learned to control my toe by myself.

00:48:41.337 --> 00:48:45.257
Oh, but still, that doesn't matter, right? You may have evolved your own folk

00:48:45.257 --> 00:48:49.857
psychology, but unfortunately, you grew up into a scientist and then you discovered

00:48:49.857 --> 00:48:51.037
that it was largely wrong.

00:48:51.197 --> 00:48:54.757
So now that we've discovered that folk psychology is wrong, and we can just

00:48:54.757 --> 00:48:56.417
put it to one side and not worry about it.

00:48:56.477 --> 00:49:00.697
No, the thing is that what I have to explain is the set of phenomena that is

00:49:00.697 --> 00:49:03.737
subjectively available to me.

00:49:03.857 --> 00:49:07.937
So the reason that people get confused about consciousness is that they have

00:49:07.937 --> 00:49:12.457
a particular kind of experience and that experience seems to conflict with what

00:49:12.457 --> 00:49:13.437
is scientifically possible,

00:49:13.677 --> 00:49:19.197
which is why people like Tononi come up with very expensive theories, right?

00:49:19.317 --> 00:49:22.437
I mean, epistemologically and ontologically very expensive theories.

00:49:22.437 --> 00:49:25.677
There is the reason why they're willing to carry their burden it's

00:49:25.677 --> 00:49:28.457
because they're not going to give up on what you

00:49:28.457 --> 00:49:32.817
so derisively call folk psychology well i

00:49:32.817 --> 00:49:38.737
think that's absolutely true but i think they should yeah oh i agree well julio

00:49:38.737 --> 00:49:43.037
ends up being a panpsychist yes oh that's yeah yeah okay that depends what kind

00:49:43.037 --> 00:49:47.437
of whose folk psychology you're buying but okay he certainly gave up mine um

00:49:47.437 --> 00:49:52.097
but but the thing is i think you try to wiggle yourself out of a problem here Joshua.

00:49:52.397 --> 00:49:57.717
And I would like to remind you of the problem, which is, I really believe that I remember.

00:49:58.637 --> 00:50:04.717
That you said earlier that the constant experience of stubbing your toe was

00:50:04.717 --> 00:50:08.897
a story you told yourself afterwards, in that example. You didn't do it instantaneous.

00:50:09.757 --> 00:50:14.257
And then I was trying to help you, but of course, I was doing the wrong thing

00:50:14.257 --> 00:50:16.957
for the right reason, right? Or maybe it was all wrong.

00:50:17.537 --> 00:50:21.197
By saying, by just trying to, you're making this fundamental assumption that

00:50:21.197 --> 00:50:24.137
we have a dual process system that have different time calls.

00:50:24.237 --> 00:50:25.997
Oh no, that's not what I'm doing.

00:50:26.237 --> 00:50:29.017
There is something else going on.

00:50:30.229 --> 00:50:33.389
There is this traditional perspective of idealism, right?

00:50:33.429 --> 00:50:37.869
Idealist philosophy basically makes the statement that we don't live in a physical

00:50:37.869 --> 00:50:40.989
world, but that we live in a dream, in a magical world, essentially.

00:50:40.989 --> 00:50:46.029
One in which symbolic interactions are possible instead of just mechanical interactions

00:50:46.029 --> 00:50:50.569
that need to be performed on some fundamental, grossly closed physical level.

00:50:50.729 --> 00:50:53.549
Let's say some quantum mechanical level or below that.

00:50:53.549 --> 00:50:59.449
And in this symbolic universe certain things are possible like our minds exist

00:50:59.449 --> 00:51:04.189
and the dualist perspective said that these two domains this physical domain

00:51:04.189 --> 00:51:06.909
of objects moving in a space or,

00:51:07.789 --> 00:51:12.109
information flowing in some kind of computational graph of the universe and

00:51:12.109 --> 00:51:16.549
this mental domain where our mental phenomena take place somehow needs to need

00:51:16.549 --> 00:51:20.669
to coexist and what i want to suggest that this is in some sense the case.

00:51:20.849 --> 00:51:25.529
We are literally living in a dream that is dreamt by a mind on a higher plane of existence.

00:51:25.889 --> 00:51:28.689
And the physical universe is that higher plane of existence.

00:51:28.869 --> 00:51:33.069
And the mind that dreams us is generated by the brain of a primate.

00:51:33.189 --> 00:51:35.649
So in some sense, these perspectives are complementary.

00:51:36.309 --> 00:51:41.289
Subjectively, I live in a dream world. The same circuits that produce a dream

00:51:41.289 --> 00:51:45.849
at night produce a dream during the day to track and predict my sensory data.

00:51:46.189 --> 00:51:50.949
And the properties of the... and model building and compression and all that stuff, right? So, uh.

00:51:51.888 --> 00:51:56.608
And the difficulty that comes up, which is the reason why people like Tononi

00:51:56.608 --> 00:52:01.308
are so confused or why strong embodimentalism has such an appeal,

00:52:01.528 --> 00:52:05.808
which basically there is a different kind of embodiment which doesn't say,

00:52:05.888 --> 00:52:09.308
let's use robots because they're so useful to ground your perception.

00:52:09.588 --> 00:52:11.128
Which I totally agree with, right?

00:52:11.168 --> 00:52:14.888
If you are a mind, you should be able to use your body. That's a useful benchmark.

00:52:15.068 --> 00:52:18.528
And it's a very good source of data if you're connected to a physical universe.

00:52:18.528 --> 00:52:21.588
There's a lot of interesting variants and invariants

00:52:21.588 --> 00:52:24.388
in it right and if you have some regulation goals that are maybe

00:52:24.388 --> 00:52:27.368
given from an organism that make you interested in in

00:52:27.368 --> 00:52:30.808
cool stuff right so you build interesting models and interesting policies

00:52:30.808 --> 00:52:35.988
i agree with that perspective but there is a deeper perspective which says basically

00:52:35.988 --> 00:52:39.828
some people get up from their meditation and realize oh my god this is so awesome

00:52:39.828 --> 00:52:44.348
this cannot be explained by a mere computational process there needs to be something

00:52:44.348 --> 00:52:48.608
going on that is more than mechanical and where does this stuff have come from?

00:52:48.688 --> 00:52:51.948
Because my brain is just performing some boring computations,

00:52:52.068 --> 00:52:56.908
probabilistic excitations in these glorified fat cells that we use for processing

00:52:56.908 --> 00:53:01.008
information that we call neurons, right? So what is that thing?

00:53:01.168 --> 00:53:04.728
And for some people, they suggest, oh, maybe it's the touch of the mystical

00:53:04.728 --> 00:53:08.308
reality substance. Maybe physical reality does more than computation.

00:53:08.568 --> 00:53:11.748
And the mind is not something that just happens in the brain,

00:53:11.848 --> 00:53:16.728
but it's the result of some magical morphic resonance between your body and your environment.

00:53:16.828 --> 00:53:21.508
By getting in touch with the mystical reality of substance, you have this emergent

00:53:21.508 --> 00:53:27.488
and possibly extended mind that spreads outside into this universe that you're perceiving.

00:53:27.648 --> 00:53:31.228
And what you don't realize is that the universe that you're perceiving is not

00:53:31.228 --> 00:53:35.688
some direct access realist universe, but it's a dream generated inside of your

00:53:35.688 --> 00:53:37.628
neocortex. It's a simulation. It's a VR.

00:53:37.888 --> 00:53:40.928
Sorry, are you endorsing the idealist perspective? back to?

00:53:41.688 --> 00:53:46.428
No, I'm basically saying we should interpret the idealist perspective as something

00:53:46.428 --> 00:53:47.468
that in some sense is true.

00:53:47.548 --> 00:53:51.408
You can perform magic, you can learn how to levitate,

00:53:51.728 --> 00:53:56.088
you can learn how to look into the future, but you do this basically by editing

00:53:56.088 --> 00:54:00.128
your memories, by changing the way you produce your neocortex produces these

00:54:00.128 --> 00:54:02.868
models, and you pay a price for this because you cannot break the bank,

00:54:02.968 --> 00:54:04.828
because outside there's a mechanical universe.

00:54:05.128 --> 00:54:08.368
I think it's very strange, because you already declared this earlier,

00:54:08.588 --> 00:54:12.248
where you say, well, Well, by necessity, we build models.

00:54:13.791 --> 00:54:17.111
And as soon as you start to rely on these models to inform your actions,

00:54:17.311 --> 00:54:19.551
you live in a virtual world. It's virtual action.

00:54:19.791 --> 00:54:24.211
And everything else is just a consequence of that. That virtual… And that's fine.

00:54:25.551 --> 00:54:28.971
The mind lives in the physical body, which lives in a physical world.

00:54:29.151 --> 00:54:30.011
Yes, that's the interface.

00:54:30.551 --> 00:54:33.631
But there's one thing. There's no magic. Exactly. No, there's no magic.

00:54:34.111 --> 00:54:35.771
There's no conspiracy. It's good news.

00:54:37.031 --> 00:54:41.711
We can summarize the last hour in just eight words.

00:54:41.711 --> 00:54:45.051
Which are the mind-built

00:54:45.051 --> 00:54:48.711
models and as a result you virtualize but

00:54:48.711 --> 00:54:54.591
now you add you say you add to the physical world you add an additional word

00:54:54.591 --> 00:54:58.371
which is the higher plane why is it necessary why is the the physical world

00:54:58.371 --> 00:55:04.231
now the higher plane why it's not just we build models that are informed by

00:55:04.231 --> 00:55:08.511
the external world but we abstract away from it and those in former action.

00:55:08.671 --> 00:55:12.891
Oh, it's because the world that I live in, subjectively, is VR,

00:55:13.191 --> 00:55:14.391
right? Similar to Minecraft.

00:55:14.471 --> 00:55:16.931
It's a computer game generated in my brain, right?

00:55:17.111 --> 00:55:21.211
And the physical world is the parent universe to this. In this sense,

00:55:21.211 --> 00:55:22.491
it's the higher plane of existence.

00:55:23.051 --> 00:55:26.951
It's the substrate plane. You seem to suggest a dualism I don't like,

00:55:27.051 --> 00:55:30.611
because in Tony's case, earlier you were sympathetic to this one-way view of

00:55:30.611 --> 00:55:35.911
mind, and that actually builds a very solid interface with you.

00:55:35.911 --> 00:55:37.651
From the beginning that cannot indicle.

00:55:38.431 --> 00:55:43.471
Since it's physically instantiated, it's an embodied mind, it's by necessity

00:55:43.471 --> 00:55:45.831
always one with its physical world.

00:55:45.991 --> 00:55:48.491
No, no, no. You're going to me. This is not true.

00:55:50.851 --> 00:55:54.751
So basically, I think if you would replace your neurons with something that

00:55:54.751 --> 00:55:57.671
is functionally equivalent and performs the same computations,

00:55:57.771 --> 00:55:59.591
you could still be living in the same VR.

00:56:00.511 --> 00:56:04.711
Look, if I allowed you to make all these assumptions, the conclusion might be

00:56:04.711 --> 00:56:09.731
accepted. But I think these assumptions are extremely strong and you will not

00:56:09.731 --> 00:56:12.251
succeed in living up to them until the end of your career.

00:56:13.413 --> 00:56:16.273
Oh, no, it's just you're not a functionalist. That's all. And you maybe,

00:56:16.413 --> 00:56:18.493
I cannot change your mind. I can try.

00:56:19.173 --> 00:56:23.193
But for the same reason that you can change the CPU in your computer to something

00:56:23.193 --> 00:56:26.493
that is functionally equivalent in the sense that it's able to run the same

00:56:26.493 --> 00:56:30.033
software on it, you can run the same program on your computer,

00:56:30.113 --> 00:56:31.173
right? That was not the argument.

00:56:31.273 --> 00:56:35.293
The point is you describe the external world as a parallel universe, right?

00:56:35.933 --> 00:56:38.913
That's the word you used. No, as a different universe. I didn't say parallel.

00:56:39.753 --> 00:56:45.613
Well, we did separate it. We did separate. No, it's a different frame of reference.

00:56:45.853 --> 00:56:49.993
So in the same sense as, for instance, the software that runs on your phone

00:56:49.993 --> 00:56:53.193
does not exist in the same physical universe as the phone itself,

00:56:53.453 --> 00:56:56.993
because it's a frame of reference that you're using to describe it,

00:56:57.053 --> 00:56:59.853
which doesn't make sense in that other universe.

00:56:59.853 --> 00:57:02.653
John, I think it's a shortcut you're making, right?

00:57:02.713 --> 00:57:07.453
Because all the words you pronounce now that are the product of this biologically

00:57:07.453 --> 00:57:12.413
instituted term machine that you believe you are, right, resonate back to you

00:57:12.413 --> 00:57:15.733
as you speak them through the real world, right?

00:57:15.793 --> 00:57:19.393
And even will hit your skin, you gesticulate.

00:57:19.693 --> 00:57:27.453
So the embodiment is implicitly there also as an integral component of these

00:57:27.453 --> 00:57:29.553
models in which you rely. It's not decoupled.

00:57:29.773 --> 00:57:31.673
Now they're independent from each other.

00:57:32.573 --> 00:57:36.073
There's continuous feedback between them. I remember from reading your paper

00:57:36.073 --> 00:57:38.333
that you are basically an identity theorist.

00:57:38.713 --> 00:57:43.953
And I don't think that's a useful way of conceptualizing that space because

00:57:43.953 --> 00:57:46.053
I am not a monkey and I'm not that body.

00:57:46.193 --> 00:57:49.033
I'm a side effect of the regulation needs of the monkey.

00:57:49.193 --> 00:57:52.973
I'm the result of that mind. I'm a story that the mind tells itself.

00:57:53.353 --> 00:57:56.093
There are people which only... Because you just joined that...

00:57:57.011 --> 00:57:58.691
But, Dushan, I reject your story.

00:58:01.811 --> 00:58:06.191
With the extent that you're constructing an I, why not construct an I that identifies

00:58:06.191 --> 00:58:08.811
with the physical self as much as with the mental self?

00:58:09.471 --> 00:58:13.911
Because this identification doesn't hold. I realize that when I think about

00:58:13.911 --> 00:58:17.631
what this physical system is, that I really don't get behind it.

00:58:17.991 --> 00:58:22.631
I have difficulty to identify with a subset of the evolution of the state vector

00:58:22.631 --> 00:58:25.971
of the universe, because this is not what I perceive myself to be.

00:58:25.971 --> 00:58:30.391
What I perceive myself to be is very similar to a character in a novel that

00:58:30.391 --> 00:58:34.051
is told in a very rich perceptual language.

00:58:34.411 --> 00:58:36.571
And this character in this novel

00:58:36.571 --> 00:58:40.671
is something that the brain uses to tell itself its own story in a way.

00:58:40.731 --> 00:58:44.851
But it's not the story of a brain or it's not the story of a primate or an organism

00:58:44.851 --> 00:58:47.231
or a bunch of cells or a bunch of particles.

00:58:47.571 --> 00:58:51.751
It's something else entirely, right? Because the thing is, a brain is a physical

00:58:51.751 --> 00:58:53.611
system. It cannot feel anything. thing.

00:58:53.691 --> 00:58:56.531
It cannot experience anything. Neurons cannot experience anything.

00:58:56.871 --> 00:59:01.131
But it would be very useful for the brain to know what it would be like if it

00:59:01.131 --> 00:59:03.151
was somebody who could experience something.

00:59:03.471 --> 00:59:06.171
So what the brain is doing, it creates this VR of a world.

00:59:06.911 --> 00:59:12.711
And then within this VR, it creates a person, a simulacrum of a person.

00:59:12.811 --> 00:59:17.371
And it does think to that simulacrum that is halfway to a simulation. Wait, it does things?

00:59:17.691 --> 00:59:21.611
Yes. So basically, as an author of that dream that you are in,

00:59:21.691 --> 00:59:26.191
that's the system, a set of processes that generate your VR and that create

00:59:26.191 --> 00:59:29.111
your personal self, your model of what you think you are.

00:59:29.351 --> 00:59:34.311
And then it does things to that model, like expectations and the result of past

00:59:34.311 --> 00:59:39.511
actions, and then witnesses the reactions of that simulation and uses it for learning.

00:59:39.711 --> 00:59:45.451
I mean, then you're identifying with this thing which is flickering on and off

00:59:45.451 --> 00:59:48.651
because this simulation of the self isn't going on all the time.

00:59:48.671 --> 00:59:51.951
Most of the time you're operating in the real world. Your body's just doing things.

00:59:53.011 --> 00:59:56.591
Yeah, so the good thing is my brain gives that thing access to the language

00:59:56.591 --> 00:59:58.211
center. And here I am and talk to you.

00:59:59.951 --> 01:00:07.031
And here we are. You don't have to. It's totally optional. It's a bonus.

01:00:11.191 --> 01:00:15.331
No, but look, this is partially all reasonable because….

01:00:19.731 --> 01:00:25.331
Because it still fits the same model building story, only now you say,

01:00:25.431 --> 01:00:28.791
look, but there is other sources of information related to the embodiment itself,

01:00:28.851 --> 01:00:30.211
which also build models.

01:00:30.591 --> 01:00:35.771
But now you have a problem because you indeed actually explicitly declare that

01:00:35.771 --> 01:00:37.611
there is something observing that model.

01:00:38.111 --> 01:00:40.811
So you have to solve one of those problems. No.

01:00:42.851 --> 01:00:47.271
Okay. How does the observer work? Who cares about the observer?

01:00:47.311 --> 01:00:48.731
Why do you need an observer at all?

01:00:49.431 --> 01:00:53.191
There is not an observer in the sense that there is something looking at it.

01:00:53.311 --> 01:00:58.631
What is there is a system that takes parameters from that simulation and uses it for learning.

01:00:59.151 --> 01:01:03.491
So there is a loss function that is generated by producing a virtual character

01:01:03.491 --> 01:01:06.231
with certain intrinsic tendencies, behavioral tendencies.

01:01:06.231 --> 01:01:10.191
So it's a model of my own internal regulation. simulation and when something

01:01:10.191 --> 01:01:13.911
happens to me in my virtual simulated world like my simulated,

01:01:14.531 --> 01:01:17.691
love interest my partner decides to

01:01:17.691 --> 01:01:20.331
desert me which is something that might happen in

01:01:20.331 --> 01:01:26.551
my mental simulation then my brain in some sense observes what happens to that

01:01:26.551 --> 01:01:29.951
character which doesn't mean it's laying back and watch that movie it just takes

01:01:29.951 --> 01:01:34.031
a few parameters from that thing and uses it for learning right so it will play

01:01:34.031 --> 01:01:38.371
that scenario into let's say working memory and then you can an experience in that scenario.

01:01:38.591 --> 01:01:42.651
Yes, so it's basically a simulated world that does things to that particular

01:01:42.651 --> 01:01:46.451
thing, because it's a very high-level compression of the state vector of the

01:01:46.451 --> 01:01:49.511
universe and available data at my systemic interface to say,

01:01:49.611 --> 01:01:53.331
okay, you have a partner and that partner just left you. What do you feel about this?

01:01:53.951 --> 01:01:56.251
Sure, but it doesn't matter whether it's high-dimensional or not.

01:01:56.471 --> 01:02:01.951
So, you have a self-model, you say, then you have a model of the world.

01:02:02.611 --> 01:02:06.591
And these two models now interact with each other, because now the self model

01:02:06.591 --> 01:02:12.391
could say okay, let's push some parameters in the world model to create some

01:02:12.391 --> 01:02:14.231
scenario that would affect me.

01:02:14.231 --> 01:02:16.671
Let's see what kind of station it would give rise to.

01:02:17.031 --> 01:02:21.291
So it's like the coupling of two models. So you can do this to some extent, right?

01:02:21.331 --> 01:02:26.551
You can imagine what it would be like if you would get a lot of chocolate in an hour from now, right?

01:02:26.691 --> 01:02:29.411
And you can evaluate your own reaction to this.

01:02:29.671 --> 01:02:34.011
But if you are working correctly, you are not allowed to believe that this is

01:02:34.011 --> 01:02:36.591
going to happen without evidence that it's going to happen, right?

01:02:36.711 --> 01:02:38.511
Because otherwise you become delusional.

01:02:39.188 --> 01:02:43.608
Which means you are not free to choose the environment that you want to live in.

01:02:43.888 --> 01:02:47.468
It's a very paradoxical thing because objectively you might suffer because the

01:02:47.468 --> 01:02:49.688
world simulation that you're in does horrible things to you.

01:02:49.768 --> 01:02:52.968
And you might wish that not to happen. It's actually your brain at some level

01:02:52.968 --> 01:02:57.488
that does these things to you. And you could go and reprogram that brain to give you paradise.

01:02:58.168 --> 01:03:02.548
But of course it might mean that this organism dies. Because the model does

01:03:02.548 --> 01:03:03.648
not give very good predictions.

01:03:03.648 --> 01:03:10.048
And yet, I think that most of what these Buddhist techniques of fixing your

01:03:10.048 --> 01:03:14.968
conflicts with environment are doing is to reprogram that simulation to get

01:03:14.968 --> 01:03:16.708
your brain to like you better,

01:03:16.788 --> 01:03:19.248
to like that person better, and do nicer things to it.

01:03:19.388 --> 01:03:23.508
The Buddhists found a very simple trick to do that, because you solve all conflicts,

01:03:23.588 --> 01:03:25.848
you don't want anything. You stop wanting stuff.

01:03:26.148 --> 01:03:29.268
So turn down all your drive systems, and the world is great.

01:03:29.608 --> 01:03:32.308
There is more than one Buddhist tradition, of course. And what you describe

01:03:32.308 --> 01:03:35.208
is the sutra. So the sutric tradition is ascetic.

01:03:35.448 --> 01:03:39.248
You basically get rid of your needs by stopping to satisfy them.

01:03:39.368 --> 01:03:44.088
And at some point, this thing realizes, I don't need to satisfy them and look forward to this.

01:03:44.168 --> 01:03:48.268
So they don't create any discrepancy between target and current value anymore.

01:03:48.408 --> 01:03:52.368
You forget about this thing and you become free. And there's the tantric tradition,

01:03:52.608 --> 01:03:57.808
which does the opposite, which says basically embrace this maximally and satisfy all your needs.

01:03:58.748 --> 01:04:03.428
Accept that they are there and explore them and eat your shadow,

01:04:03.548 --> 01:04:06.508
integrate this all into a thing that actually really works for you.

01:04:07.588 --> 01:04:11.888
So can I ask you a question? So I mean, since I've been in Barcelona,

01:04:11.948 --> 01:04:15.128
I've been having vivid dreams. Now, you wanna share?

01:04:15.968 --> 01:04:19.268
I can tell you more later. But when I wake up in the morning.

01:04:20.337 --> 01:04:24.437
I experience the world in a different way than I've experienced my model of the world in my dream.

01:04:24.537 --> 01:04:29.577
Well, my model of the world seems very vivid and very real when I'm dreaming it.

01:04:29.757 --> 01:04:34.917
But then I wake up in the real world, and if that is a model of the world that

01:04:34.917 --> 01:04:38.977
I'm living in now, it's different in some interesting and important qualitative

01:04:38.977 --> 01:04:40.377
ways from my dream world.

01:04:40.717 --> 01:04:45.477
So, I mean, whereas, would you recognize that there's a distinction there?

01:04:45.477 --> 01:04:54.337
That the waking world that I'm in is somehow closer to maybe this physical world

01:04:54.337 --> 01:04:58.517
that you're permanently separated from.

01:04:58.797 --> 01:05:03.197
So I think you can, and certainly we've talked a lot about the mind-building

01:05:03.197 --> 01:05:06.637
models, but we can also talk about the world being its own model.

01:05:06.637 --> 01:05:11.577
But it's closer in the sense, ideally it's predictive of the patterns on your systemic interface.

01:05:11.577 --> 01:05:14.937
Phase so uh at night you are dissociated from

01:05:14.937 --> 01:05:18.217
your sensory input yeah which means that your dreams spin

01:05:18.217 --> 01:05:21.537
freely and you have no way to invalidate uh your

01:05:21.537 --> 01:05:24.397
hypothesis of the world state that you're in also dreams

01:05:24.397 --> 01:05:28.337
tend to relax your priors which means your model has more degrees of freedom

01:05:28.337 --> 01:05:31.817
than it would have during which gets us back to grounding so when i'm awake

01:05:31.817 --> 01:05:37.217
and when i'm not daydreaming i'm in a grounded state where i can't believe arbitrary

01:05:37.217 --> 01:05:42.777
things or or see cockroaches climbing off the walls, which I did the other night.

01:05:42.857 --> 01:05:45.457
I can't do that because they're just not there.

01:05:46.017 --> 01:05:50.857
So the world is grounding me in this awake state.

01:05:51.137 --> 01:05:56.017
So this makes me closer to the physical world than I would be if I was entirely

01:05:56.017 --> 01:05:57.257
living in a mental simulation.

01:05:57.657 --> 01:05:59.917
Do you think it's a very useful category to say closer?

01:06:00.417 --> 01:06:07.057
I mean, how close are you to quantum mechanics? But I can believe with some

01:06:07.057 --> 01:06:14.177
confidence that my perceptions in my everyday state reflect something about the physical world.

01:06:14.317 --> 01:06:16.557
So it's not that I'm trapped in my mind.

01:06:17.937 --> 01:06:21.737
Let's not mistake the physical world for quantum mechanics. They're two very different things.

01:06:22.534 --> 01:06:25.394
No, quantum mechanics is probably not an ontological theory.

01:06:25.554 --> 01:06:29.314
It's a particular kind of formal theory that allows you to describe low-level

01:06:29.314 --> 01:06:30.554
stuff. The only statement is valid.

01:06:31.054 --> 01:06:34.314
No, what I wanted to say is the world is more weird than quantum mechanics.

01:06:35.234 --> 01:06:39.974
The physical world is probably some computational system that progresses from

01:06:39.974 --> 01:06:45.674
state to state in a particular way, and we are an emergent pattern in this, as physical systems.

01:06:45.794 --> 01:06:48.354
Yeah, but nevertheless, there are things that can't happen in my awake life,

01:06:48.354 --> 01:06:52.234
unless they're true hallucinations. Yes.

01:06:52.354 --> 01:06:57.474
I mean, it's very much constrained what I can perceive by the physical world.

01:06:57.614 --> 01:07:01.874
Ideally, you should only hallucinate things that are matching the patterns on

01:07:01.874 --> 01:07:05.534
your retina, which have predictive value for the patterns on your retina.

01:07:05.674 --> 01:07:09.514
But the things that you're hallucinating are not actually like the things happening

01:07:09.514 --> 01:07:13.034
on your retina or the things that give rise to what happens on your retina.

01:07:13.134 --> 01:07:13.894
But they are interfaces.

01:07:14.234 --> 01:07:18.214
So, I mean, they're interfaces so that physical events can affect my own mind.

01:07:18.354 --> 01:07:23.394
An impact on my mind you might you might you might come to uh interpolation

01:07:23.394 --> 01:07:28.974
between states of the retina that as such have never occurred why not i don't

01:07:28.974 --> 01:07:33.674
see that problem you're very literal in that sense and then the other the other

01:07:33.674 --> 01:07:37.774
problem i have um among the many problems i have,

01:07:38.274 --> 01:07:43.514
with the one i will share with you now is that you agree with this idea of the

01:07:43.514 --> 01:07:46.574
embodied view on mind, so the machine view on mind.

01:07:47.114 --> 01:07:50.754
But I think we disagree very deeply about what machine means,

01:07:50.874 --> 01:07:56.434
because I think I would even accuse you of only thinking about the machine in

01:07:56.434 --> 01:07:57.214
terms of a Turing machine.

01:07:57.634 --> 01:08:01.514
And that's why I think also that computation is so central for you in the way

01:08:01.514 --> 01:08:02.814
you describe these phenomena.

01:08:03.354 --> 01:08:09.574
Well, again, I would see the Turing machine as just a thought experiment as

01:08:09.574 --> 01:08:12.594
a very ideal kind of… Yeah, it's way too powerful.

01:08:12.774 --> 01:08:15.334
We need finite state machines. No, but also physically they don't exist.

01:08:16.194 --> 01:08:20.734
At best it's approximated. It's a such… No, I agree. There are no infinite states.

01:08:21.054 --> 01:08:24.734
Exactly. So therefore, it's only a metaphorical machine.

01:08:25.614 --> 01:08:29.334
What I want to talk about is flash and bond machines that really physically

01:08:29.334 --> 01:08:30.214
exist and have an apology.

01:08:31.114 --> 01:08:39.074
So, isn't there a bias in that sense in analysis if we implicitly define machine as Turing machine?

01:08:39.214 --> 01:08:44.114
Wouldn't it be more useful to just consider scenarios where we're less tied

01:08:44.114 --> 01:08:46.794
to a notion of computation? Would you say the universe is computation?

01:08:46.994 --> 01:08:50.774
No. That's the way you can describe it. Okay. Let's try to disassemble this.

01:08:50.954 --> 01:08:54.774
So first of all, a machine is not part of the territory.

01:08:54.934 --> 01:09:00.254
It's part of your map, right? It's a certain part of the language that I use to describe the world.

01:09:01.000 --> 01:09:05.660
And what I mean by this particular concept in my language is a system that I

01:09:05.660 --> 01:09:09.940
can describe as a state space and a transition function that correlates these

01:09:09.940 --> 01:09:12.240
spaces in deterministic or probabilistic ways.

01:09:12.620 --> 01:09:16.820
And what's characteristic usually about machine descriptions is that the machine

01:09:16.820 --> 01:09:19.080
is able to exactly revisit earlier states.

01:09:20.080 --> 01:09:23.100
So, at least conceptually, it should be able to do that. So,

01:09:23.160 --> 01:09:25.480
it's a particular... But already that is problematic.

01:09:25.840 --> 01:09:31.040
Yes, it is. But this is not our point of contention here. So the important thing

01:09:31.040 --> 01:09:34.540
is machine is part of the language that I use to make sense of the world.

01:09:34.760 --> 01:09:39.420
And mathematics is the set of all languages, right? It's the domain of all possible languages.

01:09:40.160 --> 01:09:44.260
And then there is a set in which you can make proofs, which are the languages

01:09:44.260 --> 01:09:47.860
that you can formalize in a particular way. So we get very interested in those.

01:09:48.020 --> 01:09:50.860
And most of what we do in mathematics is these formal languages.

01:09:51.120 --> 01:09:55.660
And it turns out that in these languages, I can specify things that cannot work,

01:09:55.800 --> 01:09:59.700
like paradoxes, things that are conflicting with themselves,

01:09:59.940 --> 01:10:03.520
like the set of all sets that don't contain themselves. This cannot be implemented.

01:10:03.860 --> 01:10:07.920
And for something to exist, it probably needs to be implementable,

01:10:08.040 --> 01:10:14.040
which means it needs to be realizable in a way that it can produce certain functions,

01:10:14.120 --> 01:10:16.160
that it can perform certain things, right?

01:10:16.260 --> 01:10:22.440
That it can go from state to state and give rise to states at my interface to my environment.

01:10:22.660 --> 01:10:26.760
But I cannot know what that is, that thing, because I only get discernible differences.

01:10:26.840 --> 01:10:30.800
And even this, I don't know exactly if I really get the discernible differences

01:10:30.800 --> 01:10:34.560
that I think I get, right? So I had to make very conditional models.

01:10:34.800 --> 01:10:39.780
And it turns out that mathematics allows me to reason about these conditional

01:10:39.780 --> 01:10:44.660
models if I have evidence that my mind is cohesive enough to afford this reasoning.

01:10:44.880 --> 01:10:49.240
And the good news is the theory that my mind is cohesive enough to perform mathematics

01:10:49.240 --> 01:10:50.420
gives very good predictions.

01:10:50.900 --> 01:10:54.880
It's a very nice theory. So I can do some mathematics with my mind can start

01:10:54.880 --> 01:10:56.340
bootstrapping certain things.

01:10:56.560 --> 01:11:00.560
And one of the first things I discover, that mathematics discovered very late,

01:11:00.700 --> 01:11:04.460
is that there is a part of mathematics that doesn't have contradictions.

01:11:04.700 --> 01:11:09.280
And this is constructive mathematics. And the modern name for this is computation.

01:11:09.780 --> 01:11:13.520
So everything that happens as a computation cannot have a contradiction.

01:11:13.580 --> 01:11:16.400
Your computer can never be in an invalid state.

01:11:17.574 --> 01:11:21.614
And the traditional definitions of computation, which use infinite tapes and

01:11:21.614 --> 01:11:24.234
other things, don't fully realize that implication.

01:11:24.334 --> 01:11:28.474
So this gives rise to the Gödel problem and the Halting problem and so on,

01:11:28.514 --> 01:11:33.674
which are features of language specifications that introduce infinity and timelessness and so on.

01:11:33.814 --> 01:11:36.054
But in computation, these things disappear.

01:11:36.354 --> 01:11:39.534
Now, in computation, a few things are different from classical mathematics.

01:11:39.914 --> 01:11:43.394
For instance, pi is suddenly not a number anymore. It's a function.

01:11:43.554 --> 01:11:47.734
And it's a function that you cannot actually compute because nobody knows the last digit of pi.

01:11:47.954 --> 01:11:51.414
In order to compute the last digit of pi, you would need an infinite amount

01:11:51.414 --> 01:11:55.954
of energy to produce these computations if you happen to live in a reversible

01:11:55.954 --> 01:11:59.374
universe like ours, but you need energy to delete bits, right?

01:11:59.554 --> 01:12:03.854
So you will never know the last digit of pi if you live in the universe that

01:12:03.854 --> 01:12:08.174
you can mathematically deduce to be probably in, right?

01:12:08.334 --> 01:12:12.494
So we are not talking about reality as such. What we are talking about is mathematical

01:12:12.494 --> 01:12:18.514
models of hypothetical worlds. and then we try to discover the class of mathematical

01:12:18.514 --> 01:12:21.274
models that can explain our observations.

01:12:21.654 --> 01:12:26.434
And subjectively, we live in members of that class, but we don't live in the

01:12:26.434 --> 01:12:31.354
low-level members of that class because our brain is not sufficiently detailed to simulate them.

01:12:31.434 --> 01:12:36.994
Your hypothesis is that you do live in a subset of that class because your mathematical

01:12:36.994 --> 01:12:42.734
languages and the statement of mathematics is only true for closed worlds.

01:12:42.734 --> 01:12:44.934
If the boundary conditions are defined.

01:12:45.834 --> 01:12:50.774
And as life forms, the power of life is that we act against the second wave

01:12:50.774 --> 01:12:52.434
of thermodynamics as open systems.

01:12:52.854 --> 01:12:57.274
So there might be aspects of these systems that you cannot capture with this

01:12:57.274 --> 01:12:59.774
set of mathematical languages that you now advance.

01:13:00.314 --> 01:13:04.534
I'm not part of that religion. I think that... No, that's fine. That's fine.

01:13:06.074 --> 01:13:13.954
But still, you must convince me or... I'm trying. But you should explain why

01:13:13.954 --> 01:13:15.854
you believe that this specific language,

01:13:17.214 --> 01:13:22.914
that computational computation would be susceptible for giving us any kind of

01:13:22.914 --> 01:13:25.194
traction to deal with these open systems that we call life.

01:13:26.074 --> 01:13:31.014
I don't think that life is about open systems. The meaning of life on Earth

01:13:31.014 --> 01:13:33.554
is the hydrogenation of carbon dioxide.

01:13:35.014 --> 01:13:38.634
Roughly speaking. Life is cells. The meaning.

01:13:42.474 --> 01:13:45.694
Catastrophes is something that is not necessary. Only in the sense that people

01:13:45.694 --> 01:13:46.934
are asking for the meaning of life.

01:13:47.214 --> 01:13:50.414
Right. Of course, there is no meaning. But in the sense that people are asking

01:13:50.414 --> 01:13:53.934
for the meaning of life, why is it there? What does it do? I'm not asking for the meaning of life.

01:13:54.494 --> 01:13:59.234
I was saying your computational language might not be appropriate to understand

01:13:59.234 --> 01:14:01.534
life, including consciousness and everything in between.

01:14:01.754 --> 01:14:04.714
Okay. I'm trying to show you why I think that is the case.

01:14:04.994 --> 01:14:07.074
Of course. I'm sorry. Yes. Okay. Sure.

01:14:08.414 --> 01:14:12.594
So, what life is about, it sells. At some point in 1650, we didn't know what

01:14:12.594 --> 01:14:15.814
was life in the same sense as people in 1950 didn't really know what was intelligence.

01:14:15.814 --> 01:14:19.734
You could point at it, but you didn't know what it structurally, functionally was.

01:14:20.354 --> 01:14:24.254
So in the same way as I would say today, and you might disagree that intelligence

01:14:24.254 --> 01:14:29.054
is the ability to make models, I would say today, or most biologists would say this, life is cells.

01:14:29.714 --> 01:14:34.134
A cell is a molecular machine that is able to extract negentropy from the very

01:14:34.134 --> 01:14:35.354
right range of environments.

01:14:36.054 --> 01:14:40.374
And it has something built into it that is almost literally a Turing machine.

01:14:40.514 --> 01:14:44.114
You have this DNA tape with a red right head. and this thing implements an operating

01:14:44.114 --> 01:14:47.634
system on the cell which allows the cell to become part of a cellular automaton

01:14:47.634 --> 01:14:50.814
to create an organism. Not necessarily roughing states of that.

01:14:51.657 --> 01:14:55.497
It's reading it, but it can also write a few things. And some of it,

01:14:55.497 --> 01:14:59.457
it also writes in methylation. So we have this epigenomics, the flash drive, yes.

01:14:59.617 --> 01:15:05.377
But neurons seem to be breaking their DNA in particular ways to write something.

01:15:05.457 --> 01:15:09.257
But largely, of course, it's a ROM. We're mostly reading from this.

01:15:09.637 --> 01:15:13.937
And we also write state, but we do this in additional parts in the cell.

01:15:13.997 --> 01:15:16.497
So with methylation and changing the layout of the genome.

01:15:16.937 --> 01:15:24.237
Basically a few bytes that we can write. And what the cell can do is it basically measures its state,

01:15:24.337 --> 01:15:28.577
which is a chemical configuration based on processes that happen inside of the

01:15:28.577 --> 01:15:32.797
cell and processes that chemicals diffuse over the cellular membrane from the

01:15:32.797 --> 01:15:35.377
environment of the cell, which might also contain other cells.

01:15:35.557 --> 01:15:38.957
And based on this, the cell is going to go into a new state.

01:15:39.197 --> 01:15:43.137
And in the process of this, it either regulates something or it differentiates

01:15:43.137 --> 01:15:46.497
itself into another cell, which will behave differently when confronted with

01:15:46.497 --> 01:15:48.137
the same set of possible states.

01:15:48.677 --> 01:15:51.677
Or it might go into apoptosis, or it might divide itself.

01:15:52.037 --> 01:15:55.817
So your cell combines a neck entropy extractor, this state machine.

01:15:56.857 --> 01:16:02.557
Differentiation mechanisms that allow them to participate in evolution, and a self-replicator.

01:16:02.697 --> 01:16:05.857
In the absence of any of these components, your cell is not going to work.

01:16:06.017 --> 01:16:09.337
Once you have these components, which are incredibly complicated and probably

01:16:09.337 --> 01:16:13.477
require an enormous amount of rolls of the cosmic dice, maybe in the same order

01:16:13.477 --> 01:16:16.597
of magnitude as their available planetary surfaces is this in a sufficiently

01:16:16.597 --> 01:16:19.777
isotropic region of the universe, right?

01:16:21.537 --> 01:16:25.497
You have that thing live. You have this evolution of things that grow up.

01:16:25.597 --> 01:16:29.637
And what life can do, what other damp chemical processes cannot do,

01:16:29.797 --> 01:16:31.777
it can produce controlled reactions.

01:16:32.357 --> 01:16:37.277
We are in competition with damp combustion processes and redox reaction on this planet.

01:16:37.417 --> 01:16:40.557
And when they start, we cannot outperform them. And as a fire,

01:16:40.617 --> 01:16:41.697
we lose against the fire.

01:16:41.797 --> 01:16:45.497
The fire is so efficient in closing like entropy gradients. But there are some

01:16:45.497 --> 01:16:48.597
things that we can do, like we can do photosynthesis if you are a cell.

01:16:48.797 --> 01:16:52.037
So you need to add a little bit of energy to harvest more energy in the end.

01:16:52.999 --> 01:16:57.599
And what happened initially was that the cells, some of them discovered after

01:16:57.599 --> 01:17:03.679
a while a way to turn energy from the sun into their own structure by taking

01:17:03.679 --> 01:17:08.339
carbon dioxide from the atmosphere and changing it into their organic molecules

01:17:08.339 --> 01:17:11.079
and putting oxygen into the atmosphere.

01:17:11.079 --> 01:17:14.399
And they did this for a while and were very successful in this.

01:17:14.479 --> 01:17:18.079
And from the geological perspective, this was the major oxygenation event,

01:17:18.159 --> 01:17:21.619
a major event where suddenly you had a lot of free oxygen in the atmosphere.

01:17:22.659 --> 01:17:29.319
We're just agreeing with each other. So Paul said that animals are machines

01:17:29.319 --> 01:17:30.339
that can physically exist.

01:17:30.679 --> 01:17:35.679
You're saying that animals are computational entities, but you're defining those

01:17:35.679 --> 01:17:39.019
in a particular way such that there can be ones that can exist.

01:17:39.239 --> 01:17:40.539
So we're largely agreeing.

01:17:41.179 --> 01:17:44.459
Yeah, so now there happened a singularity at this point. At some point,

01:17:44.559 --> 01:17:50.699
some cells decided to combine together and form organisms, which made them participate

01:17:50.699 --> 01:17:51.819
in the multilevel selection.

01:17:52.119 --> 01:17:56.219
And they mostly became carnivores. They started eating other cells.

01:17:56.519 --> 01:17:59.999
We're not making progress here because the challenge was a somewhat different one.

01:18:00.679 --> 01:18:03.959
Oh, I'm giving you a computational description, and I'm trying to explain to

01:18:03.959 --> 01:18:06.039
you at some point why life is not an open system.

01:18:06.279 --> 01:18:09.399
You don't give me a computational description because you only admitted that

01:18:09.399 --> 01:18:12.939
the state machine of the cell is not a Turing machine, but it might be.

01:18:13.139 --> 01:18:16.599
No, it's just a finite state machine. It doesn't have an infinite tape.

01:18:16.699 --> 01:18:17.839
It's less than a Turing machine.

01:18:18.699 --> 01:18:21.539
It's not more. It's also not something different.

01:18:21.779 --> 01:18:25.599
Computational language might give you leverage in describing certain processes.

01:18:25.759 --> 01:18:26.879
I give you that. Not a problem.

01:18:27.399 --> 01:18:30.579
But you seem to give us some ontological stages. That was my concern.

01:18:31.739 --> 01:18:37.199
But we cannot discuss ontology. What we can discuss is epistemology and metaphysics.

01:18:37.199 --> 01:18:44.879
No, we can because you seem to give it some exclusive descriptive power that

01:18:44.879 --> 01:18:46.259
I want to question a bit more.

01:18:46.859 --> 01:18:51.119
That's why I was challenging you on this notion, is your notional machine a Turing machine?

01:18:51.339 --> 01:18:54.439
Because maybe if we think about, also we were talking about mind,

01:18:54.539 --> 01:18:58.839
synthesizing mind and consciousness, there might be properties to that that

01:18:58.839 --> 01:19:02.959
go beyond what we can capture easily from this computational perspective.

01:19:03.139 --> 01:19:07.659
It's basically, in the limit, it becomes a Turing machine. if you give it unlimited

01:19:07.659 --> 01:19:13.959
computational resources then the tree becomes another level it's also in a sort

01:19:13.959 --> 01:19:18.359
of open system energy is flowing across the boundaries otherwise it would die

01:19:18.359 --> 01:19:22.859
so this is just another way of describing what it's doing.

01:19:24.560 --> 01:19:29.080
So maybe there is a conflict here, but I think in some way that it reflects

01:19:29.080 --> 01:19:35.060
your view of the mind as being pure computation and sort of sitting inside this

01:19:35.060 --> 01:19:40.420
body and this physical world with which you can never have any direct contact

01:19:40.420 --> 01:19:44.180
because it's purely in the virtual sphere.

01:19:44.180 --> 01:19:49.080
Whereas I think Paul, myself, have more of a view that these are really,

01:19:49.140 --> 01:19:50.600
in the end, the same thing.

01:19:51.260 --> 01:19:54.700
The physical world generates the virtual mind.

01:19:55.100 --> 01:19:58.820
Oh, basically what I would say, there are in some sense the same thing,

01:19:58.880 --> 01:20:00.380
but they're described with a different language.

01:20:01.660 --> 01:20:05.780
So basically the language in which you describe mental events and in which you

01:20:05.780 --> 01:20:09.460
experience mental events is a language that is different from the language that

01:20:09.460 --> 01:20:10.920
we use to describe physics.

01:20:11.540 --> 01:20:14.960
So mental events don't exist in the physical realm.

01:20:15.420 --> 01:20:19.540
And it's also not all physical events exist in all languages of the physical

01:20:19.540 --> 01:20:22.400
realm. For instance, wetness doesn't exist at the level of molecules.

01:20:23.240 --> 01:20:25.820
Heat doesn't exist at the level of molecules. Mental events,

01:20:26.000 --> 01:20:30.260
like memory states, are physical events, right? There's an identity relation.

01:20:31.440 --> 01:20:36.020
You will know examples of that, right? From physiology, you look at play cells.

01:20:37.587 --> 01:20:41.947
My challenge is more to say, look, fine, you have a computational language,

01:20:42.067 --> 01:20:45.307
and Turing machines are fantastic, and Alan Turing is really a great...

01:20:45.307 --> 01:20:47.007
Wait a moment. No, no, I'm not convinced of this.

01:20:47.267 --> 01:20:50.867
I'm not convinced. No, no, that's an important distinction. There might be something

01:20:50.867 --> 01:20:52.367
going on here. It's not true.

01:20:52.647 --> 01:20:57.007
So, it could be that there are certain, for instance, that I have memories of

01:20:57.007 --> 01:21:00.527
conscious states that the physical system has actually never been in.

01:21:01.487 --> 01:21:05.367
From my perspective, this is a conscious state that I inhabited because I have

01:21:05.367 --> 01:21:08.587
a memory of having had this in the focus of my attention.

01:21:08.907 --> 01:21:12.987
But you might be able to show that my physical system actually was never there.

01:21:13.187 --> 01:21:17.267
So from my perspective, I have a language that describes something that's subjectively

01:21:17.267 --> 01:21:22.387
totally valid, the experience that is being valid, but it actually never took

01:21:22.387 --> 01:21:23.607
place in the physical universe.

01:21:23.767 --> 01:21:28.567
Because none of your memories ever took place. It's that all reconstructions

01:21:28.567 --> 01:21:33.747
of things and how you would like to think they happen.

01:21:33.807 --> 01:21:37.887
No, I think what you're doing is in order to say if your identity theory,

01:21:38.127 --> 01:21:39.887
you basically come up with this

01:21:39.887 --> 01:21:43.647
derisive notion of folk psychology to invalidate my actual experience.

01:21:45.767 --> 01:21:49.667
I'm saying that none of you… No, my experience is not identical to what happens

01:21:49.667 --> 01:21:52.127
physically. It's something else.

01:21:53.227 --> 01:21:57.947
The nature of memory… Because I have experiences that are not consistent with

01:21:57.947 --> 01:21:58.987
my understanding of physics.

01:21:59.307 --> 01:22:01.387
Yeah, but that's a very introspective argument.

01:22:01.827 --> 01:22:06.027
Yes. But basically, I caught myself editing my own memories,

01:22:06.127 --> 01:22:10.367
and I've had mystical experiences that are probably the result of my brain messing

01:22:10.367 --> 01:22:12.507
up its own memories or editing them.

01:22:13.687 --> 01:22:19.247
It's a necessary part of being a physical machine that you can't go back in

01:22:19.247 --> 01:22:20.667
the past and relive that experience.

01:22:20.747 --> 01:22:24.407
All you can do is reconstruct it, which is what memory is, and it's always going

01:22:24.407 --> 01:22:26.387
to be different from the actual experience.

01:22:26.527 --> 01:22:28.787
Yes, but how do you know that I ever experienced that thing?

01:22:29.707 --> 01:22:33.687
You don't. Yes, so actually... You don't need to. I could have.

01:22:33.807 --> 01:22:39.507
Based on the timing, I know that most of the things that I experience cannot

01:22:39.507 --> 01:22:42.087
physically take place in the way that I experienced them.

01:22:42.707 --> 01:22:45.387
And that's all fine. That's the fallibility of memory.

01:22:46.147 --> 01:22:48.967
No, it's not just the fallibility of memory. Memory is also creative.

01:22:49.207 --> 01:22:51.927
It's constructive. You compress, you build models, right? So,

01:22:51.927 --> 01:22:53.107
okay, big deal. No problem.

01:22:53.687 --> 01:22:59.267
It's basically, it's something else. There is the distinction between a novel and the book.

01:22:59.547 --> 01:23:02.367
If you look at the book and you open it, you see characters in it,

01:23:02.427 --> 01:23:07.187
which are actually just colored parts of the categories. Right. Yeah.

01:23:07.507 --> 01:23:11.127
There is no story in that book except from the perspective of a mind who reads it.

01:23:12.165 --> 01:23:16.705
And a similar thing happens to me I don't exist expect from the perspective

01:23:16.705 --> 01:23:22.265
of a mind who generates me and my own perspective who reflects on what I think I am,

01:23:22.845 --> 01:23:27.245
and that thing is distinct from this thing so there is no identity at some level

01:23:27.245 --> 01:23:28.485
between the novel and the book,

01:23:29.125 --> 01:23:33.505
you could say there is one because the novel doesn't have a being that is distinct

01:23:33.505 --> 01:23:41.765
from the book and the reader metaphor breaks down because your mind is not the story that's in the book,

01:23:42.545 --> 01:23:46.065
it's a bad metaphor because the book always assumes there's a reader.

01:23:46.265 --> 01:23:48.685
So now immediately you have your homogenous problem.

01:23:49.105 --> 01:23:52.305
You don't want to have this homogenous problem if you want to have a scientific theory.

01:23:52.805 --> 01:24:00.045
So I like the metaphor to give a gist of this idea of the problem of natural category is fine.

01:24:00.685 --> 01:24:05.345
That was the idea of that metaphor. So you see that. As such, I accept it. Yes.

01:24:05.445 --> 01:24:10.005
But you cannot push it too far and say, therefore, it's a story I tell myself

01:24:10.005 --> 01:24:12.925
about who I am, me, me as a story.

01:24:13.365 --> 01:24:17.005
I'm worried that you're going too far in an attempt to save identity theory.

01:24:17.925 --> 01:24:20.905
I don't think it's very elegant. I don't think it works.

01:24:21.505 --> 01:24:25.965
Yeah, but that functionalism... Wait, wait, wait. We don't have a disagreement about functionalism.

01:24:25.985 --> 01:24:29.345
There's the weak and strong functionalism, and you can be,

01:24:29.345 --> 01:24:33.045
you know, weakly functionalist and say that you can

01:24:33.045 --> 01:24:36.845
have mental states in things rather than brains but there's

01:24:36.845 --> 01:24:39.485
a lot of constraints around what those sorts of

01:24:39.485 --> 01:24:42.285
things can be uh and uh you know

01:24:42.285 --> 01:24:45.485
i think that's probably where i would be i don't know if paul is an identity theorist

01:24:45.485 --> 01:24:48.265
i think he probably is at least weekly functionalist because he wants to

01:24:48.265 --> 01:24:51.885
build robots who can have mental states so um

01:24:51.885 --> 01:24:54.985
but but then we can discuss what those constraints might

01:24:54.985 --> 01:24:57.725
be and i think we're making some progress on that

01:24:57.725 --> 01:25:00.505
yeah but whereas you're quite strong functionalist so you

01:25:00.505 --> 01:25:03.525
think you could you know take your mind maybe you could put

01:25:03.525 --> 01:25:06.585
it into some very different kind of computer and it

01:25:06.585 --> 01:25:13.405
will still be your mind uh whereas for me maybe for paul it's instantiated by

01:25:13.405 --> 01:25:16.705
brain it's very hard to imagine how it can be instantiated another any other

01:25:16.705 --> 01:25:21.585
kind of device but i'm still a functionalist because you know in theory at least

01:25:21.585 --> 01:25:26.565
it might be possible so uh to take a small step back from this thing.

01:25:26.665 --> 01:25:31.045
When we describe reality, we don't aim for a single narrative, right?

01:25:31.165 --> 01:25:36.085
What we aim for is the map of all possible theories that could explain the observations

01:25:36.085 --> 01:25:37.965
to the degree that we can discover them.

01:25:38.085 --> 01:25:41.045
And then we want to assign confidences in that theory space.

01:25:42.809 --> 01:25:46.369
To our best knowledge and meta-confidences on why we assign the confidences

01:25:46.369 --> 01:25:49.709
if we can, right? So this is somehow the process that we are doing.

01:25:49.889 --> 01:25:54.789
So when we have a disagreement, we should be able to understand the perspective

01:25:54.789 --> 01:26:00.209
of the other one in such a way that we could state it in ways that the other

01:26:00.209 --> 01:26:01.949
would agree that we stated this correctly.

01:26:02.489 --> 01:26:06.889
And then we should be able to exchange arguments about the different confidences, right?

01:26:07.049 --> 01:26:11.549
So for instance, there are people that are embodimentalists in the sense that

01:26:11.549 --> 01:26:16.089
I think that the mind must be the result of some resonance between body and environment,

01:26:16.229 --> 01:26:20.629
and brains are only incidentally present in that whole thing,

01:26:20.729 --> 01:26:22.649
which is a very strong thing.

01:26:22.769 --> 01:26:25.669
But I've met a number of philosophers that are in that camp.

01:26:26.849 --> 01:26:31.329
And then there are people which are strong idealists in the sense that I think

01:26:31.329 --> 01:26:34.309
there is no physical system behind that dream.

01:26:34.369 --> 01:26:37.009
There's nothing that computes God's mind, and I live in God's mind,

01:26:37.089 --> 01:26:40.609
and this is the lowest level of the thing that exists.

01:26:40.889 --> 01:26:43.789
There right and then we have different ideas that

01:26:43.789 --> 01:26:46.569
for instance there's some computational reducibility going on in

01:26:46.569 --> 01:26:51.089
the brain that leads to mechanisms that are so intricate that they produce the

01:26:51.089 --> 01:26:54.849
mind in a way that cannot be functionally abstracted in something that has fewer

01:26:54.849 --> 01:26:58.989
moving parts than the brain and then there's disagreement about how many moving

01:26:58.989 --> 01:27:03.049
parts we need right and this is now it starts to get interesting because now

01:27:03.049 --> 01:27:04.689
we get in the domain of things that

01:27:04.729 --> 01:27:07.769
we can possibly empirically test, right? Fair enough.

01:27:08.029 --> 01:27:13.609
So at this point, we can start maybe identifying interesting disagreements.

01:27:13.729 --> 01:27:19.489
So I cannot know whether it's possible to reduce a brain into functional units,

01:27:19.589 --> 01:27:23.329
for instance, at the level of cortical columns, before somebody has done this, right?

01:27:23.449 --> 01:27:25.649
So it's only a hunch that I'm having.

01:27:25.829 --> 01:27:31.269
And my hunch is currently that we can probably run a person in less than a terabyte

01:27:31.269 --> 01:27:36.809
of computer memory if we had the right kinds of algorithms, which we at the moment probably don't.

01:27:39.262 --> 01:27:44.322
Oh, it's just an upper bound based on the amount of data that you collect in your lifetime,

01:27:44.562 --> 01:27:48.582
a number of concepts that you can form, the number of computations that a neuron

01:27:48.582 --> 01:27:51.702
can make per second, functional organization into cortical columns,

01:27:51.902 --> 01:27:57.402
the kissing number of the cortical columns, and so on, which basically gives you the address space.

01:27:57.762 --> 01:28:00.882
So this is a guesstimate, let's say. Yes, it's all a guesstimate, right?

01:28:01.482 --> 01:28:04.882
It's all a ballpark figure. and when I

01:28:04.882 --> 01:28:08.162
was at MIT and taught a class about the future of AI I

01:28:08.162 --> 01:28:10.882
asked the students in that class to come up with their

01:28:10.882 --> 01:28:14.842
own estimate what they think so before we had discussions about different ways

01:28:14.842 --> 01:28:19.282
of making that estimate and the estimates range between something like 200 gigabytes

01:28:19.282 --> 01:28:25.202
on the low end and several exabytes on the high end depending on what people

01:28:25.202 --> 01:28:29.262
thought the level of granularity is the majority clustered around the low end.

01:28:30.142 --> 01:28:34.482
But of course this doesn't mean that we know that this is the true answer right

01:28:34.482 --> 01:28:36.662
we will only know if and when we succeed,

01:28:37.362 --> 01:28:41.842
in doing this and then we must look at arguments that are not driven by our

01:28:41.842 --> 01:28:46.042
intuitions or human vanity because we are that's a special species that cannot

01:28:46.042 --> 01:28:51.262
be possibly explained by a dumb computer it depends a lot on the physical machine

01:28:51.262 --> 01:28:56.942
in which you're instantiating the person that's right and there's another thing about memory that,

01:28:57.562 --> 01:29:02.022
as we discussed right Also from the model building, it exploits models to be

01:29:02.022 --> 01:29:03.502
regenerative and constructive.

01:29:04.102 --> 01:29:09.362
That's why you can imagine stuff you never experienced. So, so that means it's

01:29:09.362 --> 01:29:13.662
more about what are the parameters of these models, right? And how can you store those?

01:29:14.402 --> 01:29:19.682
Um, but to store a model might be more than just storing just flat bits, right?

01:29:19.822 --> 01:29:22.402
It's also relations now and functions that I have to.

01:29:23.599 --> 01:29:26.339
Yeah, transitional function. So, it might be gone.

01:29:26.459 --> 01:29:30.839
But for me, the question on the theory, we can make the theory space,

01:29:30.999 --> 01:29:33.099
we can start to compare and see what the disagreements are.

01:29:33.259 --> 01:29:37.859
But I think at the bottom of it, the fundamental challenge of any theory is

01:29:37.859 --> 01:29:41.319
to explain something and to make testable prediction and to control stuff.

01:29:41.539 --> 01:29:43.339
These are the three criteria for a theory, right?

01:29:44.619 --> 01:29:51.839
So, with respect to your theory, well, I see we just discussed 5% of it by now,

01:29:51.899 --> 01:29:54.059
but okay, Okay, we can do a follow-up of our podcast.

01:29:54.379 --> 01:29:57.739
We're still on lecture one, I think. Exactly. No, yeah, slide one.

01:29:57.999 --> 01:29:59.679
No, I think we're at slide two now, right?

01:30:02.019 --> 01:30:07.079
What's the prediction? What's the most convincing prediction about the theory

01:30:07.079 --> 01:30:09.899
which has been validated, right?

01:30:10.459 --> 01:30:14.859
And what has been the most fundamental aspect of mind that you think you have explained?

01:30:18.159 --> 01:30:23.099
So that's a difficult thing to say. For me, I've answered many questions that

01:30:23.099 --> 01:30:25.259
I had, and this was mostly conception analysis.

01:30:26.139 --> 01:30:29.579
So, for instance, when I started to go into the field of AI,

01:30:29.779 --> 01:30:32.939
I didn't know how it's possible that a system can have an emotion.

01:30:33.659 --> 01:30:35.799
What does it mean to be in an emotional state?

01:30:36.939 --> 01:30:39.999
Because I thought of an emotional state as something like a parameter within

01:30:39.999 --> 01:30:42.719
the system. What does make this parameter special?

01:30:43.039 --> 01:30:49.439
What makes an emotion different from a number or from a scalar or from a bunch of scalars, right?

01:30:50.039 --> 01:30:53.659
And at some point, I discovered that a way to make sense of this is that the

01:30:53.659 --> 01:30:55.059
emotion is a modulation of cognition.

01:30:55.539 --> 01:31:02.339
So you have parameters that are themselves not emotions, like arousal and valence and focus and...

01:31:03.541 --> 01:31:06.821
The securing rate which tells you you focus your

01:31:06.821 --> 01:31:09.681
attention inwards or outwards and so on and these together

01:31:09.681 --> 01:31:13.581
put you into a particular kind of configuration and this configuration means

01:31:13.581 --> 01:31:17.001
that your cognition is going to work differently you will perceive the world

01:31:17.001 --> 01:31:20.861
as differently you will perceive yourself as differently if you have a self-concept

01:31:20.861 --> 01:31:25.121
and these changes is what we actually mean by emotion so basically i realized

01:31:25.121 --> 01:31:29.541
yes this is explains sufficiently what i meant is this notion of emotion.

01:31:29.781 --> 01:31:34.021
So I could before point at a certain phenomenon, and now this language allows

01:31:34.021 --> 01:31:36.901
me to perform a conception analysis that makes it possible,

01:31:37.561 --> 01:31:41.581
for me to understand, oh, this is actually how emotion is probably implemented

01:31:41.581 --> 01:31:46.181
in an organism as a set of modulators that configure the cognition of that organism

01:31:46.181 --> 01:31:49.181
depending on the environmental and internal circumstances.

01:31:50.361 --> 01:31:54.661
So in this sense, it's not a predictive theory. It's an integrative model.

01:31:54.941 --> 01:31:59.081
And I've done this for a lot of things. For the predictive things,

01:31:59.181 --> 01:32:02.821
you can say, okay, we can come up with a machine learning algorithm that,

01:32:02.861 --> 01:32:04.581
for instance, identifies a cost function.

01:32:04.741 --> 01:32:09.461
You should think that the value of a cognition is proportional to the increase

01:32:09.461 --> 01:32:14.321
in expected reward if you perform it, minus the utility cost of not doing something

01:32:14.321 --> 01:32:16.281
else with these computational units, right?

01:32:16.401 --> 01:32:21.321
So if you implement this principle and you have lots of computational agents,

01:32:21.501 --> 01:32:27.201
do they perform useful modeling or useful computations in a given domain?

01:32:27.541 --> 01:32:32.441
This is an experiment that you can test. And many things in machine learning are in that area.

01:32:32.801 --> 01:32:37.901
But they're not theories of how the mind works. They are theories on how to

01:32:37.901 --> 01:32:38.921
implement a learning system.

01:32:39.221 --> 01:32:43.121
And when you look at reinforcement learning, for instance, many of the breakthroughs

01:32:43.121 --> 01:32:45.581
that we had there were the result of such considerations.

01:32:46.061 --> 01:32:51.461
So in a sense, AI has made many interesting predictions and useful theories.

01:32:51.461 --> 01:32:56.661
But without being able to build a thing that works like us, we will not know

01:32:56.661 --> 01:33:00.901
whether our models are actually models of a mind, or of something else that

01:33:00.901 --> 01:33:03.401
only performs certain functions that are very mind-like.

01:33:03.661 --> 01:33:07.361
The slide that I put up on day one was interesting.

01:33:08.016 --> 01:33:10.576
Said we can have an analytic approach, a synthetic approach.

01:33:11.056 --> 01:33:14.076
The synthetic approach can give rise to an existence.

01:33:14.556 --> 01:33:18.796
And the way I would look at what you're doing, you're following a synthetic

01:33:18.796 --> 01:33:22.996
approach, which is inspired by biology, but not necessarily literature.

01:33:23.436 --> 01:33:29.096
And you're looking to build existence proofs of your ideas that do things which

01:33:29.096 --> 01:33:32.536
animals and humans do. We might call them intelligent.

01:33:33.576 --> 01:33:38.036
And that's how you validate your theories. Yeah, I think that's fair.

01:33:38.136 --> 01:33:42.996
The problem is, if I would use an analytic approach, I would have to produce

01:33:42.996 --> 01:33:46.416
theories which have an enormous amount of free variables.

01:33:46.736 --> 01:33:49.576
And I would be very prone to overfitting of my models.

01:33:49.656 --> 01:33:53.816
It's very hard to make predictive models with very many free variables when

01:33:53.816 --> 01:33:57.516
you analytically look at the brain. So you gave, for instance,

01:33:57.556 --> 01:34:03.296
this example with making a model of the conic tomb of C. elegans, right?

01:34:03.436 --> 01:34:08.716
So we know it has 309 neurons, and we have mapped out the conic tomb.

01:34:08.796 --> 01:34:12.076
Unfortunately, we haven't really mapped out which of the neurons are excitatory

01:34:12.076 --> 01:34:15.876
or inhibitory, because we haven't really gotten to a model to the level of the

01:34:15.876 --> 01:34:17.016
individual vesicles there.

01:34:17.196 --> 01:34:22.836
So the search space is still very large. And if we map this out correctly,

01:34:23.136 --> 01:34:26.896
I think it's a good guess to say that this thing is probably going to behave

01:34:26.896 --> 01:34:28.296
somewhat C-elegance-like.

01:34:29.256 --> 01:34:32.256
At least it would be very surprising if it doesn't. Very interesting.

01:34:33.016 --> 01:34:37.056
But there are good reasons why right now our models are not really good enough for this.

01:34:37.156 --> 01:34:40.556
At least as far as I know, I haven't looked in that space in the last year or

01:34:40.556 --> 01:34:43.776
so and don't know of the new publications that have cracked the problem.

01:34:44.176 --> 01:34:48.476
But you probably would have heard about it, right? Yes. So it's also possible

01:34:48.476 --> 01:34:54.496
that the connectome of C elegans is computationally irreducible, which means that,

01:34:55.152 --> 01:34:58.952
it doesn't compress. It's possible that we don't find a description of what

01:34:58.952 --> 01:35:02.212
C. elegans is doing that is shorter than the full conic tome, right?

01:35:02.472 --> 01:35:06.332
So we would still expect that this thing does what C.

01:35:06.832 --> 01:35:11.352
Elegans is doing, but we cannot say anything beyond this, except it's a function

01:35:11.352 --> 01:35:17.472
which has 309 moving parts and complex interaction between those parts, right?

01:35:17.592 --> 01:35:21.672
Which means our minds are not very well equipped to do a further analysis on that model.

01:35:21.972 --> 01:35:25.792
But also this is a very bold thing, and we probably can compress it a little

01:35:25.792 --> 01:35:30.552
bit and identify some patterns in this and make sense of this on a higher level

01:35:30.552 --> 01:35:33.052
description, right? So this is what we would expect.

01:35:33.312 --> 01:35:37.992
But for me, the main prediction that I'm making right now is that at some point,

01:35:38.092 --> 01:35:40.252
humanity will build machines,

01:35:40.612 --> 01:35:44.712
building the current computational paradigms, which means building finite state

01:35:44.712 --> 01:35:48.612
machines that process information that are able to perform similar intellectual

01:35:48.612 --> 01:35:51.772
feats as we do, if global warming doesn't kill us before that.

01:35:51.912 --> 01:35:55.012
That's basically my prediction. It will be testable in some sense.

01:35:56.252 --> 01:36:00.112
But on a bit longish timescale. Yeah, we don't know that.

01:36:00.292 --> 01:36:05.052
So, there's of course this thing, Daddy, do you think that computers will ever be intelligent? Yes.

01:36:05.652 --> 01:36:09.172
Before I die. A little bit before that. Right, exactly.

01:36:10.112 --> 01:36:17.932
So, Joshua, before we hit the finish line, we made quite a tour through many

01:36:17.932 --> 01:36:21.272
aspects of the mind, how we can model minds, how we can think about it also

01:36:21.272 --> 01:36:22.132
from a theoretical perspective.

01:36:23.272 --> 01:36:27.612
And you have been really working very, very deeply and thinking very deeply

01:36:27.612 --> 01:36:28.832
about these challenges.

01:36:30.752 --> 01:36:36.172
So if we would like to follow in your approach to cracking this hard nut,

01:36:36.352 --> 01:36:38.012
what would be Joshua's law?

01:36:46.052 --> 01:36:50.372
It's a very interesting question what would be my law in the sense what uh what

01:36:50.372 --> 01:36:54.972
strategy should i be using to get to results what strategy should anyone use

01:36:54.972 --> 01:37:00.972
yeah recommend but what should tony be doing tomorrow what should those groups

01:37:00.972 --> 01:37:04.172
be doing instead of the stuff that they're doing now,

01:37:05.632 --> 01:37:08.572
unique direction give us a law,

01:37:09.252 --> 01:37:15.352
Joshua's law it really depends I mean from a career perspective my advice would

01:37:15.352 --> 01:37:22.492
be very different it should be printable on a t-shirt cannot have too many words yeah.

01:37:29.277 --> 01:37:32.457
So uh i think at some level you should

01:37:32.457 --> 01:37:36.837
be aware that you're you are a story that your mind tells itself and when you

01:37:36.837 --> 01:37:42.077
make a model of the world you do this by identifying relationships between information

01:37:42.077 --> 01:37:47.757
the most important rule there is that you have to understand that the meaning

01:37:47.757 --> 01:37:51.137
of information is its relationship to change another information and when you

01:37:51.137 --> 01:37:52.297
try to evaluate your a model,

01:37:52.397 --> 01:37:57.657
the first rule for that is that the weight of your confidence should equal the

01:37:57.657 --> 01:37:59.437
strength of the evidence to support it.

01:37:59.557 --> 01:38:03.437
There are no valid beliefs without priors. You basically need to track all of

01:38:03.437 --> 01:38:09.937
your priors until you form a ring, until you know the rules of your reasoning,

01:38:10.097 --> 01:38:13.837
until you know the basic axioms that go into your mathematical languages.

01:38:14.157 --> 01:38:18.957
And then you try to see if you can find a model that matches the data,

01:38:19.057 --> 01:38:21.617
that matches the patterns that you get and has predictive value.

01:38:21.997 --> 01:38:24.797
And once you've done this, you try to see if there are other rings,

01:38:24.937 --> 01:38:29.017
if there are other self-contained systems that have the same power. How's the law?

01:38:29.737 --> 01:38:36.277
So the law is, in some sense, follow the evidence. Try to make models that track the priors.

01:38:36.477 --> 01:38:40.777
Always follow your priors and resolve them until your belief stops being a verb.

01:38:41.677 --> 01:38:45.477
Because that system is now self-contained and no longer has a relationship to you.

01:38:45.597 --> 01:38:50.937
Once you understand a way to define arithmetic, that definition of arithmetic

01:38:50.937 --> 01:38:54.257
becomes independent of you. Your priors until you're a circular.

01:38:54.557 --> 01:38:58.157
Yes, until basically you get to a mathematically closed system.

01:38:58.257 --> 01:39:01.277
Every mathematically closed system in some sense is a tautology.

01:39:01.777 --> 01:39:05.737
Mathematics is a set of tautologies. Your destiny is to be a solipsist.

01:39:07.675 --> 01:39:10.315
No, why? I don't even exist, how

01:39:10.315 --> 01:39:14.435
can there be a solipsist? A solipsist is somebody who thinks they exist.

01:39:14.755 --> 01:39:18.175
No, that's a good law. I've pretty successfully deconstructed my own existence.

01:39:19.235 --> 01:39:24.155
Don't take that away from me. You're saying what Russell said in that clip,

01:39:24.195 --> 01:39:27.455
focus on the facts, not what you want to believe to be true.

01:39:27.695 --> 01:39:33.335
But in some sense, I realized that Tony and Paul and me are characters in a

01:39:33.335 --> 01:39:36.775
dream that is driven by a higher mind on a higher plane of existence.

01:39:36.775 --> 01:39:42.395
And the best explanation for that thing is that it was created in the brain

01:39:42.395 --> 01:39:46.795
of a primate in the course of an evolutionary trajectory, right?

01:39:46.955 --> 01:39:50.835
So this seems to be a theory that gives good predictions, seems to be sound,

01:39:50.975 --> 01:39:55.295
seems to be compatible with all the solid scientific results I'm getting.

01:39:56.975 --> 01:40:01.535
You're not going to our law now. Oh, no, no. No, the basic law is,

01:40:01.555 --> 01:40:05.255
I think, still this what Francis Bacon discovered, the first law of epistemology.

01:40:05.575 --> 01:40:08.815
The weight of the confidence must equal the strength of the evidence.

01:40:09.255 --> 01:40:14.095
And I think from this follows everything else, if you take this thing,

01:40:14.155 --> 01:40:17.075
because you can take this and check all the alternatives once you have this

01:40:17.075 --> 01:40:20.895
hypothesis that you should be doing this. The first law is Bacon's law. Yeah. Okay.

01:40:21.175 --> 01:40:24.815
Evidence is about… Oh, I cannot claim originality. I'm not a very original mind.

01:40:24.935 --> 01:40:27.495
Okay. I'm not that smart. But then, so the last question to you,

01:40:27.775 --> 01:40:32.995
Tony Wolf will chase you up four years from now.

01:40:33.375 --> 01:40:37.635
You might be in San Francisco by then or somewhere else. We don't know yet.

01:40:37.875 --> 01:40:41.355
Or still at Cambridge. I don't know what you're up to.

01:40:41.955 --> 01:40:46.555
But it sure will be a great place. But Tony is going to check whether you succeeded

01:40:46.555 --> 01:40:52.375
in falsifying or verifying a specific hypothesis that you're going to share with us today.

01:40:52.375 --> 01:40:58.775
So what the most urgent hypothesis that you would like to see really fully tested

01:40:58.775 --> 01:41:02.115
in this four-year timeframe, with respect to your theory of mind?

01:41:03.935 --> 01:41:04.575
Prediction.

01:41:10.259 --> 01:41:15.399
So one of the things that I find very interesting right now is whether the purpose

01:41:15.399 --> 01:41:17.319
of consciousness is attention-based learning.

01:41:17.559 --> 01:41:20.999
So there seems to be a problem. Our brains are not differentiable.

01:41:21.159 --> 01:41:25.959
They are not easily decomposed in a succession of layers through which we can

01:41:25.959 --> 01:41:30.399
pipe a loss function with the chain rule. So something else needs to be going on, right?

01:41:30.739 --> 01:41:35.339
And I suspect what's going on there is, among a few other things that I'm not

01:41:35.339 --> 01:41:38.299
seeing right now, an attention-based learning algorithm.

01:41:38.299 --> 01:41:42.319
And that might work by capturing the current binding state, the current latent

01:41:42.319 --> 01:41:45.719
variables that define the partial configuration of your brain,

01:41:45.779 --> 01:41:49.939
and a change that you are making to this configuration with respect to a particular

01:41:49.939 --> 01:41:51.319
goal and the expected outcome,

01:41:51.559 --> 01:41:55.539
and triggering conditions that tell you when you have a possibility to evaluate

01:41:55.539 --> 01:41:56.879
whether the change was successful.

01:41:57.339 --> 01:42:00.719
And then you commit this to an index,

01:42:00.899 --> 01:42:03.919
and conscious attention might be the ability to make index memories.

01:42:03.919 --> 01:42:09.499
And then you go to the future, you basically experience a few other world states,

01:42:09.619 --> 01:42:13.119
decompose this binding state to do other things, and then at some point the

01:42:13.119 --> 01:42:17.799
result manifests in the world, you reinstantiate this original binding state,

01:42:17.959 --> 01:42:23.819
and based on the outcome of your change, you reinforce the change or you undo it.

01:42:24.439 --> 01:42:29.039
And this could be a learning algorithm that sounds a little bit awkward because

01:42:29.039 --> 01:42:32.739
we need to do so many steps, but eventually we need to touch way fewer links

01:42:32.739 --> 01:42:34.979
than our current machine learning algorithms do.

01:42:35.219 --> 01:42:39.379
And what I would like to see is if this kind of algorithm actually works and

01:42:39.379 --> 01:42:43.779
produces useful results, and if we see a new family of learning algorithms that

01:42:43.779 --> 01:42:45.279
implement this or similar principles.

01:42:45.679 --> 01:42:49.819
Okay, very good. Lucia Boch, thank you very much for this conversation. Thank you too.

01:42:53.819 --> 01:42:58.879
The CSN podcast was produced by the Convergent Science Network of Biometrics

01:42:58.879 --> 01:43:05.639
and Biohybrid Systems, a project funded by the European 7th Research Framework Programme.

01:43:06.839 --> 01:43:12.159
For more interviews, recorded lectures or upcoming conferences in the field

01:43:12.159 --> 01:43:18.219
of biometrics and biohybrid systems, go to csnnetwork.com.

01:43:18.480 --> 01:43:26.800
Music.