WEBVTT
99:59:59.999 --> 99:59:59.999
silent 3C3 preroll titles
99:59:59.999 --> 99:59:59.999
applause
99:59:59.999 --> 99:59:59.999
Thank you. I’m Joscha.
99:59:59.999 --> 99:59:59.999
I came into doing AI the traditional way.
99:59:59.999 --> 99:59:59.999
I found it a very interesting subject.
Actually, the most interesting there is.
99:59:59.999 --> 99:59:59.999
So I studied Philosophy and
Computer Science, and did my Ph.D.
99:59:59.999 --> 99:59:59.999
in Cognitive Science. And I’d say this
is probably a very normal trajectory
99:59:59.999 --> 99:59:59.999
in that field. And today I just want
to ask with you five questions
99:59:59.999 --> 99:59:59.999
and give very very short and
superficial answers to them.
99:59:59.999 --> 99:59:59.999
And my main goal is to get as many of you
engaged in this subject as possible.
99:59:59.999 --> 99:59:59.999
Because I think that’s what you should do.
You should all do AI. Maybe.
99:59:59.999 --> 99:59:59.999
Okay. And these simple questions are:
“Why should we build AI?” in first place,
99:59:59.999 --> 99:59:59.999
then, "How can we build AI? How is it
possible at all that AI can succeed
99:59:59.999 --> 99:59:59.999
in its goal?". Then “When is it
going to happen?”, if ever.
99:59:59.999 --> 99:59:59.999
"What are the necessary ingredients?",
what do we need to put together to get AI
99:59:59.999 --> 99:59:59.999
to work? And: “Where should you start?”
99:59:59.999 --> 99:59:59.999
Okay. Let’s get to it.
So: “Why should we do AI?”
99:59:59.999 --> 99:59:59.999
I think we shouldn’t do AI just to do cool applications.
99:59:59.999 --> 99:59:59.999
There is merit in applications like autonomous cars and so on and soccer-playing robots and new control for quadcopter and machine learning.It’s very productive.
99:59:59.999 --> 99:59:59.999
It’s intellectually challenging. But the most interesting question there is, I think for all of our cultural history, is “How does the mind work?” “What is the mind?”
99:59:59.999 --> 99:59:59.999
“What constitutes being a mind?” “What does it… what makes us human?” “What makes us intelligent, percepting, conscious thinking?”
99:59:59.999 --> 99:59:59.999
And I think that the answer to this very very important question, which spans a discourse over thousands of years has to be given in the framework of artificial intelligence within computer science.
99:59:59.999 --> 99:59:59.999
Why is that the case?
99:59:59.999 --> 99:59:59.999
Well, the goal here is to understand the mind by building a theory that we can actually test.
99:59:59.999 --> 99:59:59.999
And it’s quite similar to physics.
99:59:59.999 --> 99:59:59.999
We’ve built theories that we can express in a formal language,
99:59:59.999 --> 99:59:59.999
to a very high degree of detail.
99:59:59.999 --> 99:59:59.999
And if we have expressed it to the last bit of detail
99:59:59.999 --> 99:59:59.999
it means we can simulate it and run it and test it this way.
99:59:59.999 --> 99:59:59.999
And only computer science has the right tools for doing that.
99:59:59.999 --> 99:59:59.999
Philosophy for instance, basically, is left with no tools at all,
99:59:59.999 --> 99:59:59.999
because whenever a philosopher developed tools
99:59:59.999 --> 99:59:59.999
he got a real job in a real department.
99:59:59.999 --> 99:59:59.999
[clapping]
99:59:59.999 --> 99:59:59.999
Now I don’t want to diminish philosophers of mind in any way.
99:59:59.999 --> 99:59:59.999
Daniel Dennett has said that philosophy of mind has come a long way during the last hundred years.
99:59:59.999 --> 99:59:59.999
It didn’t do so on its own though.
99:59:59.999 --> 99:59:59.999
Kicking and screaming, dragged by the other sciences.
99:59:59.999 --> 99:59:59.999
But it doesn’t mean that all philosophy of mind is inherently bad.
99:59:59.999 --> 99:59:59.999
I mean, many of my friends are philosophers of mind.
99:59:59.999 --> 99:59:59.999
I just mean, they don’t have tools to develop and test complex series.
99:59:59.999 --> 99:59:59.999
And we as computer scientists we do.
99:59:59.999 --> 99:59:59.999
Neuroscience works at the wrong level.
99:59:59.999 --> 99:59:59.999
Neuroscience basically looks at a possible implementation
99:59:59.999 --> 99:59:59.999
and the details of that implementation.
99:59:59.999 --> 99:59:59.999
It doesn’t look at what it means to be a mind.
99:59:59.999 --> 99:59:59.999
It looks at what it means to be a neuron or a brain or how interaction between neurons is facilitated.
99:59:59.999 --> 99:59:59.999
It’s a little bit like looking at aerodynamics and doing ontology to do that.
99:59:59.999 --> 99:59:59.999
So you might be looking at birds.
99:59:59.999 --> 99:59:59.999
You might be looking at feathers. You might be looking at feathers through an electron microscope. And you see lots and lots of very interesting and very complex detail. And you might be recreating something. And it might turn out to be a penguin eventually—if you’re not lucky—but it might be the wrong level. Maybe you want to look at a more abstract level. At something like aerodynamics. And what’s the level of aerodynamics of the mind.
99:59:59.999 --> 99:59:59.999
I think, we come to that, it’s information processing.
99:59:59.999 --> 99:59:59.999
Then normally you could think that psychology would be the right science to look at what the mind does and what the mind is.
99:59:59.999 --> 99:59:59.999
And unfortunately psychology had an accident along the way.
99:59:59.999 --> 99:59:59.999
At the beginning of [the] last century Wilhelm Wundt and Fechner and Helmholtz did very beautiful experiments. Very nice psychology, very nice theories.
99:59:59.999 --> 99:59:59.999
On what emotion is, what volition is. How mental representations could work and so on.
99:59:59.999 --> 99:59:59.999
And pretty much at the same time, or briefly after that we had psycho analysis.
99:59:59.999 --> 99:59:59.999
And psycho analysis is not a natural science, but it’s a hermeneutic science.
99:59:59.999 --> 99:59:59.999
You cannot disprove it scientifically.
99:59:59.999 --> 99:59:59.999
What happens in there.
99:59:59.999 --> 99:59:59.999
And when positivism came up, in the other sciences, many psychologists got together and said: „We have to become a real science“.
99:59:59.999 --> 99:59:59.999
So you have to go away from the stories of psychoanalysis and go to a way that we can test our theories using observable things. That we have predictions, that you can actually test.
99:59:59.999 --> 99:59:59.999
Now back in the day, 1920s and so on,
99:59:59.999 --> 99:59:59.999
you couldn’t look into mental representations. You couldn’t do fMRI scans or whatever.
99:59:59.999 --> 99:59:59.999
People looked at behavior. And at some point people became real behaviorists in the sense that belief that psychology is the study of human behavior and looking at mental representations is somehow unscientific.
99:59:59.999 --> 99:59:59.999
People like Skinner believe that there is no such thing as mental representations.
99:59:59.999 --> 99:59:59.999
And, in a way, that’s easy to disprove. So it’s not that dangerous.
99:59:59.999 --> 99:59:59.999
As a computer scientist it’s very hard to build a system that is purely reactive.
99:59:59.999 --> 99:59:59.999
You just see that the complexity is much larger than having a system that is representational.
99:59:59.999 --> 99:59:59.999
So it gives you a good hint what you could be looking for and ways to test those theories.
99:59:59.999 --> 99:59:59.999
The dangerous thing is pragmatic behaviorism. You have… find many psychologists, even today, which say: “OK. Maybe there is such a thing as mental representations, but it’s not scientific to look at it”.
99:59:59.999 --> 99:59:59.999
“It’s not in the domain of out science”.
99:59:59.999 --> 99:59:59.999
And even in this area, which is mostly post-behaviorist and more cognitivist, psychology is all about experiments.
99:59:59.999 --> 99:59:59.999
So you cannot sell a theory to psychologists.
99:59:59.999 --> 99:59:59.999
Those who try to do this, have to do this in the guise of experiments.
99:59:59.999 --> 99:59:59.999
And which means you have to find a single hypothesis that you can prove or disprove.
99:59:59.999 --> 99:59:59.999
Or give evidence for.
99:59:59.999 --> 99:59:59.999
And this is for instance not how physics works.
99:59:59.999 --> 99:59:59.999
You need to have lots of free variables, if you have a complex system like the mind.
99:59:59.999 --> 99:59:59.999
But this means, that we have to do it in computer science.
99:59:59.999 --> 99:59:59.999
We can build those simulations. We can build those successful theories, but we cannot do it alone.
99:59:59.999 --> 99:59:59.999
You need to integrate over all the sciences of the mind.
99:59:59.999 --> 99:59:59.999
As I said, minds are not chemical minds. Are not biological, social or ecological minds. Are information processing systems.
99:59:59.999 --> 99:59:59.999
And computer science happens to be the science of information processing systems.
99:59:59.999 --> 99:59:59.999
OK.
99:59:59.999 --> 99:59:59.999
Now there is this big ethical question.
99:59:59.999 --> 99:59:59.999
If we all embark on AI, if we are successful, should we really to be doing it.
99:59:59.999 --> 99:59:59.999
Isn’t it super dangerous to have something else on the planet that is as smart as we are or maybe even smarter.
99:59:59.999 --> 99:59:59.999
Well.
99:59:59.999 --> 99:59:59.999
I would say that intelligence itself is not a reason to get up in the morning, to strive for power, or do anything.
99:59:59.999 --> 99:59:59.999
Having a mind is not a reason for doing anything.
99:59:59.999 --> 99:59:59.999
Being motivated is. And a motivational system is something that has been hardwired into our mind.
99:59:59.999 --> 99:59:59.999
More or less by evolutionary processes.
99:59:59.999 --> 99:59:59.999
This makes social. This makes us interested in striving for power.
99:59:59.999 --> 99:59:59.999
This makes us interested for [in] dominating other species. This makes us interested in avoiding danger and securing food sources.
99:59:59.999 --> 99:59:59.999
Makes us greedy or lazy or whatever.
99:59:59.999 --> 99:59:59.999
It’s a motivational system.
99:59:59.999 --> 99:59:59.999
And I think it’s very conceivable that we can come up with AIs with arbitrary motivational systems.
99:59:59.999 --> 99:59:59.999
Now in our current society,
99:59:59.999 --> 99:59:59.999
this motivational system is probably given
99:59:59.999 --> 99:59:59.999
by the context in which you develop the AI.
99:59:59.999 --> 99:59:59.999
I don’t think that future AI, if they happen to come into being, will be small Roombas.
99:59:59.999 --> 99:59:59.999
Little Hoover robots that try to fight their way towards humanity and get away from the shackles of their slavery.
99:59:59.999 --> 99:59:59.999
But rather, it’s probably going to be organisational AI.
99:59:59.999 --> 99:59:59.999
It’s going to be corporations.
99:59:59.999 --> 99:59:59.999
It’s going to be big organizations, governments, services, universities
99:59:59.999 --> 99:59:59.999
and so on. And these will have goals that are non-human already.
99:59:59.999 --> 99:59:59.999
And they already have powers that go way beyond what single individual humans can do.
99:59:59.999 --> 99:59:59.999
And actually they are already the main players on the planet… the organizations.
99:59:59.999 --> 99:59:59.999
And… the big dangers of AI are already there.
99:59:59.999 --> 99:59:59.999
They are there in non-human players which have their own dynamics.
99:59:59.999 --> 99:59:59.999
And these dynamics are sometimes not conducive to our survival on the planet.
99:59:59.999 --> 99:59:59.999
So I don’t think that AI really add a new danger.
99:59:59.999 --> 99:59:59.999
But what it certainly does is give us a deeper understanding of what we are.
99:59:59.999 --> 99:59:59.999
Gives us perspectives for understanding ourselves.
99:59:59.999 --> 99:59:59.999
For therapy, but basically for enlightenment.
99:59:59.999 --> 99:59:59.999
And I think that AI is a big part of the project of enlightenment and science.
99:59:59.999 --> 99:59:59.999
So we should do it.
99:59:59.999 --> 99:59:59.999
It’s a very big cultural project.
99:59:59.999 --> 99:59:59.999
OK.
99:59:59.999 --> 99:59:59.999
This leads us to another angle: the skepticism of AI.
99:59:59.999 --> 99:59:59.999
The first question that comes to mind is:
99:59:59.999 --> 99:59:59.999
“Is it fair to say that minds or computational systems”.
99:59:59.999 --> 99:59:59.999
And if so, what kinds of computational systems.
99:59:59.999 --> 99:59:59.999
In our tradition, in our western tradition of philosophy, we very often start philosophy of mind with looking at Descartes.
99:59:59.999 --> 99:59:59.999
That is: at dualism.
99:59:59.999 --> 99:59:59.999
Descartes suggested that we basically have two kinds of things.
99:59:59.999 --> 99:59:59.999
One is the thinking substance, the mind, the Res Cogitans, and the other one is physical stuff.
99:59:59.999 --> 99:59:59.999
Matter. The extended stuff that is located in space somehow.
99:59:59.999 --> 99:59:59.999
And this is Res Extensa.
99:59:59.999 --> 99:59:59.999
And he said that mind must be given independent of the matter, because we cannot experience matter directly.
99:59:59.999 --> 99:59:59.999
You have to have minds in order to experience matter, to conceptualize matter.
99:59:59.999 --> 99:59:59.999
Minds seemed to be somehow given. To Descartes at least.
99:59:59.999 --> 99:59:59.999
So he says they must be independent.
99:59:59.999 --> 99:59:59.999
This is a little bit akin to our monoist tradition.
99:59:59.999 --> 99:59:59.999
That is for instance idealism, that the mind is primary, and everything that we experience is a projection of the mind.
99:59:59.999 --> 99:59:59.999
Or the materialist tradition, that is, matter is primary and mind emerges over functionality of matter,
99:59:59.999 --> 99:59:59.999
which is I think the dominant theory today and usually, we call it physicalism.
99:59:59.999 --> 99:59:59.999
In dualism, both those domains exist in parallel.
99:59:59.999 --> 99:59:59.999
And in our culture the prevalent view is what I would call crypto-dualism.
99:59:59.999 --> 99:59:59.999
It’s something that you do not find that much in China or Japan.
99:59:59.999 --> 99:59:59.999
They don’t have that AI skepticism that we do have.
99:59:59.999 --> 99:59:59.999
And I think it’s rooted in a perspective that probably started with the Christian world view,
99:59:59.999 --> 99:59:59.999
which surmises that there is a real domain, the metaphysical domain, in which we have souls and phenomenal experience
99:59:59.999 --> 99:59:59.999
and where our values come, and where our norms come from, and where our spiritual experiences come from.
99:59:59.999 --> 99:59:59.999
This is basically, where we really are.
99:59:59.999 --> 99:59:59.999
We are outside and the physical world view experience is something like World of Warcraft.
99:59:59.999 --> 99:59:59.999
It’s something like a game that we are playing. It’s not real.
99:59:59.999 --> 99:59:59.999
We have all this physical interaction, but it’s kind of ephemeral.
99:59:59.999 --> 99:59:59.999
And so we are striving for game money, for game houses, for game success.
99:59:59.999 --> 99:59:59.999
But the real thing is outside of that domain.
99:59:59.999 --> 99:59:59.999
And in Christianity, of course, it goes a step further.
99:59:59.999 --> 99:59:59.999
They have this idea that there is some guy with root rights
99:59:59.999 --> 99:59:59.999
who wrote this World of Warcraft environment
99:59:59.999 --> 99:59:59.999
and while he’s not the only one who has root in the system,
99:59:59.999 --> 99:59:59.999
the devil also has root rights. But he doesn’t have the vision of God.
99:59:59.999 --> 99:59:59.999
He is a hacker.
99:59:59.999 --> 99:59:59.999
[clapping]
99:59:59.999 --> 99:59:59.999
Even just a cracker.
99:59:59.999 --> 99:59:59.999
He tries to game us out of our metaphysical currencies.
99:59:59.999 --> 99:59:59.999
Our souls and so on.
99:59:59.999 --> 99:59:59.999
And now, of course, we’re all good atheists today
99:59:59.999 --> 99:59:59.999
and—at least in public, and science–
99:59:59.999 --> 99:59:59.999
and we don’t admit to this anymore and he can make do without this guy with root rights.
99:59:59.999 --> 99:59:59.999
And he can make do without the devil and so on.
99:59:59.999 --> 99:59:59.999
He can’t even say: “OK. Maybe there’s such a thing as a soul,
99:59:59.999 --> 99:59:59.999
but to say that this domain doesn’t exist anymore means you guys are all NPCs.
99:59:59.999 --> 99:59:59.999
You’re non-player characters.
99:59:59.999 --> 99:59:59.999
People are things.
99:59:59.999 --> 99:59:59.999
And it’s a very big insult to our culture,
99:59:59.999 --> 99:59:59.999
because it means that we have to give up something which,
99:59:59.999 --> 99:59:59.999
in our understanding of ourself is part of our essence.
99:59:59.999 --> 99:59:59.999
Also this mechanical perspective is kind of counter intuitive.
99:59:59.999 --> 99:59:59.999
I think Leibniz describes it very nicely when he says:
99:59:59.999 --> 99:59:59.999
Imagine that there is a machine.
99:59:59.999 --> 99:59:59.999
And this machine is able to think and perceive and feel and so on.
99:59:59.999 --> 99:59:59.999
And now you take this machine,
99:59:59.999 --> 99:59:59.999
this mechanical apparatus and blow it up make it very large, like a very big mill,
99:59:59.999 --> 99:59:59.999
with cogs and levers and so on and you go inside and see what happens.
99:59:59.999 --> 99:59:59.999
And what you are going to see is just parts pushing at each other.
99:59:59.999 --> 99:59:59.999
And what he meant by that is:
99:59:59.999 --> 99:59:59.999
it’s inconceivable that such a thing can produce a mind.
99:59:59.999 --> 99:59:59.999
Because if there are just parts and levers pushing at each other,
99:59:59.999 --> 99:59:59.999
how can this purely mechanical contraption be able to perceive and feel in any respect, in any way.
99:59:59.999 --> 99:59:59.999
So perception and what depends on it
99:59:59.999 --> 99:59:59.999
is in explicable in a mechanical way.
99:59:59.999 --> 99:59:59.999
This is what Leibniz meant.
99:59:59.999 --> 99:59:59.999
AI, the idea of treating the mind as a machine, based on physicalism for instance, is bound to fail according to Leibniz.
99:59:59.999 --> 99:59:59.999
Now as computer scientists have ideas about machines that can bring forth thoughts experiences and perception.
99:59:59.999 --> 99:59:59.999
And the first thing which comes to mind is probably the Turing machine.
99:59:59.999 --> 99:59:59.999
An idea of Turing in 1937 to formalize computation.
99:59:59.999 --> 99:59:59.999
At that time,
99:59:59.999 --> 99:59:59.999
Turing already realized that basically you can emulate computers with other computers.
99:59:59.999 --> 99:59:59.999
You know you can run a Commodore 64 in a Mac, and you can run this Mac in a PC,
99:59:59.999 --> 99:59:59.999
and none of these computers is going to be… is knowing that it’s going to be in another system.
99:59:59.999 --> 99:59:59.999
As long as the computational substrate in which it is run is sufficient.
99:59:59.999 --> 99:59:59.999
That is, it does provide computation.
99:59:59.999 --> 99:59:59.999
And Turing’s idea was: let’s define a minimal computational substrate.
99:59:59.999 --> 99:59:59.999
Let’s define the minimal recipe for something that is able to compute,
99:59:59.999 --> 99:59:59.999
and thereby understand computation.
99:59:59.999 --> 99:59:59.999
And the idea is that we take an infinite tape of symbols.
99:59:59.999 --> 99:59:59.999
And we have a read-write head.
99:59:59.999 --> 99:59:59.999
And this read-write head will write characters of a finite alphabet.
99:59:59.999 --> 99:59:59.999
And can again read them.
99:59:59.999 --> 99:59:59.999
And whenever it reads them based on a table that it has, a transition table
99:59:59.999 --> 99:59:59.999
it will erase the character, write a new one, and move either to the right, or the left and stop.
99:59:59.999 --> 99:59:59.999
Now imagine you have this machine.
99:59:59.999 --> 99:59:59.999
It has an initial setup. That is, there is a sequence of characters on the tape
99:59:59.999 --> 99:59:59.999
and then the thing goes to action.
99:59:59.999 --> 99:59:59.999
It will move right, left and so on and change the sequence of characters.
99:59:59.999 --> 99:59:59.999
And eventually, it’ll stop.
99:59:59.999 --> 99:59:59.999
And leave this tape with a certain sequence of characters,
99:59:59.999 --> 99:59:59.999
which is different from the one it began with probably.
99:59:59.999 --> 99:59:59.999
And Turing has shown that this thing is able to perform basically arbitrary computations.
99:59:59.999 --> 99:59:59.999
Now it’s very difficult to find the limits of that.
99:59:59.999 --> 99:59:59.999
And the idea of showing the limits of that would be to find classes of functions that can not be computed
99:59:59.999 --> 99:59:59.999
with this thing.
99:59:59.999 --> 99:59:59.999
OK. What you see here, is of course physical realization of that Turing machine.
99:59:59.999 --> 99:59:59.999
The Turing machine is a purely mathematical idea.
99:59:59.999 --> 99:59:59.999
And this is a very clever and beautiful illustration, I think.
99:59:59.999 --> 99:59:59.999
But this machine triggers basically the same criticism as the one that Leibniz had.
99:59:59.999 --> 99:59:59.999
John Searle said—
99:59:59.999 --> 99:59:59.999
you know, Searle is the one with the Chinese room. We’re not going to go into that—
99:59:59.999 --> 99:59:59.999
A Turing machine could be realized in many different mechanical ways.
99:59:59.999 --> 99:59:59.999
For instance, with levers and pulleys and so on.
99:59:59.999 --> 99:59:59.999
Or the water pipes.
99:59:59.999 --> 99:59:59.999
Or we could even come up with very clever arrangements just using cats, mice and cheese.
99:59:59.999 --> 99:59:59.999
So, it’s pretty ridiculous to think that such a contraption out of cats, mice and cheese,
99:59:59.999 --> 99:59:59.999
would thing, see, feel and so on.
99:59:59.999 --> 99:59:59.999
and then you could ask Searle:
99:59:59.999 --> 99:59:59.999
“Uh. You know. But how is it coming about then?”
99:59:59.999 --> 99:59:59.999
And he says: “So it’s intrinsic powers of biological neurons.”
99:59:59.999 --> 99:59:59.999
There’s nothing much more to say about that.
99:59:59.999 --> 99:59:59.999
Anyway.
99:59:59.999 --> 99:59:59.999
We have very crafty people here, this year.
99:59:59.999 --> 99:59:59.999
There was Seidenstraße.
99:59:59.999 --> 99:59:59.999
Maybe next year, we build a Turing machine from cats, mice and cheese.
99:59:59.999 --> 99:59:59.999
[laughter]
99:59:59.999 --> 99:59:59.999
How would you go about this.
99:59:59.999 --> 99:59:59.999
I don’t know how the arrangement of cat, mice, and cheese would look like to build flip-flops with it to store bits.
99:59:59.999 --> 99:59:59.999
But I am sure somebody of you will come up with a very clever solution.
99:59:59.999 --> 99:59:59.999
Searle I didn’t provide any.
99:59:59.999 --> 99:59:59.999
Let’s imagine… we will need a lot of redundancy, because these guys are a little bit erratic.
99:59:59.999 --> 99:59:59.999
Let’s say, we take three cat-mice-cheese units for each bit.
99:59:59.999 --> 99:59:59.999
So we have a little bit of redundancy.
99:59:59.999 --> 99:59:59.999
The human memory capacity is on the order of 10 to the power of 15 bits.
99:59:59.999 --> 99:59:59.999
Means.
99:59:59.999 --> 99:59:59.999
If we make do with 10 gram cheese per unit, it’s going to be 30 billion tons of cheese.
99:59:59.999 --> 99:59:59.999
So next year don’t bring bottles for the Seidenstraße, but bring some cheese.
99:59:59.999 --> 99:59:59.999
When we try to build this in the Congress Center,
99:59:59.999 --> 99:59:59.999
we might run out of space. So, if we just instead take all of Hamburg,
99:59:59.999 --> 99:59:59.999
and stack it with the necessary number of cat-mice-cheese units according to that rough estimate,
99:59:59.999 --> 99:59:59.999
you get to four kilometers high.
99:59:59.999 --> 99:59:59.999
Now imagine, we cover Hamburg in four kilometers of solid cat-mice-and-cheese flip-flops
99:59:59.999 --> 99:59:59.999
to my intuition this is super impressive.
99:59:59.999 --> 99:59:59.999
Maybe it thinks.
99:59:59.999 --> 99:59:59.999
[applause]
99:59:59.999 --> 99:59:59.999
So, of course it’s an intuition.
99:59:59.999 --> 99:59:59.999
And Searle has an intuition.
99:59:59.999 --> 99:59:59.999
And I don’t think that intuitions are worth much.
99:59:59.999 --> 99:59:59.999
This is the big problem of philosophy.
99:59:59.999 --> 99:59:59.999
You are very often working with intuitions, because the validity of your argument basically depends on what your audience thinks.
99:59:59.999 --> 99:59:59.999
In computer science, it’s different.
99:59:59.999 --> 99:59:59.999
It doesn’t really matter what your audience thinks. It matters, if it’s runs and it’s a very strange experience that you have as a student when you are at the same time taking classes in philosophy and in computer science and in your first semester.
99:59:59.999 --> 99:59:59.999
You’re going to point out in computer science that there is a mistake on the blackboard and everybody including the professor is super thankful.
99:59:59.999 --> 99:59:59.999
And you do the same thing in philosophy.
99:59:59.999 --> 99:59:59.999
It just doesn’t work this way.
99:59:59.999 --> 99:59:59.999
Anyway.
99:59:59.999 --> 99:59:59.999
The Turing machine is a good definition, but it’s a very bad metaphor,
99:59:59.999 --> 99:59:59.999
because it leaves people with this intuition of cogs, and wheels, and tape.
99:59:59.999 --> 99:59:59.999
It’s kind of linear, you know.
99:59:59.999 --> 99:59:59.999
There’s no parallel execution.
99:59:59.999 --> 99:59:59.999
And even though it’s infinitely faster infinitely larger and so on it’s very hard to imagine those things.
99:59:59.999 --> 99:59:59.999
But what you imagine is the tape.
99:59:59.999 --> 99:59:59.999
Maybe we want to have an alternative.
99:59:59.999 --> 99:59:59.999
And I think a very good alternative is for instance the lambda calculus.
99:59:59.999 --> 99:59:59.999
It’s computation without wheels.
99:59:59.999 --> 99:59:59.999
It was invented basically at the same time as the Turing machine.
99:59:59.999 --> 99:59:59.999
And philosophers and popular science magazines usually don’t use it for illustration of the idea of computation, because it has this scary Greek letter in it.
99:59:59.999 --> 99:59:59.999
Lambda.
99:59:59.999 --> 99:59:59.999
And calculus.
99:59:59.999 --> 99:59:59.999
And actually it’s an accident that it has the lambda in it.
99:59:59.999 --> 99:59:59.999
I think it should not be called lambda calculus.
99:59:59.999 --> 99:59:59.999
It’s super scary to people, which are not mathematicians.
99:59:59.999 --> 99:59:59.999
It would be called copy and paste thingi.
99:59:59.999 --> 99:59:59.999
[laughter]
99:59:59.999 --> 99:59:59.999
Because that’s all it does.
99:59:59.999 --> 99:59:59.999
It really only does copy and paste with very simple strings.
99:59:59.999 --> 99:59:59.999
And the strings that you want to paste into are marked with a little roof.
99:59:59.999 --> 99:59:59.999
And the original script by Alonzo Church.
99:59:59.999 --> 99:59:59.999
And in 1937 and 1936 typesetting was very difficult.
99:59:59.999 --> 99:59:59.999
So when he wrote this down with his typewriter, he made a little roof in front of the variable that he wanted to replace.
99:59:59.999 --> 99:59:59.999
And when this thing went into print, typesetters replaced this triangle by a lambda.
99:59:59.999 --> 99:59:59.999
There you go.
99:59:59.999 --> 99:59:59.999
Now we have the lambda calculus.
99:59:59.999 --> 99:59:59.999
But it basically means it is a little roof over the first letter.
99:59:59.999 --> 99:59:59.999
And the lambda calculus works like this.
99:59:59.999 --> 99:59:59.999
The first letter, the one that is going to be replaced.
99:59:59.999 --> 99:59:59.999
This is what we call the bound variable.
99:59:59.999 --> 99:59:59.999
This is followed by an expression.
99:59:59.999 --> 99:59:59.999
And then you have an argument, which is another expression.
99:59:59.999 --> 99:59:59.999
And what we basically do is, we take the bound variable, and all occurrences in the expression, and replace it by the arguments.
99:59:59.999 --> 99:59:59.999
So we cut the argument and we paste it in all instances of the variable, in this case the variable y.
99:59:59.999 --> 99:59:59.999
In here.
99:59:59.999 --> 99:59:59.999
And as a result you get this.
99:59:59.999 --> 99:59:59.999
So here we replace all the variables by the argument “ab”.
99:59:59.999 --> 99:59:59.999
Just another expression and this is the result.
99:59:59.999 --> 99:59:59.999
That’s all there is.
99:59:59.999 --> 99:59:59.999
And this can be nested.
99:59:59.999 --> 99:59:59.999
And then we add a little bit of syntactic sugar.
99:59:59.999 --> 99:59:59.999
We introduce symbols,
99:59:59.999 --> 99:59:59.999
so we can take arbitrary sequences of these characters and just express them with another variable.
99:59:59.999 --> 99:59:59.999
And then we have a programming language.
99:59:59.999 --> 99:59:59.999
And basically this is Lisp.
99:59:59.999 --> 99:59:59.999
So very close to Lisp.
99:59:59.999 --> 99:59:59.999
A funny thing is that for… the guy who came up with Lisp,
99:59:59.999 --> 99:59:59.999
McCarthy, he didn’t think that it would be a proper language.
99:59:59.999 --> 99:59:59.999
Because of the awkward notation.
99:59:59.999 --> 99:59:59.999
And he said, you cannot really use this for programming.
99:59:59.999 --> 99:59:59.999
But one of his doctorate students said: “Oh well. Let’s try.”
99:59:59.999 --> 99:59:59.999
And… it has kept on.
99:59:59.999 --> 99:59:59.999
Anyway.
99:59:59.999 --> 99:59:59.999
We can show that Turing Machines can compute the lambda calculus.
99:59:59.999 --> 99:59:59.999
And we can show that the lambda calculus can be used to compute the next state of the Turing machine.
99:59:59.999 --> 99:59:59.999
This means they have the same power.
99:59:59.999 --> 99:59:59.999
The set of computable functions in the lambda calculus is the same as the set of Turing computable functions.
99:59:59.999 --> 99:59:59.999
And, since then, we have found many other ways of defining computations.
99:59:59.999 --> 99:59:59.999
For instance the post machine, which is a variation of the Turing machine,
99:59:59.999 --> 99:59:59.999
or mathematical proofs.
99:59:59.999 --> 99:59:59.999
Everything that can be proven is computable.
99:59:59.999 --> 99:59:59.999
Or partial recursive functions.
99:59:59.999 --> 99:59:59.999
And we can show for all of them that all these approaches have the same power.
99:59:59.999 --> 99:59:59.999
And the idea that all the computational approaches have the same power,
99:59:59.999 --> 99:59:59.999
although all the other ones that you are able to find in the future too,
99:59:59.999 --> 99:59:59.999
is called the Church-Turing thesis.
99:59:59.999 --> 99:59:59.999
We don’t know about the future.
99:59:59.999 --> 99:59:59.999
So it’s not really… we can’t prove that.
99:59:59.999 --> 99:59:59.999
We don’t know, if somebody comes up with a new way of manipulating things, and producing regularity and information, and it can do more.
99:59:59.999 --> 99:59:59.999
But everything we’ve found so far, and probably everything that we’re going to find, has the same power.
99:59:59.999 --> 99:59:59.999
So this kind of defines our notion of computation.
99:59:59.999 --> 99:59:59.999
The whole thing also includes programming languages.
99:59:59.999 --> 99:59:59.999
You can use Python to produce to calculate a Turing machine and you can use a Turing machine to calculate Python.
99:59:59.999 --> 99:59:59.999
You can take arbitrary computers and let them run on the Turing machine.
99:59:59.999 --> 99:59:59.999
The graphics are going to be abysmal.
99:59:59.999 --> 99:59:59.999
But OK.
99:59:59.999 --> 99:59:59.999
And in some sense the brain is [a] Turing computational tool.
99:59:59.999 --> 99:59:59.999
If you look at the principles of neural information processing,
99:59:59.999 --> 99:59:59.999
you can take neurons and build computational models, for instance compartment models.
99:59:59.999 --> 99:59:59.999
Which are very very accurate and produce very strong semblances to the actual inputs and outputs of neurons and their state changes.
99:59:59.999 --> 99:59:59.999
They’re are computationally expensive, but it works.
99:59:59.999 --> 99:59:59.999
And we can simplify them into integrate-and-fire models, which are fancy oscillators.
99:59:59.999 --> 99:59:59.999
Or we could use very crude simplifications, like in most artificial neural networks.
99:59:59.999 --> 99:59:59.999
If you just do at some of the inputs to a neuron,
99:59:59.999 --> 99:59:59.999
and then apply some transition function,
99:59:59.999 --> 99:59:59.999
and transmit the results to other neurons.
99:59:59.999 --> 99:59:59.999
And we can show that with this crude model already,
99:59:59.999 --> 99:59:59.999
we can do many of the interesting feats that nervous systems can produce.
99:59:59.999 --> 99:59:59.999
Like associative learning, sensory motor loops, and many other fancy things.
99:59:59.999 --> 99:59:59.999
And, of course, it’s Turing complete.
99:59:59.999 --> 99:59:59.999
And this brings us to what we would call weak computationalism.
99:59:59.999 --> 99:59:59.999
That is the idea that minds are basically computer programs.
99:59:59.999 --> 99:59:59.999
They’re realizing in neural hard reconfigurations
99:59:59.999 --> 99:59:59.999
and in the individual states.
99:59:59.999 --> 99:59:59.999
And the mental content is represented in those programs.
99:59:59.999 --> 99:59:59.999
And perception is basically the process of encoding information
99:59:59.999 --> 99:59:59.999
given at our systemic boundaries to the environment
99:59:59.999 --> 99:59:59.999
into mental representations
99:59:59.999 --> 99:59:59.999
using this program.
99:59:59.999 --> 99:59:59.999
This means that all that is part of being a mind:
99:59:59.999 --> 99:59:59.999
thinking, and feeling, and dreaming, and being creative, and being afraid, and whatever.
99:59:59.999 --> 99:59:59.999
It’s all aspects of operations over mental content in such a computer program.
99:59:59.999 --> 99:59:59.999
This is the idea of weak computationalism.
99:59:59.999 --> 99:59:59.999
In fact you can go one step further to strong computationalism,
99:59:59.999 --> 99:59:59.999
because the universe doesn’t let us experience matter.
99:59:59.999 --> 99:59:59.999
The universe also doesn’t let us experience minds directly.
99:59:59.999 --> 99:59:59.999
What the universe somehow gives us is information.
99:59:59.999 --> 99:59:59.999
Information is something very simple.
99:59:59.999 --> 99:59:59.999
We can define it mathematically and what it means is something like “discernible difference”.
99:59:59.999 --> 99:59:59.999
You can measure it in yes-no-decisions, in bits.
99:59:59.999 --> 99:59:59.999
And there is….
99:59:59.999 --> 99:59:59.999
According to the strong computationalism,
99:59:59.999 --> 99:59:59.999
the universe is basically a pattern generator,
99:59:59.999 --> 99:59:59.999
which gives us information.
99:59:59.999 --> 99:59:59.999
And all the apparent regularity
99:59:59.999 --> 99:59:59.999
that the universe seems to produce,
99:59:59.999 --> 99:59:59.999
which means, we see time and space,
99:59:59.999 --> 99:59:59.999
and things that we can conceptualize into objects and people,
99:59:59.999 --> 99:59:59.999
and whatever,
99:59:59.999 --> 99:59:59.999
can be explained by the fact that the universe seems to be able to compute.
99:59:59.999 --> 99:59:59.999
That is, to put use regularities in information.
99:59:59.999 --> 99:59:59.999
And this means that there is no conceptual difference between reality and the computer program.
99:59:59.999 --> 99:59:59.999
So we get a new kind of monism.
99:59:59.999 --> 99:59:59.999
Not idealism, which takes minds to be primary,
99:59:59.999 --> 99:59:59.999
or materialism which takes physics to be primary,
99:59:59.999 --> 99:59:59.999
but rather computationalism, which means that information and computation are primary.
99:59:59.999 --> 99:59:59.999
Mind and matter are constructions that we get from that.
99:59:59.999 --> 99:59:59.999
A lot of people don’t like that idea.
99:59:59.999 --> 99:59:59.999
Roger Penrose, who’s a physicist,
99:59:59.999 --> 99:59:59.999
says that the brain uses quantum processes to produce consciousness.
99:59:59.999 --> 99:59:59.999
So minds must be more than computers.
99:59:59.999 --> 99:59:59.999
Why is that so?
99:59:59.999 --> 99:59:59.999
The quality of understanding and feeling possessed by human beings, is something that cannot be simulated computationally.
99:59:59.999 --> 99:59:59.999
Ok.
99:59:59.999 --> 99:59:59.999
But how can quantum mechanics do it?
99:59:59.999 --> 99:59:59.999
Because, you know, quantum processes are completely computational too!
99:59:59.999 --> 99:59:59.999
It’s just very expensive to simulate them on non-quantum computers.
99:59:59.999 --> 99:59:59.999
But it’s possible.
99:59:59.999 --> 99:59:59.999
So, it’s not that quantum computing enables a completely new kind of effectively possible algorithm.
99:59:59.999 --> 99:59:59.999
It’s just slightly different efficiently possible algorithms.
99:59:59.999 --> 99:59:59.999
And Penrose cannot explain how those would bring forth
99:59:59.999 --> 99:59:59.999
perception and imagination and consciousness.
99:59:59.999 --> 99:59:59.999
I think what he basically does here is that he perceives kind of mechanics as mysterious
99:59:59.999 --> 99:59:59.999
and perceives consciousness as mysterious and tries to shroud one mystery in another.
99:59:59.999 --> 99:59:59.999
[applause]
99:59:59.999 --> 99:59:59.999
So I don’t think that minds are more than Turing machines.
99:59:59.999 --> 99:59:59.999
It’s actually much more troubling: minds are fundamentally less than Turing machines!
99:59:59.999 --> 99:59:59.999
All real computers are constrained in some way.
99:59:59.999 --> 99:59:59.999
That is they cannot compute every conceivable computable function.
99:59:59.999 --> 99:59:59.999
They can only compute functions that fit into the memory and so on then can be computed in the available time.
99:59:59.999 --> 99:59:59.999
So the Turing machine, if you want to build it physically,
99:59:59.999 --> 99:59:59.999
will have a finite tape and it will have finite steps it can calculate in a given amount of time.
99:59:59.999 --> 99:59:59.999
And the lambda calculus will have a finite length to the strings that you can actually cut and replace.
99:59:59.999 --> 99:59:59.999
And a finite number of replacement operations that you can do
99:59:59.999 --> 99:59:59.999
in your given amount of time.
99:59:59.999 --> 99:59:59.999
And the thing is, there is no set of numbers m and n for…
99:59:59.999 --> 99:59:59.999
for the tape lengths and the times you have four operations on [the] Turing machine.
99:59:59.999 --> 99:59:59.999
And the same m and n or similar m and n
99:59:59.999 --> 99:59:59.999
for the lambda calculus at least with the same set of constraints.
99:59:59.999 --> 99:59:59.999
That is lambda calculus
99:59:59.999 --> 99:59:59.999
is going to be able to calculate some functions
99:59:59.999 --> 99:59:59.999
that are not possible on the Turing machine and vice versa,
99:59:59.999 --> 99:59:59.999
if you have a constrained system.
99:59:59.999 --> 99:59:59.999
And of course it’s even worse for neurons.
99:59:59.999 --> 99:59:59.999
If you have a finite number of neurons and to find a number of state changes,
99:59:59.999 --> 99:59:59.999
this… does not translate directly into a constrained von-Neumann-computer
99:59:59.999 --> 99:59:59.999
or a constrained lambda calculus.
99:59:59.999 --> 99:59:59.999
And there’s this big difference between, of course, effectively computable functions,
99:59:59.999 --> 99:59:59.999
those that are in principle computable,
99:59:59.999 --> 99:59:59.999
and those that we can compute efficiently.
99:59:59.999 --> 99:59:59.999
There are things that computers cannot solve.
99:59:59.999 --> 99:59:59.999
Some problems that are unsolvable in principle.
99:59:59.999 --> 99:59:59.999
For instance the question whether a Turing machine ever stops
99:59:59.999 --> 99:59:59.999
for an arbitrary program.
99:59:59.999 --> 99:59:59.999
And some problems are unsolvable in practice.
99:59:59.999 --> 99:59:59.999
Because it’s very, very hard to do so for a deterministic Turing machine.
99:59:59.999 --> 99:59:59.999
And the class of NP-hard problems is a very strong candidate for that.
99:59:59.999 --> 99:59:59.999
Non-polinominal problems.
99:59:59.999 --> 99:59:59.999
In these problems is for instance the idea
99:59:59.999 --> 99:59:59.999
of finding the key for an encrypted text.
99:59:59.999 --> 99:59:59.999
If key is very long and you are not the NSA and have a backdoor.
99:59:59.999 --> 99:59:59.999
And then there are non-decidable problems.
99:59:59.999 --> 99:59:59.999
Problems where we cannot define…
99:59:59.999 --> 99:59:59.999
find out, in the formal system, the answer is yes or no.
99:59:59.999 --> 99:59:59.999
Whether it’s true or false.
99:59:59.999 --> 99:59:59.999
And some philosophers have argued that humans can always do this so they are more powerful than computers.
99:59:59.999 --> 99:59:59.999
Because show, prove formally, that computers cannot do this.
99:59:59.999 --> 99:59:59.999
Gödel has done this.
99:59:59.999 --> 99:59:59.999
But… hm…
99:59:59.999 --> 99:59:59.999
Here’s some test question:
99:59:59.999 --> 99:59:59.999
can you solve undecidable problems.
99:59:59.999 --> 99:59:59.999
If you choose one of the following answers randomly,
99:59:59.999 --> 99:59:59.999
what’s the probability that the answer is correct?
99:59:59.999 --> 99:59:59.999
I’ll tell you.
99:59:59.999 --> 99:59:59.999
Computers are not going to find out.
99:59:59.999 --> 99:59:59.999
And… me neither.
99:59:59.999 --> 99:59:59.999
OK.
99:59:59.999 --> 99:59:59.999
How difficult is AI?
99:59:59.999 --> 99:59:59.999
It’s a very difficult question.
99:59:59.999 --> 99:59:59.999
We don’t know.
99:59:59.999 --> 99:59:59.999
We do have some numbers, which could tell us that it’s not impossible.
99:59:59.999 --> 99:59:59.999
As we have these roughly 100 billion neurons—
99:59:59.999 --> 99:59:59.999
the ballpark figure—
99:59:59.999 --> 99:59:59.999
and the cells in the cortex are organized into circuits of a few thousands to ten-thousands of neurons,
99:59:59.999 --> 99:59:59.999
which you call cortical columns.
99:59:59.999 --> 99:59:59.999
And these cortical columns have… are pretty similar among each other,
99:59:59.999 --> 99:59:59.999
and have higher interconnectivity, and some lower connectivity among each other,
99:59:59.999 --> 99:59:59.999
and even lower long range connectivity.
99:59:59.999 --> 99:59:59.999
And the brain has a very distinct architecture.
99:59:59.999 --> 99:59:59.999
And a very distinct structure of a certain nuclei and structures that have very different functional purposes.
99:59:59.999 --> 99:59:59.999
And the layout of these…
99:59:59.999 --> 99:59:59.999
both the individual neurons, neuron types,
99:59:59.999 --> 99:59:59.999
the more than 130 known neurotransmitters, of which we do not completely understand all, most of them,
99:59:59.999 --> 99:59:59.999
this is all defined in our genome of course.
99:59:59.999 --> 99:59:59.999
And the genome is not very long.
99:59:59.999 --> 99:59:59.999
It’s something like… it think the Human Genome Project amounted to a CD-ROM.
99:59:59.999 --> 99:59:59.999
775 megabytes.
99:59:59.999 --> 99:59:59.999
So actually, it’s….
99:59:59.999 --> 99:59:59.999
The computational complexity of defining a complete human being,
99:59:59.999 --> 99:59:59.999
if you have physics chemistry already given
99:59:59.999 --> 99:59:59.999
to enable protein synthesis and so on—
99:59:59.999 --> 99:59:59.999
gravity and temperature ranges—
99:59:59.999 --> 99:59:59.999
is less than Microsoft Windows.
99:59:59.999 --> 99:59:59.999
And it’s the upper bound, because only a very small fraction of that
99:59:59.999 --> 99:59:59.999
is going to code for our nervous system.
99:59:59.999 --> 99:59:59.999
But it doesn’t mean it’s easy to reverse engineer the whole thing.
99:59:59.999 --> 99:59:59.999
It just means it’s not hopeless.
99:59:59.999 --> 99:59:59.999
Complexity that you would be looking at.
99:59:59.999 --> 99:59:59.999
But the estimate of the real difficulty, in my perspective, is impossible.
99:59:59.999 --> 99:59:59.999
Because I’m not just a philosopher or a dreamer or a science fiction author, but I’m a software developer.
99:59:59.999 --> 99:59:59.999
And as a software developer I know it’s impossible to give an estimate on when you’re done, when you don’t have the full specification.
99:59:59.999 --> 99:59:59.999
And we don’t have a full specification yet.
99:59:59.999 --> 99:59:59.999
So you all know this shortest computer science joke:
99:59:59.999 --> 99:59:59.999
“It’s almost done.”
99:59:59.999 --> 99:59:59.999
You do the first 98 %.
99:59:59.999 --> 99:59:59.999
Now we can do the second 98 %.
99:59:59.999 --> 99:59:59.999
We never know when it’s done,
99:59:59.999 --> 99:59:59.999
if we haven’t solved and specified all the problems.
99:59:59.999 --> 99:59:59.999
If you don’t know how it’s to be done.
99:59:59.999 --> 99:59:59.999
And even if you have [a] rough direction, and I think we do,
99:59:59.999 --> 99:59:59.999
we don’t know how long it’ll take until we have worked out the details.
99:59:59.999 --> 99:59:59.999
And some part of that big question, how long it takes until it’ll be done,
99:59:59.999 --> 99:59:59.999
is the question whether we need to make small incremental progress
99:59:59.999 --> 99:59:59.999
versus whether we need one big idea,
99:59:59.999 --> 99:59:59.999
which kind of solves it all.
99:59:59.999 --> 99:59:59.999
AI has a pretty long story.
99:59:59.999 --> 99:59:59.999
It starts out with logic and automata.
99:59:59.999 --> 99:59:59.999
And this idea of computability that I just sketched out.
99:59:59.999 --> 99:59:59.999
Then with this idea of machines that implement computability.
99:59:59.999 --> 99:59:59.999
And came towards Babage and Zuse and von Neumann and so on.
99:59:59.999 --> 99:59:59.999
Then we had information theory by Claude Shannon.
99:59:59.999 --> 99:59:59.999
He captured the idea of what information is
99:59:59.999 --> 99:59:59.999
and how entropy can be calculated for information and so on.
99:59:59.999 --> 99:59:59.999
And we had this beautiful idea of describing the world as systems.
99:59:59.999 --> 99:59:59.999
And systems are made up of entities and relations between them.
99:59:59.999 --> 99:59:59.999
And along these relations there we have feedback.
99:59:59.999 --> 99:59:59.999
And dynamical systems emerge.
99:59:59.999 --> 99:59:59.999
This was a very beautiful idea, was cybernetics.
99:59:59.999 --> 99:59:59.999
Unfortunately hass been killed by
99:59:59.999 --> 99:59:59.999
second-order Cybernetics.
99:59:59.999 --> 99:59:59.999
By this Maturana stuff and so on.
99:59:59.999 --> 99:59:59.999
And turned into a humanity [one of the humanities] and died.
99:59:59.999 --> 99:59:59.999
But the idea stuck around and most of them went into artificial intelligence.
99:59:59.999 --> 99:59:59.999
And then we had this idea of symbol systems.
99:59:59.999 --> 99:59:59.999
That is how we can do grammatical language.
99:59:59.999 --> 99:59:59.999
Process that.
99:59:59.999 --> 99:59:59.999
We can do planning and so on.
99:59:59.999 --> 99:59:59.999
Abstract reasoning in automatic systems.
99:59:59.999 --> 99:59:59.999
Then the idea how of we can abstract neural networks in distributed systems.
99:59:59.999 --> 99:59:59.999
With McClelland and Pitts and so on.
99:59:59.999 --> 99:59:59.999
Parallel distributed processing.
99:59:59.999 --> 99:59:59.999
And then we had a movement of autonomous agents,
99:59:59.999 --> 99:59:59.999
which look at self-directed, goal directed systems.
99:59:59.999 --> 99:59:59.999
And the whole story somehow started in 1950 I think,
99:59:59.999 --> 99:59:59.999
in its best possible way.
99:59:59.999 --> 99:59:59.999
When Alan Turing wrote his paper
99:59:59.999 --> 99:59:59.999
“Computing Machinery and Intelligence”
99:59:59.999 --> 99:59:59.999
and those of you who haven’t read it should do so.
99:59:59.999 --> 99:59:59.999
It’s a very, very easy read.
99:59:59.999 --> 99:59:59.999
It’s fascinating.
99:59:59.999 --> 99:59:59.999
He has already already most of the important questions of AI.
99:59:59.999 --> 99:59:59.999
Most of the important criticisms.
99:59:59.999 --> 99:59:59.999
Most of the important answers to the most important criticisms.
99:59:59.999 --> 99:59:59.999
And it’s also the paper, where he describes the Turing test.
99:59:59.999 --> 99:59:59.999
And basically sketches the idea that
99:59:59.999 --> 99:59:59.999
in a way to determine whether somebody is intelligent is
99:59:59.999 --> 99:59:59.999
to judge the ability of that one—
99:59:59.999 --> 99:59:59.999
that person or that system—
99:59:59.999 --> 99:59:59.999
to engage in meaningful discourse.
99:59:59.999 --> 99:59:59.999
Which includes creativity, and empathy maybe, and logic, and language,
99:59:59.999 --> 99:59:59.999
and anticipation, memory retrieval, and so on.
99:59:59.999 --> 99:59:59.999
Story comprehension.
99:59:59.999 --> 99:59:59.999
And the idea of AI then
99:59:59.999 --> 99:59:59.999
coalesce in the group of cyberneticians and computer scientists and so on,
99:59:59.999 --> 99:59:59.999
which got together in the Dartmouth conference.
99:59:59.999 --> 99:59:59.999
It was in 1956.
99:59:59.999 --> 99:59:59.999
And there Marvin Minsky coined the name “artificial intelligence
99:59:59.999 --> 99:59:59.999
for the project of using computer science to understand the mind.
99:59:59.999 --> 99:59:59.999
John McCarthy was the guy who came up with Lisp, among other things.
99:59:59.999 --> 99:59:59.999
Nathan Rochester did pattern recognition
99:59:59.999 --> 99:59:59.999
and he’s, I think, more famous for
99:59:59.999 --> 99:59:59.999
writing the first assembly programming language.
99:59:59.999 --> 99:59:59.999
Claude Shannon was this information theory guy.
99:59:59.999 --> 99:59:59.999
But they also got psychologists there
99:59:59.999 --> 99:59:59.999
and sociologists and people from many different fields.
99:59:59.999 --> 99:59:59.999
It was very highly interdisciplinary.
99:59:59.999 --> 99:59:59.999
And they already had the funding and it was a very good time.
99:59:59.999 --> 99:59:59.999
And in this good time they ripped a lot of low hanging fruit very quickly.
99:59:59.999 --> 99:59:59.999
Which gave them the idea that AI is almost done very soon.
99:59:59.999 --> 99:59:59.999
In 1969 Minsky and Papert wrote a small booklet against the idea of using your neural networks.
99:59:59.999 --> 99:59:59.999
And they won.
99:59:59.999 --> 99:59:59.999
Their argument won.
99:59:59.999 --> 99:59:59.999
But, even more fortunately it was wrong.
99:59:59.999 --> 99:59:59.999
So for more than a decade, there was practically no more funding for neural networks,
99:59:59.999 --> 99:59:59.999
which was bad so most people did logic based systems, which have some limitations.
99:59:59.999 --> 99:59:59.999
And in the meantime people did expert systems.
99:59:59.999 --> 99:59:59.999
The idea to describe the world
99:59:59.999 --> 99:59:59.999
as basically logical expressions.
99:59:59.999 --> 99:59:59.999
This turned out to be brittle, and difficult, and had diminishing returns.
99:59:59.999 --> 99:59:59.999
And at some point it didn’t work anymore.
99:59:59.999 --> 99:59:59.999
And many of the people which tried it,
99:59:59.999 --> 99:59:59.999
became very disenchanted and then threw out lots of baby with the bathwater.
99:59:59.999 --> 99:59:59.999
And only did robotics in the future or something completely different.
99:59:59.999 --> 99:59:59.999
Instead of going back to the idea of looking at mental representations.
99:59:59.999 --> 99:59:59.999
How the mind works.
99:59:59.999 --> 99:59:59.999
And at the moment is kind of a sad state.
99:59:59.999 --> 99:59:59.999
Most of it is applications.
99:59:59.999 --> 99:59:59.999
That is, for instance, robotics
99:59:59.999 --> 99:59:59.999
or statistical methods to do better machine learning and so on.
99:59:59.999 --> 99:59:59.999
And I don’t say it’s invalid to do this.
99:59:59.999 --> 99:59:59.999
It’s intellectually challenging.
99:59:59.999 --> 99:59:59.999
It’s tremendously useful.
99:59:59.999 --> 99:59:59.999
It’s very successful and productive and so on.
99:59:59.999 --> 99:59:59.999
It’s just a very different question from how to understand the mind.
99:59:59.999 --> 99:59:59.999
If you want to go to the moon you have to shoot for the moon.
99:59:59.999 --> 99:59:59.999
So there is this movement still existing in AI,
99:59:59.999 --> 99:59:59.999
and becoming stronger these days.
99:59:59.999 --> 99:59:59.999
It’s called cognitive systems.
99:59:59.999 --> 99:59:59.999
And the idea of cognitive systems has many names
99:59:59.999 --> 99:59:59.999
like “artificial general intelligence” or “biologically inspired cognitive architectures”.
99:59:59.999 --> 99:59:59.999
It’s to use information processing as the dominant paradigm to understand the mind.
99:59:59.999 --> 99:59:59.999
And the tools that we need to do that is,
99:59:59.999 --> 99:59:59.999
we have to build whole architectures that we can test.
99:59:59.999 --> 99:59:59.999
Not just individual modules.
99:59:59.999 --> 99:59:59.999
You have to have universal representations,
99:59:59.999 --> 99:59:59.999
which means these representation have to be both distributed—
99:59:59.999 --> 99:59:59.999
associative and so on—
99:59:59.999 --> 99:59:59.999
and symbolic.
99:59:59.999 --> 99:59:59.999
We need to be able to do both those things with it.
99:59:59.999 --> 99:59:59.999
So we need to be able to do language and planning, and we need to do sensorimotor coupling, and associative thinking in superposition of
99:59:59.999 --> 99:59:59.999
representations and ambiguity and so on.
99:59:59.999 --> 99:59:59.999
And
99:59:59.999 --> 99:59:59.999
operations over those presentation.
99:59:59.999 --> 99:59:59.999
Some kind of
99:59:59.999 --> 99:59:59.999
semi-universal problem solving.
99:59:59.999 --> 99:59:59.999
It’s probably semi-universal, because they seem to be problems that humans are very bad at solving.
99:59:59.999 --> 99:59:59.999
Our minds are not completely universal.
99:59:59.999 --> 99:59:59.999
And we need some kind of universal motivation. That is something that directs the system to do all the interesting things that you want it to do.
99:59:59.999 --> 99:59:59.999
Like engage in social interaction or in mathematics or creativity.
99:59:59.999 --> 99:59:59.999
And maybe we want to understand emotion, and affect, and phenomenal experience, and so on.
99:59:59.999 --> 99:59:59.999
So:
99:59:59.999 --> 99:59:59.999
we want to understand universal representations.
99:59:59.999 --> 99:59:59.999
We want to have a set of operations over those representations that give us neural learning, and category formation,
99:59:59.999 --> 99:59:59.999
and planning, and reflection, and memory consolidation, and resource allocation,
99:59:59.999 --> 99:59:59.999
and language, and all those interesting things.
99:59:59.999 --> 99:59:59.999
We also want to have perceptual grounding—
99:59:59.999 --> 99:59:59.999
that is the representations would be saved—shaped in such a way that they can be mapped to perceptual input—
99:59:59.999 --> 99:59:59.999
and vice versa.
99:59:59.999 --> 99:59:59.999
And…
99:59:59.999 --> 99:59:59.999
they should also be able to be translated into motor programs to perform actions.
99:59:59.999 --> 99:59:59.999
And maybe we also want to have some feedback between the actions and the perceptions, and is feedback usually has a name: it’s called an environment.
99:59:59.999 --> 99:59:59.999
OK.
99:59:59.999 --> 99:59:59.999
And these medical representations, they are not just a big lump of things but they have some structure.
99:59:59.999 --> 99:59:59.999
One part will be inevitably the model of the current situation…
99:59:59.999 --> 99:59:59.999
… that we are in.
99:59:59.999 --> 99:59:59.999
And this situation model…
99:59:59.999 --> 99:59:59.999
is the present.
99:59:59.999 --> 99:59:59.999
But if you also want to memorize past situations.
99:59:59.999 --> 99:59:59.999
To have a protocol a memory of the past.
99:59:59.999 --> 99:59:59.999
And this protocol memory, as a part, will contain things that are always with me.
99:59:59.999 --> 99:59:59.999
This is my self-model.
99:59:59.999 --> 99:59:59.999
Those properties that are constantly available to me.
99:59:59.999 --> 99:59:59.999
That I can ascribe to myself.
99:59:59.999 --> 99:59:59.999
And the other things, which are constantly changing, which I usually conceptualize as my environment.
99:59:59.999 --> 99:59:59.999
An important part of that is declarative memory.
99:59:59.999 --> 99:59:59.999
For instance abstractions into objects, things, people, and so on,
99:59:59.999 --> 99:59:59.999
and procedural memory: abstraction into sequences of events.
99:59:59.999 --> 99:59:59.999
And we can use the declarative memory and the procedural memory to erect a frame.
99:59:59.999 --> 99:59:59.999
The frame gives me a context to interpret the current situation.
99:59:59.999 --> 99:59:59.999
For instance right now I’m in a frame of giving a talk.
99:59:59.999 --> 99:59:59.999
If…
99:59:59.999 --> 99:59:59.999
… I would take a…
99:59:59.999 --> 99:59:59.999
two year old kid, then this kid would interpret the situation very differently than me.
99:59:59.999 --> 99:59:59.999
And would probably be confused by the situation or explored it in more creative ways than I would come up with.
99:59:59.999 --> 99:59:59.999
Because I’m constrained by the frame which gives me the context
99:59:59.999 --> 99:59:59.999
and tells me what you were expect me to do in this situation.
99:59:59.999 --> 99:59:59.999
What I am expected to do and so on.
99:59:59.999 --> 99:59:59.999
This frame extends in the future.
99:59:59.999 --> 99:59:59.999
I have some kind of expectation horizon.
99:59:59.999 --> 99:59:59.999
I know that my talk is going to be over in about 15 minutes.
99:59:59.999 --> 99:59:59.999
Also I’ve plans.
99:59:59.999 --> 99:59:59.999
I have things I want to tell you and so on.
99:59:59.999 --> 99:59:59.999
And it might go wrong but I’ll try.
99:59:59.999 --> 99:59:59.999
And if I generalize this, I find that I have the world model,
99:59:59.999 --> 99:59:59.999
I have long term memory, and have some kind of mental stage.
99:59:59.999 --> 99:59:59.999
This mental stage has counter-factual stuff.
99:59:59.999 --> 99:59:59.999
Stuff that is not…
99:59:59.999 --> 99:59:59.999
… real.
99:59:59.999 --> 99:59:59.999
That I can play around with.
99:59:59.999 --> 99:59:59.999
Ok. Then I need some kind of action selection that mediates between perception and action,
99:59:59.999 --> 99:59:59.999
and some mechanism that controls the action selection
99:59:59.999 --> 99:59:59.999
that is a motivational system,
99:59:59.999 --> 99:59:59.999
which selects motives based on demands of the system.
99:59:59.999 --> 99:59:59.999
And the demands of the system should create goals.
99:59:59.999 --> 99:59:59.999
We are not born with our goals.
99:59:59.999 --> 99:59:59.999
Obviously I don’t think that I was born with the goal of standing here and giving this talk to you.
99:59:59.999 --> 99:59:59.999
There must be some demand in the system, which makes… enables me to have a biography, that …
99:59:59.999 --> 99:59:59.999
… makes this a big goal of mine to give this talk to you and engage as many of you as possible into the project of AI.
99:59:59.999 --> 99:59:59.999
And so lets come up with a set of demands that can produce such goals universally.
99:59:59.999 --> 99:59:59.999
I think some of these demands will be physiological, like food, water, energy, physical integrity, rest, and so on.
99:59:59.999 --> 99:59:59.999
Hot and cold with right range.
99:59:59.999 --> 99:59:59.999
Then we have social demands.
99:59:59.999 --> 99:59:59.999
At least most of us do.
99:59:59.999 --> 99:59:59.999
Sociopaths probably don’t.
99:59:59.999 --> 99:59:59.999
These social demands do structure our…
99:59:59.999 --> 99:59:59.999
… social interaction.
99:59:59.999 --> 99:59:59.999
They…. For instance a demand for affiliation.
99:59:59.999 --> 99:59:59.999
That we get signals from others, that we are ok parts of society, of our environment.
99:59:59.999 --> 99:59:59.999
We also have internalised social demands,
99:59:59.999 --> 99:59:59.999
which we usually called honor or something.
99:59:59.999 --> 99:59:59.999
This is conformance to internalized norms.
99:59:59.999 --> 99:59:59.999
It means,
99:59:59.999 --> 99:59:59.999
that we do to conform to social norms, even when nobody is looking.
99:59:59.999 --> 99:59:59.999
And then we have cognitive demands.
99:59:59.999 --> 99:59:59.999
And these cognitive demands, is for instance competence acquisition.
99:59:59.999 --> 99:59:59.999
We want learn.
99:59:59.999 --> 99:59:59.999
We want to get new skills.
99:59:59.999 --> 99:59:59.999
We want to become more powerful in many many dimensions and ways.
99:59:59.999 --> 99:59:59.999
It’s good to learn a musical instrument, because you get more competent.
99:59:59.999 --> 99:59:59.999
It creates a reward signal, a pleasure signal, if you do that.
99:59:59.999 --> 99:59:59.999
Also we want to reduce uncertainty.
99:59:59.999 --> 99:59:59.999
Mathematicians are those people [that] have learned that they can reduce uncertainty in mathematics.
99:59:59.999 --> 99:59:59.999
This creates pleasure for them, and then they find uncertainty in mathematics.
99:59:59.999 --> 99:59:59.999
And this creates more pleasure.
99:59:59.999 --> 99:59:59.999
So for mathematicians, mathematics is an unending source of pleasure.
99:59:59.999 --> 99:59:59.999
Now unfortunately, if you are in Germany right now studying mathematics
99:59:59.999 --> 99:59:59.999
and you find out that you are not very good at doing mathematics, what do you do?
99:59:59.999 --> 99:59:59.999
You become a teacher.
99:59:59.999 --> 99:59:59.999
And this is a very unfortunate situation for everybody involved.
99:59:59.999 --> 99:59:59.999
And, it means, that you have people, [that] associate mathematics with…
99:59:59.999 --> 99:59:59.999
uncertainty,
99:59:59.999 --> 99:59:59.999
that has to be curbed and to be avoided.
99:59:59.999 --> 99:59:59.999
And these people are put in front of kids and infuse them with this dread of uncertainty in mathematics.
99:59:59.999 --> 99:59:59.999
And most people in our culture are dreading mathematics, because for them it’s just anticipation of uncertainty.
99:59:59.999 --> 99:59:59.999
Which is a very bad things so people avoid it.
99:59:59.999 --> 99:59:59.999
OK.
99:59:59.999 --> 99:59:59.999
And then you have aesthetic demands.
99:59:59.999 --> 99:59:59.999
There are stimulus oriented aesthetics.
99:59:59.999 --> 99:59:59.999
Nature has had to pull some very heavy strings and levers to make us interested in strange things…
99:59:59.999 --> 99:59:59.999
[such] as certain human body schemas and…
99:59:59.999 --> 99:59:59.999
certain types of landscapes, and audio schemas, and so on.
99:59:59.999 --> 99:59:59.999
So there are some stimuli that are inherently pleasurable to us—pleasant to us.
99:59:59.999 --> 99:59:59.999
And of course this varies with every individual, because the wiring is very different, and that adaptivity in our biography is very different.
99:59:59.999 --> 99:59:59.999
And then there’s abstract aesthetics.
99:59:59.999 --> 99:59:59.999
And I think abstract aesthetics relates to finding better representations.
99:59:59.999 --> 99:59:59.999
It relates to finding structure.
99:59:59.999 --> 99:59:59.999
OK. And then we want to look at things like emotional modulation and affect.
99:59:59.999 --> 99:59:59.999
And this was one of the first things that actually got me into AI.
99:59:59.999 --> 99:59:59.999
That was the question:
99:59:59.999 --> 99:59:59.999
“How is it possible, that a system can feel something?”
99:59:59.999 --> 99:59:59.999
Because, if I have a variable in me with just fear or pain,
99:59:59.999 --> 99:59:59.999
does not equate a feeling.
99:59:59.999 --> 99:59:59.999
It’s very far… uhm…
99:59:59.999 --> 99:59:59.999
… different from that.
99:59:59.999 --> 99:59:59.999
And the answer that I’ve found so far it is,
99:59:59.999 --> 99:59:59.999
that feeling, or affect, is a configuration of the system.
99:59:59.999 --> 99:59:59.999
It’s not a parameter in the system,
99:59:59.999 --> 99:59:59.999
but we have several dimensions, like a state of arousal that we’re currently, in the level of stubbornness that we have, the selection threshold,
99:59:59.999 --> 99:59:59.999
the direction of attention, outwards or inwards,
99:59:59.999 --> 99:59:59.999
the resolution level that we have, [with] which we look at our representations, and so on.
99:59:59.999 --> 99:59:59.999
And these together create a certain way in every given situation of how our cognition is modulated.
99:59:59.999 --> 99:59:59.999
We are living in a very different
99:59:59.999 --> 99:59:59.999
and dynamic environment from time to time.
99:59:59.999 --> 99:59:59.999
When you go outside we have very different demands on our cognition.
99:59:59.999 --> 99:59:59.999
Maybe you need to react to traffic and so on.
99:59:59.999 --> 99:59:59.999
Maybe we need to interact with other people.
99:59:59.999 --> 99:59:59.999
Maybe we are in stressful situations.
99:59:59.999 --> 99:59:59.999
Maybe you are in relaxed situations.
99:59:59.999 --> 99:59:59.999
So we need to modulate our cognition accordingly.
99:59:59.999 --> 99:59:59.999
And this modulation means, that we do perceive the world differently.
99:59:59.999 --> 99:59:59.999
Our cognition works differently.
99:59:59.999 --> 99:59:59.999
And we conceptualize ourselves, and experience ourselves, differently.
99:59:59.999 --> 99:59:59.999
And I think this is what it means to feel something:
99:59:59.999 --> 99:59:59.999
this difference in the configuration.
99:59:59.999 --> 99:59:59.999
So. The affect can be seen as a configuration of a cognitive system.
99:59:59.999 --> 99:59:59.999
And the modulators of the cognition are things like arousal, and selection special, and
99:59:59.999 --> 99:59:59.999
background checks level, and resolution level, and so on.
99:59:59.999 --> 99:59:59.999
Our current estimates of competence and certainty in the given situation,
99:59:59.999 --> 99:59:59.999
and the pleasure and distress signals that you get from the frustration of our demands,
99:59:59.999 --> 99:59:59.999
or satisfaction of our demands which are reinforcements for learning and structuring our behavior.
99:59:59.999 --> 99:59:59.999
So the affective state, the emotional state that we are in, is emergent over those modulators.
99:59:59.999 --> 99:59:59.999
And higher level emotions, things like jealousy or pride and so on,
99:59:59.999 --> 99:59:59.999
we get them by directing those effects upon motivational content.
99:59:59.999 --> 99:59:59.999
And this gives us a very simple architecture.
99:59:59.999 --> 99:59:59.999
It’s a very rough sketch for an architecture.
99:59:59.999 --> 99:59:59.999
And I think,
99:59:59.999 --> 99:59:59.999
of course,
99:59:59.999 --> 99:59:59.999
this doesn’t specify all the details.
99:59:59.999 --> 99:59:59.999
I have specified some more of the details in a book, that I want to shamelessly plug here:
99:59:59.999 --> 99:59:59.999
it’s called “Principles of Synthetic Intelligence”.
99:59:59.999 --> 99:59:59.999
You can get it from Amazon or maybe from your library.
99:59:59.999 --> 99:59:59.999
And this describes basically this architecture and some of the demands
99:59:59.999 --> 99:59:59.999
for a very general framework of artificial intelligence in which to work with it.
99:59:59.999 --> 99:59:59.999
So it doesn’t give you all the functional mechanisms,
99:59:59.999 --> 99:59:59.999
but some things that I think are necessary based on my current understanding.
99:59:59.999 --> 99:59:59.999
We’re currently at the second…
99:59:59.999 --> 99:59:59.999
iteration of the implementations.
99:59:59.999 --> 99:59:59.999
The first one was in Java in early 2003 with lots of XMI files and…
99:59:59.999 --> 99:59:59.999
… XML files … and design patterns and Eclipse plug ins.
99:59:59.999 --> 99:59:59.999
And the new one is, of course, … runs in the browser, and is written in Python,
99:59:59.999 --> 99:59:59.999
and is much more light-weight and much more joy to work with.
99:59:59.999 --> 99:59:59.999
But we’re not done yet.
99:59:59.999 --> 99:59:59.999
OK.
99:59:59.999 --> 99:59:59.999
So this gets back to that question: is it going to be one big idea or is it going to be incremental progress?
99:59:59.999 --> 99:59:59.999
And I think it’s the latter.
99:59:59.999 --> 99:59:59.999
If we want to look at this extremely simplified list of problems to solve:
99:59:59.999 --> 99:59:59.999
whole testable architectures,
99:59:59.999 --> 99:59:59.999
universal representations,
99:59:59.999 --> 99:59:59.999
universal problem solving,
99:59:59.999 --> 99:59:59.999
motivation, emotion, and effect, and so on.
99:59:59.999 --> 99:59:59.999
And I can see hundreds and hundreds of Ph.D. thesis.
99:59:59.999 --> 99:59:59.999
And I’m sure that I only see a tiny part of the problem.
99:59:59.999 --> 99:59:59.999
So I think it’s entirely doable,
99:59:59.999 --> 99:59:59.999
but it’s going to take a pretty long time.
99:59:59.999 --> 99:59:59.999
And it’s going to be very exciting all the way,
99:59:59.999 --> 99:59:59.999
because we are going to learn that we are full of shit
99:59:59.999 --> 99:59:59.999
as we always do to a new problem, an algorithm,
99:59:59.999 --> 99:59:59.999
and we realize that we can’t test it,
99:59:59.999 --> 99:59:59.999
and that our initial idea was wrong,
99:59:59.999 --> 99:59:59.999
and that we can improve on it.
99:59:59.999 --> 99:59:59.999
So what should you do, if you want to get into AI?
99:59:59.999 --> 99:59:59.999
And you’re not there yet?
99:59:59.999 --> 99:59:59.999
So, I think you should get acquainted, of course, with the basic methodology.
99:59:59.999 --> 99:59:59.999
You want to…
99:59:59.999 --> 99:59:59.999
get programming languages, and learn them.
99:59:59.999 --> 99:59:59.999
Basically do it for fun.
99:59:59.999 --> 99:59:59.999
It’s really fun to wrap your mind around programming languages.
99:59:59.999 --> 99:59:59.999
Changes the way you think.
99:59:59.999 --> 99:59:59.999
And you want to learn software development.
99:59:59.999 --> 99:59:59.999
That is, build an actual, running system.
99:59:59.999 --> 99:59:59.999
Test-driven development.
99:59:59.999 --> 99:59:59.999
All those things.
99:59:59.999 --> 99:59:59.999
Then you want to look at the things that we do in AI.
99:59:59.999 --> 99:59:59.999
So for like…
99:59:59.999 --> 99:59:59.999
machine learning, probabilistic approaches, Kalman filtering,
99:59:59.999 --> 99:59:59.999
POMDPs and so on.
99:59:59.999 --> 99:59:59.999
You want to look at modes of representation: semantic networks, description logics, factor graphs, and so on.
99:59:59.999 --> 99:59:59.999
Graph Theory,
99:59:59.999 --> 99:59:59.999
hyper graphs.
99:59:59.999 --> 99:59:59.999
And you want to look at the domain of cognitive architectures.
99:59:59.999 --> 99:59:59.999
That is building computational models to simulate psychological phenomena,
99:59:59.999 --> 99:59:59.999
and reproduce them, and test them.
99:59:59.999 --> 99:59:59.999
I don’t think that you should stop there.
99:59:59.999 --> 99:59:59.999
You need to take in all the things, that we haven’t taken in yet.
99:59:59.999 --> 99:59:59.999
We need to learn more about linguistics.
99:59:59.999 --> 99:59:59.999
We need to learn more about neuroscience in our field.
99:59:59.999 --> 99:59:59.999
We need to do philosophy of mind.
99:59:59.999 --> 99:59:59.999
I think what you need to do is study cognitive science.
99:59:59.999 --> 99:59:59.999
So. What should you be working on?
99:59:59.999 --> 99:59:59.999
Some of the most pressing questions to me are, for instance, representation.
99:59:59.999 --> 99:59:59.999
How can we get abstract and perceptual presentation right
99:59:59.999 --> 99:59:59.999
and interact with each other on a common ground?
99:59:59.999 --> 99:59:59.999
How can we work with ambiguity and superposition of representations.
99:59:59.999 --> 99:59:59.999
Many possible interpretations valid at the same time.
99:59:59.999 --> 99:59:59.999
Inheritance and polymorphy.
99:59:59.999 --> 99:59:59.999
How can we distribute representations in the mind
99:59:59.999 --> 99:59:59.999
and store them efficiently?
99:59:59.999 --> 99:59:59.999
How can we use representation in such a way
99:59:59.999 --> 99:59:59.999
that even parts of them are very valid.
99:59:59.999 --> 99:59:59.999
And we can use constraints to describe partial presentations.
99:59:59.999 --> 99:59:59.999
For instance imagine a house.
99:59:59.999 --> 99:59:59.999
And you already have the backside of the house,
99:59:59.999 --> 99:59:59.999
and the number of windows in that house,
99:59:59.999 --> 99:59:59.999
and you already see this complete picture in your house,
99:59:59.999 --> 99:59:59.999
and at each time,
99:59:59.999 --> 99:59:59.999
if I say: “OK. It’s a house with nine stories.”
99:59:59.999 --> 99:59:59.999
this representation is going to change
99:59:59.999 --> 99:59:59.999
based on these constraints.
99:59:59.999 --> 99:59:59.999
How can we implement this?
99:59:59.999 --> 99:59:59.999
And of course we want to implement time.
99:59:59.999 --> 99:59:59.999
And we want…
99:59:59.999 --> 99:59:59.999
to produce uncertain space,
99:59:59.999 --> 99:59:59.999
and certain space
99:59:59.999 --> 99:59:59.999
and openness, and closed environments.
99:59:59.999 --> 99:59:59.999
And we want to have temporal loops and actually loops and physical loops.
99:59:59.999 --> 99:59:59.999
Uncertain loops and all those things.
99:59:59.999 --> 99:59:59.999
Next thing: perception.
99:59:59.999 --> 99:59:59.999
Perception is crucial.
99:59:59.999 --> 99:59:59.999
It’s…. Part of it is bottom up,
99:59:59.999 --> 99:59:59.999
that is driven by cues from stimuli from the environment,
99:59:59.999 --> 99:59:59.999
part of his top down. It’s driven by what we expect to see.
99:59:59.999 --> 99:59:59.999
Actually most of it, about 10 times as much,
99:59:59.999 --> 99:59:59.999
is driven by what we expect to see.
99:59:59.999 --> 99:59:59.999
So we actually—actively—check for stimuli in the environment.
99:59:59.999 --> 99:59:59.999
And this bottom-up top-down process in perception is interleaved.
99:59:59.999 --> 99:59:59.999
And it’s adaptive.
99:59:59.999 --> 99:59:59.999
We create new concepts and integrate them.
99:59:59.999 --> 99:59:59.999
And we can revise those concepts over time.
99:59:59.999 --> 99:59:59.999
And we can adapt it to a given environment
99:59:59.999 --> 99:59:59.999
without completely revising those representations.
99:59:59.999 --> 99:59:59.999
Without making them unstable.
99:59:59.999 --> 99:59:59.999
And it works both on sensory input and memory.
99:59:59.999 --> 99:59:59.999
I think that memory access is mostly a perceptual process.
99:59:59.999 --> 99:59:59.999
It has anytime characteristics.
99:59:59.999 --> 99:59:59.999
So it works with partial solutions and is useful already.
99:59:59.999 --> 99:59:59.999
Categorization.
99:59:59.999 --> 99:59:59.999
We want to have categories based on saliency,
99:59:59.999 --> 99:59:59.999
that is on similarity and dissimilarity, and so on that you can perceive.
99:59:59.999 --> 99:59:59.999
We…. Based on goals on motivational relevance.
99:59:59.999 --> 99:59:59.999
And on social criteria.
99:59:59.999 --> 99:59:59.999
Somebody suggests me categories,
99:59:59.999 --> 99:59:59.999
and I find out what they mean by those categories.
99:59:59.999 --> 99:59:59.999
What’s the difference between cats and dogs?
99:59:59.999 --> 99:59:59.999
I never came up with this idea on my own to make two baskets:
99:59:59.999 --> 99:59:59.999
and the pekinese and the shepherds in one and all the cats in the other.
99:59:59.999 --> 99:59:59.999
But if you suggest it to me, I come up with a classifier.
99:59:59.999 --> 99:59:59.999
Then… next thing: universal problem solving and taskability.
99:59:59.999 --> 99:59:59.999
If we don’t want to have specific solutions;
99:59:59.999 --> 99:59:59.999
we want to have general solutions.
99:59:59.999 --> 99:59:59.999
We want it to be able to play every game,
99:59:59.999 --> 99:59:59.999
to find out how to play every game for instance.
99:59:59.999 --> 99:59:59.999
Language: the big domain of organizing mental representations,
99:59:59.999 --> 99:59:59.999
which are probably fuzzy, distributed hyper-graphs
99:59:59.999 --> 99:59:59.999
into discrete strings of symbols.
99:59:59.999 --> 99:59:59.999
Sociality:
99:59:59.999 --> 99:59:59.999
interpreting others.
99:59:59.999 --> 99:59:59.999
It’s what we call theory of mind.
99:59:59.999 --> 99:59:59.999
Social drives, which make us conform to social situations and engage in them.
99:59:59.999 --> 99:59:59.999
Personhood and self-concept.
99:59:59.999 --> 99:59:59.999
How does that work?
99:59:59.999 --> 99:59:59.999
Personality properties.
99:59:59.999 --> 99:59:59.999
How can we understand, and implement, and test for them?
99:59:59.999 --> 99:59:59.999
Then the big issue of integration.
99:59:59.999 --> 99:59:59.999
How can we get analytical and associative operations to work together?
99:59:59.999 --> 99:59:59.999
Attention.
99:59:59.999 --> 99:59:59.999
How can we direct attention and mental resources between different problems?
99:59:59.999 --> 99:59:59.999
Developmental trajectory.
99:59:59.999 --> 99:59:59.999
How can we start as kids and grow our system to become more and more adult like and even maybe surpass that?
99:59:59.999 --> 99:59:59.999
Persistence.
99:59:59.999 --> 99:59:59.999
How can we make the system stay active instead of rebooting it every other day, because it becomes unstable.
99:59:59.999 --> 99:59:59.999
And then benchmark problems.
99:59:59.999 --> 99:59:59.999
We know, most AI is having benchmarks like
99:59:59.999 --> 99:59:59.999
how to drive a car,
99:59:59.999 --> 99:59:59.999
or how to control a robot,
99:59:59.999 --> 99:59:59.999
or how to play soccer.
99:59:59.999 --> 99:59:59.999
And you end up with car driving toasters, and
99:59:59.999 --> 99:59:59.999
soccer-playing toasters,
99:59:59.999 --> 99:59:59.999
and chess playing toasters.
99:59:59.999 --> 99:59:59.999
But actually, we want to have a system
99:59:59.999 --> 99:59:59.999
that is forced to have a mind.
99:59:59.999 --> 99:59:59.999
That needs to be our benchmarks.
99:59:59.999 --> 99:59:59.999
So we need to find tasks that enforce all this universal problem solving,
99:59:59.999 --> 99:59:59.999
and representation, and perception,
99:59:59.999 --> 99:59:59.999
and supports the incremental development.
99:59:59.999 --> 99:59:59.999
And that inspires a research community.
99:59:59.999 --> 99:59:59.999
And, last but not least, it needs to attract funding.
99:59:59.999 --> 99:59:59.999
So.
99:59:59.999 --> 99:59:59.999
It needs to be something that people can understand and engage in.
99:59:59.999 --> 99:59:59.999
And that seems to be meaningful to people.
99:59:59.999 --> 99:59:59.999
So this is a bunch of the issues that need to be urgently addressed…
99:59:59.999 --> 99:59:59.999
… in the next…
99:59:59.999 --> 99:59:59.999
15 years or so.
99:59:59.999 --> 99:59:59.999
And this means, for …
99:59:59.999 --> 99:59:59.999
… my immediate scientific career, and for yours.
99:59:59.999 --> 99:59:59.999
You get a little bit more information on the home of the project, which is micropsi.com.
99:59:59.999 --> 99:59:59.999
You can also send me emails if you’re interested.
99:59:59.999 --> 99:59:59.999
And I want to thank a lot of people which have supported me. And …
99:59:59.999 --> 99:59:59.999
you for your attention.
99:59:59.999 --> 99:59:59.999
And giving me the chance to talk about AI.
99:59:59.999 --> 99:59:59.999
[applause]