Tip:
Highlight text to annotate it
X
Edward Frenkel Interview Chapter 4
Do you think that an AI can be developed that can pass the Turing Test? If so, can this AI be considered the true equivalent of the human mind?
There's a very important issue which is often overlooked.
What we mean by intelligence?
And I think that today most people mean by that...
...what encapsulates... the capacity which encapsulates thinking.
And that by itself, implicitly denies the possibility that there are certain functions of the human mind...
...which are not controlled by thinking.
And that I think is very big issue here.
It's almost...
When I talk to some people, it's almost like people look at this in disbelief.
What do you mean there is something which cannot be reached by thinking? You know?
So, that is one of the cornerstones I think of...
This is where sort of we get off the track.
The idea that everything ultimately can be rationalized...
...in the process of our consciousness, in the process of functioning of the human mind.
And things like intuition and instinct and insight,
...the prevailing point... prevailing view I believe,
...in the sort of scientific community, in the community of people who work with artificial intelligence and so on,
...is that this does not represent a different modality of functioning for the mind,
...that actually it is completely controlled by thinking.
So when they say intelligence they actually equate it with thinking.
Now, if we are talking narrowly about thinking,
...about thinking alone,
...then I suppose there is no... I don't see any...
...anything to prevent a machine, in principle, with the growth of computational power...
...to reach higher and higher levels of imitating thinking.
But, I know that thinking is a very small part of the functioning of the human psyche.
This I know.
And usually there's several examples. I mean, it is something that each of us has to discover on their own,
...because if I... whatever I say... is...
I don't want people to believe me,
...because I was speaking about my own experience; I am speaking about my own knowledge.
And, I don't want them to believe me when I say there is much more to human mind than thinking;
...I want them to discover it by themselves, but I can give some hints.
And one of the hints I would like to give is the nature of discovery in mathematical research.
Being a mathematician... a professional mathematician... this is what I've been doing all my life,
...I can speak about this with authority.
Anybody can google me and find out what kind of discoveries I have made in mathematics...
...to see that I am not making this up.
And I can share my experience. And this is something which I didn't realize for many years,
...and only recently came to realize... is that actually at the moment of discovery,
...even in mathematics... even in mathematics which is the most cerebral subject, one would say, right?
At the moment of discovery... at the actual moment of discovery,
...our thinking stops, something else comes into play.
Now thinking is important. Of course, I have to think and think and think about it, you know,
...but then, you know, like Einstein used to say:
"You think and think and then you go for a walk and whistle, and that's when the idea comes to you."
It's not by chance, it's not by accident.
Sometimes people read the stories and they say: "Well that's weird,"
"...chance occurrence that somehow the thought didn't come to him when he was thinking,"
"...but the thought came when he was whistling," but actually it's not by chance: That's how it works.
I was reading a biography of the great Indian mathematician Srinivasa Ramanujan, the other day.
And explains very clearly... He was famous for making discoveries which nobody could prove.
And, in fact, he didn't even have a formal math education,
...and he himself couldn't prove them.
So he was this math prodigy... very poor... from a poor family in India...
...and then eventually was discovered by great English mathematician,
...G.H. Hardy and brought to Cambridge,
...became a fellow of the Trinity College and member of the Royal Society and so on.
But interesting thing is that when he was asked how he made these discoveries,
...he said that these discoveries, these magical formulas,
...which mathematicians have been studying for the last hundred years,
...came to him in his dreams.
The goddess Namagiri would come to him in his dreams...
...and tell him these formulas which he would then write down when he wakes up.
You see? So... that's just an indication.
Now, in the same interview of Demis Hassabis,
...who is the head of DeepMind, I was reading that, different offices at his headquarters...
...in London are called after various people,
...and there is one office, one room which is called Ramanujan.
So, I wonder whether Demis Hassabis knows the story of how Ramanujan,
...by his own account, came to these discoveries,
...and whether when he finds out this would change his view,
...that everything about the human mind can be described...
...by deep learning algorithms.
So, this to me is a good indication that there is a lot more to human mind than thinking.
As far as thinking... just sort of bare thinking is concerned,
...I think that, yes, machines could imitate,
...could imitate and perhaps pass the Turing Test.
Turing himself by the way, made that mistake.
If you read his article... I read it a few months ago again.
He is very much sort of expressing this point of view that it's all about thinking.
He is not allowing for other aspects to come forth in this discussion.
And look, you've got Garry Kasparov being beaten by Deep Blue computer;
...you've got a computer by the same DeepMind and, independently, another group of researchers...
...beating a grandmaster in Go.
When people look at these achievements, and these are definite achievements,
...and I congratulate the scientists who are involved in this research.
It is really an amazing achievement. Absolutely fascinating. I'm really fascinated by this.
But a lot of people don't realize that,
...these algorithms are not operating in the same way...
...in which Kasparov is playing chess or a Go master is playing chess.
These algorithms are essentially built on calculating all possible positions,
...just whereas Kasparov cuts to the main move right away...
...using his other capacities: His intuition, his instincts.
Kasparov actually wrote a fascinating article in the New York Review of Books a few years ago,
...in which he very explicitly explained that. Anybody can find it online and read it.
And so, you know, the way these algorithms work is,
...these learning algorithms which everybody's so excited about,
...it's a really very simple idea: You have a big sample,
...you know, hundreds of thousands of positions or photographs...
...for some recognition software... image recognition software... lots of programs.
And then you feed these into the algorithm and let it choose...
...a particular algorithm which approximates a human-made choice.
So, you feed the images in which you say this is a cat, this is a dog...
...so the computer knows that this is a cat, this is a dog, and it tries to come up with the most efficient algorithm...
...to decide whether it's a cat or a dog by analyzing the pixels.
But I'm sorry, that's not what a two-month-old child or six-month-old child is doing.
A six-month-old child knows very well which one is cat and which is a dog,
...and nobody has fed this child a hundred thousand pictures, and he's doing much better.
So, at least admit and accept that humans are learning...
...in a different way, and they're using different... other capacities.
That would be first step to kind of like dispelling those myths... dispelling those false ideas.
But, yes, certain things can be done... In certain areas such as board games or video games,
...the human capacity for intuition... and perhaps even in some areas of image recognition and so on, for sure,
...human capacity for intuition and insight, can be compensated by excessive computation.
But it doesn't mean that this is the way we function;
...it doesn't mean that this is a way in which you can reproduce human intelligence.
So, that's my answer.