Music

There is essentially two ways of attacking intelligence.

And you can kind of tabulate them between what their coverage is and the depth of

analysis or functionality. When I say narrow then I mean that usually you can

only do things for a very limited domain. Multiplying hundred digit numbers.

Okay, that's a very limited domain and it's something we completely conquer.

Right, as I said my wristwatch, my smartphone, my computer all can do it,

just my brain can't. Other things, wider coverage is what the touring tests.

You can talk about anything. Then there's shallow and deep. A good

example for that is your text editor. Think about Word. Very nice. You type a

little bit and it makes a nicely formatted text out of it and it does AI.

It has a spell checker. That was AI at one point. Very very shallow, very simple.

You just take a lexicon and see whether every word is in there and if it's not

you make a little wavy line under it. And just the existence of a lexicon was

essentially an AI technology at some point. Very early because we understood

what it was we spun it off into computer science. That was shallow

functionality. Deep functionality might be something of the form this when word

would tell me and I'm writing a love letter saying Michael don't write this

love letter to this woman. Probably not a good idea.

At least my word never told me. I'm not using word by the way so maybe it

didn't have a chance. But okay that would be deep functionality. A

functionality that needs a lot of understanding and if you think

about who could give you this kind of advice in a kind of a controlled way my

brother. He could. He's good at deep analysis. Word cannot. Okay so that's kind

of for me the difference between deep and shallow. And what we want is wide

coverage deep like humans. And you can reach this in two ways. One of them is

stay shallow and first to wide coverage and the other one is stay narrow and

first to deep analysis. Okay doesn't quite get you there but it's something

you can try. And essentially what we're going to do is we're going to look at

these kind of techniques which we call symbolic AI or a good old-fashioned AI

go-fi in this semester on statistical methods that scale into wide coverage

better but usually stay somewhat shallow in the next semester. And my private

belief is that we will need kind of something some cooperation of shallow

data-driven statistical methods which is something we're going to look at in the

next semester and deep narrow knowledge-based methods symbolic AI which

is what we're going to do this semester. There is research here and there are

many people who kind of try to understand it at the framework level.

Much of it I consider underwhelming which means I don't think very highly of it.

Of course I can't do it better either but and there's one thing where I

have the feeling that could be it. Unfortunately I don't understand the

mathematics but I know the people who are actually working on this quite well

and so that that's fascinating but that's kind of ten years out until

something very useful at all useful will come out of that. There is an example

where this cooperation works very well and it is indeed AlphaGo. So AlphaGo

uses a symbolic technique Monte Carlo Tree Search which is something we will

learn in a couple of weeks and discuss which is kind of a symbolic framework

where we think of game playing as a special case of search and the problem

is that search is an exponential problem. All algorithms we know are exponential

which means that you have to look through huge set huge spaces of possible

solutions and you can do that if you have a very good sense of smell. You know