Google's Deep Mind "Alpha Zero" - amazing AI concept prooved in GO and chess - page 2

 
Amir Yacoby:
Yeh, but the underlying code does not tell the machine how to do it. It finds a way by itself. Moreover, the developers don't have a way to know what the machine learned.

...

Of course it did. Not in a specific way as we are doing in most of our code but in a generalized way. The "AI" doesn't find anything "by itself", it follows a code. If something is not provided in this code it will not be able to deal with it. But the code is very smart and the hardward very fast. There is no "magic".

 
Amir Yacoby:
By the way, about the speed:
Stockfish, the non-AI chess software - calculates 200 million positions per second (in the contest between them, given the hardware).
Alpha-Zero calculated 70 thousand position per second. It is 2000 times less which is a huge differrence.

The reason is it does not rely on brute force at all. The AI is giving it a kind of "intuition", and it only calculates as little as to make sure that it's "intuition" is correct.

Smarter underlying code. The coders of Alpha-Zero can be very proud.

 
Alain Verleyen:

Of course it did. Not in a specific way as we are doing in most of our code but in a generalized way. The "AI" doesn't find anything "by itself", it follows a code. If something is not provided in this code it will not be able to deal with it. But the code is very smart and the hardward very fast. There is no "magic".

No, Alain. It is not correct. What do you mean in a generlized way?
AI does not supply anything about the problem to the system, it just has features that help it learn by itself.
As I said before, the programmer does not know how to play chess, and it did not tell the program nothing about how to win in chess. Just told it the rules, and let it play against it self and learn (by scoring wins and losses) how to play.
 

Yes. AI is a hype buzz word. There is no intelligents, and it is not artificial in any of the mentioned concepts. Machine learning, yes. Pattern recognition, yes. That's about it.

As to AI trading the Forex market. It will learn from it's mistakes only to make a new mistake because the conditions have changed rendering what it has learned useless, then the process will repeat.

It will be one big a$$ lagging indicator. 

 
Alain Verleyen:

Interesting.

 All the rest is a matter of input, output, probability and very high speed.


When the code starts to re-write itself, we are summoning the demon...

There is the magic.

Many people do not understand this, this is the border where it leaves the dimension of 'just' being clockwork...

It will start to develop new languages on it's own, languages unknown to us humans, but many times more efficient, and you can no longer state that it follows the original programming. 

Nobody knows how it eventually came to it's output anymore.

So there might be more to it then just thinking in static code. 

 
Enrique Dangeroux:

Yes. AI is a hype buzz word. There is no intelligents, and it is not artificial in any of the mentioned concepts. Machine learning, yes. Pattern recognition, yes. That's about it.

As to AI trading the Forex market. It will learn from it's mistakes only to make a new mistake because the conditions have changed rendering what it has learned useless, then the process will repeat.

It will be one big a$$ lagging indicator. 

Machine learning - yes, pattern recognition - yes.
Exactly. Huge part of inteligence is contained in these two.
The GO game is the most complex game humanity knows (most options). No one up to 2016 would have tried to create a computer program for it, because the rules are so complex and the possibilities vast.
The GO world champion, a south korean guy, would not be able to explain why a move he takes in the game is considered good or bad. It is very complex.
The principles of why people play the way they play have been gathered in the last 3000 years.
Yet, an AI learned those and surpassed that knowldege in 40 days.

About forex, I am not sure at all. First, regarding money management (the lot sizes), it will probably find something better then the systems we have.
About deal management, as well (moving the stops and the takes).

And I think that also it might find ways to trade (gross example is a grid, but might be much better).
All together, I think it will acquire an edge over today's market participants.

 
Amir Yacoby:
No, Alain. It is not correct. What do you mean in a generlized way?
AI does not supply anything about the problem to the system, it just has features that help it learn by itself.
As I said before, the programmer does not know how to play chess, and it did not tell the program nothing about how to win in chess. Just told it the rules, and let it play against it self and learn (by scoring wins and losses) how to play.

This is what I mean...the programmers coded it in a generalized way, not specifically chess. Sorry my English vocabulary is too limited, I don't find a better way to explain what I mean. If you don't give it the rules it will not be able to do anything about Chess. Something intelligent, an human intelligence, is able to lean chess by itself, it will learn the rules just by looking others playing. (It's just an example, this AI can probably by configured to learn Chess rules too).

My point is there is no intelligence, it's an algorithm, it follows the algorithm, it can't create algorithms. 

Of course as I said at the start, all depends of what we mean by intelligence. A lot of people are still thinking QI is reflecting intelligence, some are even identifying QI and intelligence. Intelligence are multiple, QI is only about one of them, and even there is far from reflecting it totally. Anyway, that would lead us too far from this site probably.

I agree with Enrique, AI is a buzz word...for now. It could eventually change in the future.

 
Amir Yacoby:
Machine learning - yes, pattern recognition - yes.
Exactly. Huge part of inteligence is contained in these two.
The GO game is the most complex game humanity knows (most options). No one up to 2016 would have tried to create a computer program for it, because the rules are so complex and the possibilities vast.
The GO world champion, a south korean guy, would not be able to explain why a move he takes in the game is considered good or bad. It is very complex.
The principles of why people play the way they play have been gathered in the last 3000 years.
Yet, an AI learned those and surpassed that knowldege in 40 days.

About forex, I am not sure at all. First, regarding money management (the lot sizes), it will probably find something better then the systems we have.
About deal management, as well (moving the stops and the takes).

And I think that also it might find ways to trade (gross example is a grid, but might be much better).
All together, I think it will acquire an edge over today's market participants.

No, because you are talking about games,  and the market is not a game, so a machine can never beat a human if it's not a game, so you can't compare
 
Alain Verleyen:

If you don't give it the rules it will not be able to do anything about Chess.


Alain this is a misconception.

The current state of the technology is that they do learn to play by themselves they re-write their own code.

mrluck1:
No, because you are talking about games,  and the market is not a game, so a machine can never beat a human if it's not a game, so you can't compare
Of course the market is a game.
 

It is not a theoretical debate, if it's really a human inteligence or not.

It is about facts, and the facts are:

AI today is an algorithm of course, but the writer of the algorithm (human) does not know the final outcome of it, and even in succeeding (the AI plays better) he does not know why the AI played a move (for example, if it plays chess, you as the programmer can't tell why the AI chose certain move, and you can not influence it after it learned).

This makes a seperation between non-AI programming - where you need to tell the program the rules, and modern AI- where the algorithm is able to learn the connections, even if you yourself don't know those connections.

Call it what you want, it does not matter. I see it as valuable in trading as well.