AlphaGo Defeats World's Best Human Player

AI Go Program Performs Better Than Expected

Anything that doesn't fit anywhere else below.

Moderators: Calilasseia, ADParker

Re: AlphaGo Defeats World's Best Human Player

#41  Postby VazScep » Sep 05, 2017 5:15 pm

GrahamH wrote:
VazScep wrote:
GrahamH wrote:
tuco wrote:That seems like a question for philosophy. Unless convinced otherwise, I will go with lets say AI scientists who consider such environment suitable for stated reasons. Ultimately, the point of such exercise is imo for AI to reach set goal(s). If we take the war element out, we are left with, well, managing economy.


I doubt the scientists give that much thought to such ethical or philosophical issues of unintended consequences

Maybe Elon Musk and Stephen hawking have a point

If AI is only good for war games and financial trading we would be better without it.
This sort of AI is good for any sort of problem that amounts to numerical optimisation, such as the logistics of food distribution or town planning, standard problems for which we generally get a good bang for the buck with machine learning.

It happens that Chess and Go were classic AI problems, and so there was a lot of kudos in beating all humans at them. Computer games are a trivial source of AI problems, since solving games is always an AI problem. Forget the war aspect. A non-cooperative game (which is the vast majority of games) can always be described in terms of warfare. This is a case of having an all two powerful metaphor.

Another reason games are targeted is because they provide a closed environment for a learning machine: all the information is there, and the game is formally specified.


I guess it would be more comforting for the AI to be trained on something other than zero-sum games because if we find ourselves in a zero-sum game with advanced AI we could well be screwed.
I'm not concerned with the war aspect as such, more that future AI may carry forward aspect of playing to win into new interactions.

Someone once speculated that setting a goal for AI of wining as many games of chess as possible could have dire unforeseen consequences if the AI took over as much computing resource as possible in order to meet that goal. Don't set a goal of eliminating human suffering in case an AI finds and implements the obvious (final) solution!
That's the paperclip problem. Maximise paperclip production -> exterminate all life and turn over the land to paperclip factories.

But deep learning isn't set up that way. I can see potential dangers at throwing AIs at problems such as maximising resources on a computer network, only because modern computer networks are hopelessly fragile and full of exploitable holes that an AI might open. But that's a problem with computer security that I've moaned about elsewhere, and I consider it a clusterfuck even without the AIs. I can see the same issues when throwing AIs into the space of smart contracts and tax systems. That stuff needs to be locked down and as formally specified as Chess before we start running decent AI at it.
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4590

United Kingdom (uk)
Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#42  Postby tuco » Sep 05, 2017 7:15 pm

Not like you care(?) but that is the reason for being alarmed by AI then? Not because its somehow inherently dangerous, but because it can find exploits easier and faster than all humans on the planet can. It was not obvious to me before. So I guess that answers this question.
tuco
 
Posts: 13802

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#43  Postby VazScep » Sep 05, 2017 8:41 pm

Yes. That has always been my concern with AI: it finds unexpected solutions. And in cases such as insecure networks and vulnerable code, the unexpected solutions are mostly vulnerabilities. They're unexpected precisely because it was the human programmer who fucked up by not expecting them or, worse, by blithely coding them in (like all other programmers in existence, I am as far from innocent of this crime as anyone)
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4590

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#44  Postby minininja » Oct 18, 2017 6:44 pm

So the new version, after being told nothing but the rules of Go, taught itself enough in three days to beat the previous version 100 games to 0. :shock:

'It's able to create knowledge itself': Google unveils AI that learns on its own

Google’s artificial intelligence group, DeepMind, has unveiled the latest incarnation of its Go-playing program, AlphaGo – an AI so powerful that it derived thousands of years of human knowledge of the game before inventing better moves of its own, all in the space of three days.

Named AlphaGo Zero, the AI program has been hailed as a major advance because it mastered the ancient Chinese board game from scratch, and with no human help beyond being told the rules. In games against the 2015 version, which famously beat Lee Sedol, the South Korean grandmaster, AlphaGo Zero won 100 to 0.

The feat marks a milestone on the road to general-purpose AIs that can do more than thrash humans at board games. Because AlphaGo Zero learns on its own from a blank slate, its talents can now be turned to a host of real-world problems.

At DeepMind, which is based in London, AlphaGo Zero is working out how proteins fold, a massive scientific challenge that could give drug discovery a sorely needed shot in the arm.


Please do not plug that thing into the internet.
[Disclaimer - if this is comes across like I think I know what I'm talking about, I want to make it clear that I don't. I'm just trying to get my thoughts down]
User avatar
minininja
 
Posts: 1186

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#45  Postby I'm With Stupid » Oct 19, 2017 10:58 am

Yeah, but can it win at Monopoly?
Image
User avatar
I'm With Stupid
 
Posts: 9042
Age: 34
Male

Country: UK
Jolly Roger (arr)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#46  Postby minininja » Oct 19, 2017 2:31 pm

I'm With Stupid wrote:Yeah, but can it win at Monopoly?

Tell it the rules and give it a day to think about them.
[Disclaimer - if this is comes across like I think I know what I'm talking about, I want to make it clear that I don't. I'm just trying to get my thoughts down]
User avatar
minininja
 
Posts: 1186

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#47  Postby Tracer Tong » Nov 23, 2017 8:04 pm

Is chess solved yet?
Die Liebe ist nicht schön – es ist nur der Traum der Liebe, der entzückt
User avatar
Tracer Tong
 
Posts: 1458
Male

Country: England
England (eng)
Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#48  Postby Matthew Shute » Nov 23, 2017 8:43 pm

Chess was a cinch compared to go. A computer called Deep Blue beat Garry Kasparov in 1997.
"Change will preserve us. It is the lifeblood of the Isles. It will move mountains! It will mount movements!" - Sheogorath
User avatar
Matthew Shute
 
Name: Matthew Shute
Posts: 3642
Age: 40

Antarctica (aq)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#49  Postby VazScep » Nov 23, 2017 9:20 pm

Tracer Tong wrote:Is chess solved yet?
Technically, no. When a game theorist says "solved", they mean that they have an algorithm that either guarantees a win or forces a draw. Noughts and crosses is solved. Chess isn't.
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4590

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#50  Postby Tracer Tong » Nov 23, 2017 9:51 pm

Matthew Shute wrote:Chess was a cinch compared to go. A computer called Deep Blue beat Garry Kasparov in 1997.


I see. Just in case, take a look at this:

https://lichess.org/analysis/3k4/1r2p3/ ... K3/8_w_-_-

What does the computer say? Feel free to leave it think.

VazScep wrote:
Tracer Tong wrote:Is chess solved yet?
Technically, no. When a game theorist says "solved", they mean that they have an algorithm that either guarantees a win or forces a draw. Noughts and crosses is solved. Chess isn't.


Yeah, I know. In fact, it's not even close to being solved. I'm also open to the idea that a human player may yet emerge who could rival the strongest engines, though we'll see.
Die Liebe ist nicht schön – es ist nur der Traum der Liebe, der entzückt
User avatar
Tracer Tong
 
Posts: 1458
Male

Country: England
England (eng)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#51  Postby Matthew Shute » Nov 24, 2017 12:17 pm

Well, a human is no closer to having an algorithm to force a win or draw than is a machine. Presumably, if one is found, it'll also run on a computer, guaranteeing the computer a win or a draw. In the meantime the best chess machines will very likely continue to beat the best humans at chess. Their processing speeds aren't declining, ours aren't increasing, and some of the smartest humans are always trying to build ever-more competent AIs. So don't hold your breath waiting for the great human hope at chess.
"Change will preserve us. It is the lifeblood of the Isles. It will move mountains! It will mount movements!" - Sheogorath
User avatar
Matthew Shute
 
Name: Matthew Shute
Posts: 3642
Age: 40

Antarctica (aq)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#52  Postby Tracer Tong » Nov 24, 2017 12:24 pm

What does the computer say about the position I linked to?
Die Liebe ist nicht schön – es ist nur der Traum der Liebe, der entzückt
User avatar
Tracer Tong
 
Posts: 1458
Male

Country: England
England (eng)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#53  Postby minininja » Nov 24, 2017 2:01 pm

Tracer Tong wrote:What does the computer say about the position I linked to?

Well obviously simple chess computers that you can run on a mobile phone and that only look at a points based advantage over a set number of moves will end up in stalemate from there. But the whole point of the deep learning AIs, and the reason they test them with Go, is that they are developing their own methods of analysing strategic positional advantage in ways that humans haven't seen before.
[Disclaimer - if this is comes across like I think I know what I'm talking about, I want to make it clear that I don't. I'm just trying to get my thoughts down]
User avatar
minininja
 
Posts: 1186

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#54  Postby Tracer Tong » Nov 24, 2017 2:17 pm

minininja wrote:
Tracer Tong wrote:What does the computer say about the position I linked to?

Well obviously simple chess computers that you can run on a mobile phone and that only look at a points based advantage over a set number of moves will end up in stalemate from there.


I’m sorry, but I really don’t know what you’re talking about here.

minininja wrote:
But the whole point of the deep learning AIs, and the reason they test them with Go, is that they are developing their own methods of analysing strategic positional advantage in ways that humans haven't seen before.


Cool.
Die Liebe ist nicht schön – es ist nur der Traum der Liebe, der entzückt
User avatar
Tracer Tong
 
Posts: 1458
Male

Country: England
England (eng)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#55  Postby minininja » Nov 25, 2017 12:15 pm

Tracer Tong wrote:
minininja wrote:
Tracer Tong wrote:What does the computer say about the position I linked to?

Well obviously simple chess computers that you can run on a mobile phone and that only look at a points based advantage over a set number of moves will end up in stalemate from there.

I’m sorry, but I really don’t know what you’re talking about here.

Then perhaps I'm misunderstanding your point. What are you saying about the position you linked to?
[Disclaimer - if this is comes across like I think I know what I'm talking about, I want to make it clear that I don't. I'm just trying to get my thoughts down]
User avatar
minininja
 
Posts: 1186

United Kingdom (uk)
Print view this post

Ads by Google


Re: AlphaGo Defeats World's Best Human Player

#56  Postby Tracer Tong » Nov 25, 2017 5:03 pm

It's a salutary reminder of the weaknesses of computers in playing chess, even 20 years after Deep Blue's (narrow) victory. The position I linked to is the final position of a chess puzzle created in 1912 by a W. E. Rudolph, in which white is given the seemingly impossible task of drawing. Nevertheless, in that final position, it should be obvious to even a beginner that the position is dead drawn: there is no way for black to make progress, since he's locked in behind his own pawns. Yet the latest version of Stockfish, probably the world's most powerful chess engine (and whose analysis you can find in my link), suggests black is completely winning. Older versions (I'm not sure about the more recent) of Houdini and Komodo suggest the same, as does the latest version of Fire. This isn't some isolated faux pas, either: there are very many completely drawn positions which these chess engines insist are winning for one side or the other.
Die Liebe ist nicht schön – es ist nur der Traum der Liebe, der entzückt
User avatar
Tracer Tong
 
Posts: 1458
Male

Country: England
England (eng)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#57  Postby VazScep » Nov 26, 2017 2:57 pm

I'd hope things have changed now, but traditional Chess AI design works by looking a set number of moves ahead on each turn. The AI then has a simple heuristic to determine the value of the board after that number of moves, and picks the best. Apparently, you only need to be at 8 moves ahead before you're generally playing at master level.

The heuristics are supposed to be dumb. You need them to run fast, not smart. All the real "intelligence" is happening because of the lookahead. Deep Blue, so I'm told, is, in some cases, doing look ahead to a depth of several hundred moves.

Point is, the heuristics are so dumb that it shouldn't be difficult to contrive board setups where they turn out to give a completely shitty value for the board state, as you say, suggesting that one player is winning when it's clearly going for a draw. But then, as I say, the real computation is happening in the lookahead, not the heuristics.

AlphaGo isn't using such a naive algorithm, as I understand it (and I don't understand it very much), and is effectively coming up with its own heuristics.
Here we go again. First, we discover recursion.
VazScep
 
Posts: 4590

United Kingdom (uk)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#58  Postby Tracer Tong » Dec 06, 2017 6:38 pm

Then, like clockwork, today DeepMind crushes Stockfish.

https://www.theverge.com/2017/12/6/1674 ... o-shogi-go
Die Liebe ist nicht schön – es ist nur der Traum der Liebe, der entzückt
User avatar
Tracer Tong
 
Posts: 1458
Male

Country: England
England (eng)
Print view this post

Re: AlphaGo Defeats World's Best Human Player

#59  Postby tuco » Dec 12, 2017 7:54 pm

Well, it cannot learn itself shooting from distance, http://www.rationalskepticism.org/news- ... 54301.html, so Stockfish is no news really.

Hopefully, it will learn itself decentralized youtube, however.
tuco
 
Posts: 13802

Print view this post

Re: AlphaGo Defeats World's Best Human Player

#60  Postby Matthew Shute » Mar 01, 2018 3:09 pm

This isn't go, chess, or Dota 2 but...

A video game-playing AI beat Q*bert in a way no one’s ever seen before.
https://www.theverge.com/tldr/2018/2/28 ... -bug-cheat


"Change will preserve us. It is the lifeblood of the Isles. It will move mountains! It will mount movements!" - Sheogorath
User avatar
Matthew Shute
 
Name: Matthew Shute
Posts: 3642
Age: 40

Antarctica (aq)
Print view this post

Previous

Return to General Science & Technology

Who is online

Users viewing this topic: No registered users and 1 guest