Recommend
9 
 Thumb up
 Hide
23 Posts

Abstract Games» Forums » General

Subject: Giraffe learns to play Chess rss

Your Tags: Add tags
Popular Tags: [View All]
Andreas Last
Germany
Langen
flag msg tools
mbmbmb
Yes, that's right! It took it 72 hours to lean to play on International Master level. And next it might take on Go. Read more here about this impress AI
5 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Stephen Tavener
United Kingdom
London
England
flag msg tools
designer
The overtext below is true.
badge
The overtext above is false.
Avatar
mbmbmbmbmb
Impressive indeed - though it required a vast number of high-quality games to train itself, so won't be very useful outside of a few lifestyle games... it'll be very interesting to see what it makes of go, though
4 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
This is a significant development that builds on large databases to train the neural network. I don't see speed as much of a barrier (I hear all kinds of stuff about quantum computers here at the Delft University of Technology:
https://en.wikipedia.org/wiki/Delft_University_of_Technology

I don't doubt the learning capacity and the algorithms enabling it either. But how to fine-tune such machines without large databases to support the learning process isn't quite clear to me. Can the process in principle start from scratch I wonder?
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Robert Wesley
Nepal
Aberdeen
Washington
flag msg tools
designer
badge
Avatar
mb
Re: "Gerricoiffed" Giraffe learns to play Chess VS???
wow ~"RAD"-scramblerouija ALERT! "Stonewall" Jackson's 'e`spirit' shall ARISE-(along as with the SOUTH-player) to engage through 'Checkers' whilst "Gerricoifed" Giraffe played CHESS!
_sauron_
..zombie ~"I'ma 'cross-back' OVER the 'river' FOR yas!"
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Corey Clark
Canada
flag msg tools
designer
mbmb
GROGnads wrote:
wow ~"RAD"-scramblerouija ALERT! "Stonewall" Jackson's 'e`spirit' shall ARISE-(along as with the SOUTH-player) to engage through 'Checkers' whilst "Gerricoifed" Giraffe played CHESS!
_sauron_
..zombie ~"I'ma 'cross-back' OVER the 'river' FOR yas!"


This is what happens when you are raised in a convent of James Joyce cultists.... Or you are Grognads
4 
 Thumb up
0.02
 tip
 Hide
  • [+] Dice rolls
Andreas Last
Germany
Langen
flag msg tools
mbmbmb
christianF wrote:
I don't doubt the learning capacity and the algorithms enabling it either. But how to fine-tune such machines without large databases to support the learning process isn't quite clear to me. Can the process in principle start from scratch I wonder?


Yes, in theory a neural network should be able to start learning from scratch. All it needs to know are the rules. The weightings of paths through the network is what it adjusts itself anyway. It would just take so much longer without the DB to get to the same point.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Dieter Stein
Germany
flag msg tools
designer
Avatar
This all is very interesting and fascinating.

But wait, there's something else. Perfect (as well as random) play is actually the absence of "play". Don't forget that a game (besides being object to game theory) is also a sensual human reality. When we, some day, all are equipped with a brain-implanted Giraffe chip, chess, Go etc. are all dead. I strongly hope there are still things humans can enjoy and love in the future.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Steve
United Kingdom
Farnham
Surrey
flag msg tools
designer
Avatar
mb
mrraow wrote:
it'll be very interesting to see what it makes of go, though

From the details in the article about assessing board position, I suspect the programmer has learned from computational Go and applied the methods to chess, so I doubt there'll be much of an impact on Go playing computing.

Go programmes shifted from computational brute force to positional heuristics decades ago.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Andreas Last
Germany
Langen
flag msg tools
mbmbmb
spielstein wrote:
But wait, there's something else. Perfect (as well as random) play is actually the absence of "play". Don't forget that a game (besides being object to game theory) is also a sensual human reality. When we, some day, all are equipped with a brain-implanted Giraffe chip, chess, Go etc. are all dead. I strongly hope there are still things humans can enjoy and love in the future.


I don't think anyone is forgetting that. I rather play myself than watching a bot play Also, Giraffe is far from perfect play. That's not even its goal, as only a brute force approach could get there. It's more the simulation of brain learning applied to a game that's interesting here. And based on that how quickly it learned even though computing much slower than brute force.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Russ Williams
Poland
Wrocław
Dolny Śląsk
flag msg tools
designer
badge
Avatar
mbmbmbmbmb
spielstein wrote:
This all is very interesting and fascinating.

But wait, there's something else. Perfect (as well as random) play is actually the absence of "play". Don't forget that a game (besides being object to game theory) is also a sensual human reality. When we, some day, all are equipped with a brain-implanted Giraffe chip, chess, Go etc. are all dead. I strongly hope there are still things humans can enjoy and love in the future.

I'm not very worried about that scenario happening. E.g. many people still run marathon races etc despite the fact that we have machines like bicycles and automobiles which can go faster than humans running with only their own strength. People still enjoy solving sudoku problems even though a computer program could solve them in a few seconds or faster. Etc.
5 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Dieter Stein
Germany
flag msg tools
designer
Avatar
russ wrote:
I'm not very worried about that scenario happening. E.g. many people still run marathon races etc despite the fact that we have machines like bicycles and automobiles which can go faster than humans running with only their own strength. People still enjoy solving sudoku problems even though a computer program could solve them in a few seconds or faster. Etc.


Yes, we still use machines as external tools. But what if these tools become part of the body - even the brain? I hope you are right and future generations still meet personally to have a talk or to exchange ideas and thoughts, acting as humans and not based on optimized algorithms - it may well be they turn off that brain extension module to play a game.
3 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
spielstein wrote:
russ wrote:
I'm not very worried about that scenario happening. E.g. many people still run marathon races etc despite the fact that we have machines like bicycles and automobiles which can go faster than humans running with only their own strength. People still enjoy solving sudoku problems even though a computer program could solve them in a few seconds or faster. Etc.


Yes, we still use machines as external tools. But what if these tools become part of the body - even the brain? I hope you are right and future generations still meet personally to have a talk or to exchange ideas and thoughts, acting as humans and not based on optimized algorithms - it may well be they turn off that brain extension module to play a game.

Unless of course the brain butler is so integrated that we wouldn't be able to distinguish betweens 'its' thoughts and 'ours' anymore.
4 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Dieter Stein
Germany
flag msg tools
designer
Avatar
Alast wrote:
Giraffe is far from perfect play. That's not even its goal, as only a brute force approach could get there. It's more the simulation of brain learning applied to a game that's interesting here. And based on that how quickly it learned even though computing much slower than brute force.


I know this. It was not my point. There are more things to come in the field of AI. I wanted to point out that in the future we as humans will have to say goodbye to things we appreciate now. We should be careful with ourselves and not blindly thrilled by progress.
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Andreas Last
Germany
Langen
flag msg tools
mbmbmb
spielstein wrote:
I know this. It was not my point. There are more things to come in the field of AI. I wanted to point out that in the future we as humans will have to say goodbye to things we appreciate now. We should be careful with ourselves and not blindly thrilled by progress.


I see. Well, I sure am not blindly thrilled AI progress is one thing, brain implants another.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Dieter Stein
Germany
flag msg tools
designer
Avatar
Alast wrote:
Well, I sure am not blindly thrilled AI progress is one thing, brain implants another.


It wasn't aimed at you, Andreas.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Richard Moxham
United Kingdom
flag msg tools
designer
Avatar
mb
spielstein wrote:
Perfect (as well as random) play is actually the absence of "play".

Well, I join in with the chorus of those who were in no danger of forgetting this, but for the opposite reason: I'd never formed the thought before.

So thank you, Dieter. Few days begin with something so good.

2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Andreas Last
Germany
Langen
flag msg tools
mbmbmb
spielstein wrote:
Alast wrote:
Well, I sure am not blindly thrilled AI progress is one thing, brain implants another.


It wasn't aimed at you, Andreas.


I didn't really think so, even though my reaction implies differently Just wanted to add how I think about it.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Stephen Tavener
United Kingdom
London
England
flag msg tools
designer
The overtext below is true.
badge
The overtext above is false.
Avatar
mbmbmbmbmb
Alast wrote:
Yes, in theory a neural network should be able to start learning from scratch. All it needs to know are the rules. The weightings of paths through the network is what it adjusts itself anyway. It would just take so much longer without the DB to get to the same point.


Thinking about it, it's not so impressive - the AI was trained on 175,000,000 positions, which is roughly equivalent to playing 5 million games of chess. I reckon if I played 5,000,000 games of chess, I'd be a lot better than International Master. The article also says that the AI "peaks.. after 72 hours"; so giving it a longer time period wouldn't significantly improve play - this is the upper level of play that can be reached using this technique at present.
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
mocko wrote:
spielstein wrote:
Perfect (as well as random) play is actually the absence of "play".

Well, I join in with the chorus of those who were in no danger of forgetting this, but for the opposite reason: I'd never formed the thought before.

So thank you, Dieter. Few days begin with something so good.


I agree about perfect play based on complete knowledge of the tree, but not quite about random. A game in which all moves are made randomly is not a game, true, but a large number of random games got someone playing with the idea and come up with the Monte Carlo method of evaluating
2 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Andreas Last
Germany
Langen
flag msg tools
mbmbmb
It's impressive in the perspective of the system needing 72h to reach that level while conventional brute force approaches needed years of fine tuning. This system fine tunes itself. Yes, Lai did use a lot of input to get there. And I have no idea how much time he actually spent on building the system. But that the AI can get to that level basically itself by "studying" positions in a similar way as humans would do is pretty impressive to me
 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Russ Williams
Poland
Wrocław
Dolny Śląsk
flag msg tools
designer
badge
Avatar
mbmbmbmbmb
Random abstract ponderings about whether making perfect moves is "play" or not:

1. If by your own brainpower you discover perfect moves each time, it seems uncontroversially like play to me. If you successfully mate my king in a normal game of Chess, finding the only way to deliver checkmate on the final move, then the fact that you found a perfect move doesn't negate the fact that we're playing. So by induction (perhaps this is a "heap of sand" fallacy?), if you successfully find the perfect moves at each previous step before the final checkmate move, isn't it still play? Isn't a primary motivation of playing this kind of game that you want to find as good a move as you can? If you succeed perfectly, that would seem a success in your playing.

2. But if you simply magically know the perfect moves with no real thought on your part (e.g. because a computer tells them to you - which I think is the specific situation Dieter is meaning), then there might STILL be an element of "play" in a different sense, of personal expression, in choosing from among them in positions which have multiple optimal moves. E.g. one player might prefer choosing perfect moves which capture enemy pieces, another player might prefer choosing perfect moves which check the enemy king, etc etc.

3. We're implicitly equating "play" with decision-making, but (semantic tangent) "play" as a concept is arguably a more general nebulous thing. Children "play" decisionless games like Candyland, after all, and consider themselves to be playing.

4. We're perhaps implicitly equating "play" with competing to win, but there are e.g. solo wargamers who play both sides of a 2-player wargame simply to explore the battle or enjoy watching the story unfold, and consider themselves to be playing. (Perverse example: I recently "played" Fire in the Lake, an asymmetric 4-player board wargame which includes manual AI systems for each of the 4 positions. Out of curiosity, in this game I ran all 4 AI players, simply observing the Vietnam War unfold on my table. Was I playing? Yes and no... I think many Abstract forum members would say "No" and many Wargame forum members would say "Yes".)


Still, having said all that philosophical rambling, I do agree that "play" in the normal (for this forum) sense of the word (making decisions in a formal rule framework to try to win) would be at odds with having a computer tell us the perfect moves.
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Néstor Romeral Andrés
Spain
flag msg tools
designer
publisher
Avatar
mbmbmbmbmb
So it will be something like...

Player 1: Human------%---AI (40% human)
Player 2: Human---%------AI (70% human)



1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
christian freeling
Netherlands
flag msg tools
designer
Avatar
mbmb
n_r_a wrote:
So it will be something like...

Player 1: Human------%---AI (40% human)
Player 2: Human---%------AI (70% human)




How about the spectators? ...
1 
 Thumb up
 tip
 Hide
  • [+] Dice rolls
Front Page | Welcome | Contact | Privacy Policy | Terms of Service | Advertise | Support BGG | Feeds RSS
Geekdo, BoardGameGeek, the Geekdo logo, and the BoardGameGeek logo are trademarks of BoardGameGeek, LLC.