I’m talking about Deep Blue the chess computer. Would white always win because it takes the first term? Would all of the matches be stalemates? Would there be a 50/50 chance of white or black winning? Can the probabilities of each event be calculated rather than actually playing the program out for a statistical distribution?
That’s a good question. I bet there’s a factual answer out there; I’m sure its developers have had it “play” itself many times.
The closest thing I was able to find (link) is this, where we see at least that they’ve played different versions against each other. Best I could do.
It wouldn’t launch the nuke.
First it would play several rounds of tic-tac-toe against itself, finally realizing that there are no winners and war is pointless.
[Joshua]Want to play a game[/Joshua]
If the program plays deterministically (i.e., it doesn’t introduce any element of randomness) and it doesn’t use machine learning techniques to adapt its techniques based on past play, then yes, one side would always win (or alternatively, it would always draw). And not only that, but every time it played, it would play the exact same game.
I think this is the ticket. And Deep Blue only adapts with human intervention; in the Kasparov-DB matches, the programmers would adjust DB’s decision-making parameters between games to avoid Kasparov being able to beat it 5000 times in a row playing the exact same game.
Now, I’m not intimately familiar with Deep Blue’s implementation (other than the fact that they “cheated”), but there are superior chess playing programs today. Most of these sorts of algorithms these days do implement some level of randomness and adaptability.
With no randomness or adaptability, the game would be identical each time, and you would know which color would win from the get go.
With no randomness at all, but with adaptability, it would potentially be be slightly different games each time, and it would hill-climb to a local maxima. Eventually, it would converge to behavior similar to the initial situation.
With some randomness,depending on how much, you MAY reach an optimal solution… eventually, if it’s large enough to step off of any local maxima, and small enough to make it not utterly random. Assuming that it is a “fair” game (which I honestly don’t know off the top of my head), it will eventually converge to an optimal strategy for both sides, which would probably result in converging to constant draws.
If you want more info on this, you may want to look at the Alpha-Beta Pruning or Minimax game-playing algorithms. Obviously, these are very basic versions, and the chess implementations involve VERY complex heuristics and massive search depths to determine the scores, but it should give you a basic idea of what’s going on.
Sorry, but minimax and alpha-beta pruning have nothing to do with randomness and adapatibility. I have no idea why you would think that people wanting more info on those two subjects should investigate these two completely different subjects.
I’m not sure how Deep Blue, and similar, are implemented. But when I programmed machine learning algorithms for other board games, to help bootstrap it, i’d play it against itself many, many times so it could learn the ‘correct’ parameters. Obviously, some amount of randomness had to be incorporated, otherwise they’d be playing the same game over and over again.
So it really depends on the program. You can make it run deterministically, or you can add some degree of randomness to it.
I’ve seen Archon play itself, with varying results.
[nitpick] It’s “Shall we play a game?” [/nitpick]
Do you mean the old 80s game, vaguely chess-like, in which the pieces were various fantasy beasties? If so, I loved that game.
It would go blind.
Played it earlier tonight. Unicorns kick ass.
Isn’t it programed to adapt by itself?
How did they “cheat”?
At least at the times it played Kasparov, no. DB’s handlers made adjustments between games; otherwise, having beaten DB once, Kasparov could have beaten it 500 times in a row by playing the exact same game.
Note that he put it in quotes. Kasparov accused Team DB of cheating by making in-game human adjustments - even almost accusing them of having his rival Anatoly Karpov helping out. Nothing was ever proven (how could it be?) and Kasparov came off looking uncharacteristically ungracious.
It would win. Also, lose.
seriously, I could beat Deep Blue with nothing but a single unicorn.
Heck, I coulda taken out the entire Soviet strategic defense system with one unicorn.
Fortunately, it never came to that.