Slashdot Mirror
Linux Chess Supercomputer Overpowers Grandmaster
Capt Bubudiu writes "Deep Blue vs. Kasparov is something most readers will remember but when Deep Blue was retired by IBM, a Dubai company took over with Hydra.
In a $150,000 6-game challenge in Wembley UK, the
games got off to a humiliation for mankind as Michael Adams, the
UK Grandmaster, was mauled in games one and three, drawing game two. Adams is ranked seventh
in the world and what ordinary mortals call a 'Super Grandmaster'."
The interesting thing is that in a man vs. machine fight, the tech folks can say "we won" as they assembled the machine. Is it a humiliation or triumph for mankind that it can build a machine that can defeat itself? I think it would rather be a failure for humans if mortals can defeat highly optimized machines.
see a Text Widget
Is it possible that a computer could compute every possible move, make a database of it, and win automatically every time?
... that computers will beat a man at chess all the time they are allowed to use a database on positions.
The time to get scared is when a 'thinking' computer chess program does it all for scratch from the first move.
Having said that, GNUChess 0wn35 me bigtime, the bugger.
go player..
Also a lil info from Wiki on Go and Computers:
Computers and Go
Main article: Computer Go
Although attempts have been made to program computers to play Go, success in that area has been moderate at best - development in this area has not reached the level of Chess programs. Even the strongest programs are no better than an average club player, and would easily be beaten by a strong player even getting a nine-stone handicap. This is attributed to many qualities of the game, including the "optimising" nature of the victory condition, the virtually unlimited placement of each stone, the large board size, the nonlocal nature of the Ko rule, and the high degree of pattern recognition involved. For this reason, many in the field of artificial intelligence consider Go to be a better measure of a computer's capacity for thought than chess.
Use of computer networks to allow humans to meet, discuss games, and play one another, although generally considered inferior to face-to-face play, is becoming much more common. There are servers and software to facilitate this; see Additional Resources below for more information.
It is one of the reasons I hate playing chess online. There are not many, but every now and then I run into someone who is running chessmaster on their computer. They just want to win, they don't want to play the game.
Chess is a relaxing game. It is supposed to be intellectually pure and honest.
It is too bad that Vegas never started with chess tables. That would be something to do, better than blowing money on the slots. There would be no cheating. Maybe in order to play you would have to register with the casinos, and they would keep track of your wins and losses and give you a score so people would know your level.
Rosco: "If brains were gunpowder, Enos couldn't blow his nose."
It can be interesting to see a championship of computers vs. computers, with similar technology but different programming.
To watch a computer defeating a man playing chess is not even interesting anymore, is like trying to do multiplications faster than a calculator (I know some people claim that).
Marcus Hutter's AIXI paper provides a proof that if an agent is a good model for human behavior, and the universe is computable, that the most intelligent program is the smallest program that losslessly compresses the set of observations of the universe.
I've formalized a prize competition based on this criterion as the C-Prize, modeled after the Methusela Mouse Prize. The big difference is that instead of lifespan the metric is intelligence. Here is the currently published C-Prize criteria:
Since all technology prize awards are geared toward solving crucial problems, the most crucial technology prize award of them all would be one that solves the rest of them:
The C-Prize -- A prize that solves the artificial intelligence problem.
The C-Prize award criterion is as follows:
Let anyone submit a program that produces, with no inputs, one of the major natural language corpora as output.
S = size of uncompressed corpus
P = size of program outputting the uncompressed corpus
R = S/P (the compression ratio).
Award monies in a manner similar to the M-Prize:
Previous record ratio: R0
New record ratio: R1=R0+X
Fund contains: $Z at noon GMT on day of new record
Winner receives: $Z * (X/(R0+X))
Compression program and decompression program are made open source.
Explanation A very severe meta-problem with artificial intelligence is the question of how one can define the quality of an artificial intelligence.
Fortunately there is an objective technique for ranking the quality of artificial intelligence:
Kolmogorov Complexity
Kolmogorov Complexity is a mathematically precise formulation of Ockham's Razor, which basically just says "Don't over-simplify or over-complicate things." More formally, the Kolmogorov Complexity of a given bit string is the minimum size of a Turing machine program required to output, with no inputs, the given bit string.
Any set of programs which purport to be the standards of artificial intelligence can be compared by simply comparing their Artificial Intelligence Quality. Their AIQs can be precisely measured as follows:
Take an arbitrarily large corpus of writings sampled from the world wide web. This corpus will establish the equivalent of an IQ test. Give the AIs the task of compressing this corpus into the smallest representation. This representation must be a program that, taking no outside inputs, produces the exact sample it compressed. The AIQ of an AI is simply the ratio of the size of the uncompressed writings to the size of the program that, when executed, produces the uncompressed writings.
In other words, the AIQ is the compression ratio achieved by the AI on the AIQ test.
The reason this works as an AI quality test is that compression requires predictive modeling. If you can predict what someone is going to say, you have modeled their mental processes and by inference have a superset of their mental faculties.
Mechanics The C-Prize is to be modeled after the Methusela Mouse Prize or M-Prize where people make pledges of money to the prize fund. If you would like to help with the set up and/or administration of this prize award similar to the M-Prize let me know by email.
Seastead this.
Or a really powerfull magnet.
But then again, I could put some CN in anyones food and have the same effect.
The differance between a person and a computer is people can learn. A computer can not. I played chess for many years, and I did not get better by reading books or studying past games. I got better by playing.
Chess can never be reduced to a number of possible moves just like art can never be reduced to a number of strokes. God gave us something which seperates us from all other things on earth. We are unlike anything else.
If all a computer can be is logic, I wonder if anyone has found a way to force a shutdown loop, to do something so illogical the computer can not continue.
Rosco: "If brains were gunpowder, Enos couldn't blow his nose."
However, the chess technology in Hydra is 8 years newer in other respects, and so Hydra is able to look about 8 moves further ahead (albeit with slightly less accuracy, but it turns out it's a pretty big win anyway). So Hydra would be expected to comfortably beat Deep Blue, should they ever meet, which is unlikely in fact (although Deep Blue still exists, IBM are hardly likely to boot it up just to lose).
-WolfWithoutAClause
"Gravity is only a theory, not a fact!"How does the human rate on performance/Watt compared to the machine? Isn't that what's important these days?
I rarely post, but I thought this might be worth reminding people of.
While computers are easily tactical masters of chess playing - in that they can immediately anaylze all possible moves availible in a given play, and determine possible outcomes, their fallacy comes in strategy, because, put simply, they don't know how to win.
What is a good move? Is it one that results in a opposing piece's defeat? If so - what value should that piece be assigned? Indeed, what is the value of _any_ piece at any given time on the boards - why should a machine choose one set of perfect moves over another - in almost every way a computer cannot determine the long term value of a move.
This is remedied somewhat by having pre-played game analysis at the disposal of the machine, but in almost every case the computer program requires serious recalibration between matches to prevent a human player from adapting to a strong tactical game. It is by no stretch that computers can be considered inferior in almost every way to a strong human player.
Kasparov posited Advanced Chess as the ultimate play form; the tactical mastery of a computer, mixed with the multilevel strategy of a grandmaster player, making for a game of sublime subtley and perfection.
The two Hydra machines did not even make it into the final sixteen. Moreover, the eventual winners were a couple of amateurs using pretty ordinary PCs running over-the-counter chess programs. On the way to the title they beat a selection of computer- and supergrandmaster-assisted grandmasters.
On this evidence the "strongest chess entity on the planet" is a team consisting of a New Hampshire database administrator + a soccer coach + 3 ordinary PCs.
Links:
Hydra knocked out
Final result
Winners debriefing
In the case of Go, computers perform abysmally compared to humans.
While very high-end computer chess machines now play strong grandmaster chess, it takes relatively little practice to beat the best Go-playing computer.
In chess, "search" is part of the computer algorithm, and it is hard in chess because the tree of possibilities gets big in a hurry.
But in Go it is far worse.
In chess there are (I think) 16 first pawn moves + 4 first knight moves, and the same holds for black --- so that there are 400 possible positions after white and black have moved.
In Go, however, there are 361 possible first moves, and 360 possible second moves; divide by 8 if you wish for rotational symmetries, resulting in some 16,000 different possible positions after the first round of moves.
Humans have a spectacular ability to detect patterns which has yet to be duplicated in machines --- as anyone practicing speech recognition or image understanding can tell you. It is in fact quite remarkable how well the chess machines are doing.
This explanation, the "No matter what it does, I hereby claim it's just a big calculator" type of explanation leaves a number of puzzling problems for its proponents.
1. If the capability is all in the engineers, why don't they beat grandmasters ? (FWIW they don't, but you can go do your own digging if you want to reassure yourself that this is the case). The machine is even capable of correctly playing a (winning) ending where real grandmasters OR engineers would agree a draw. They can't see the way out but the machine can.
2. If I teach a child to play chess very well, is it really -me- who is defeating the child's opponents?
3. Why are you so sure that you are not just an expensive chemical heater which happens to peform calculations fixed by someone else? Just because the illusion is convincing to you doesn't mean anyone else should believe you.
Hofstadter was surprised that we proved able to produce a "dumb" computer that beats grandmasters at chess, but perhaps he shouldn't have been. So far as we can tell the grandmaster is also just a "dumb" computer, albeit a fantastically complicated one, and chess is a pretty specialised sub-function of that grandmaster. So it makes sense that a relatively uncomplicated "dumb" computer can match this one specialised task.
BTW you may wrongly imagine that you're a much more subtle beast than any computer, and that your behaviour is very hard to predict. This isn't true, nearly all human behaviour is boringly predictable, but the person doing the behaving never notices! This suggests that our experience of consciousness is at least partly a lie, we do many things without any control over them and they are retrospectively explained away by the conscious mind.
demolished the field of the international computer rock paper scissors contest with its 6 levels of sicillian reasoning. (I know what you're thinking; but you know that I know what your're thinking; but since I know that you know that I know what you're thinking - I can beat you).
Chess matches between humans and computers are not what they seem- the computer has a huge advantage that amounts to a cheat, specifically, the computer has access to all of its opponent's (the human's) previous games, while the human has no such access to the computer's previous games. Typically, the computer is programmed to play against THAT specific player through analyzing that player's past performances. The human, on the other hand, is not given the computer's games to analyze, they have to show up and play "blind", so to speak. This is a HUGE advantage in favor of the computer. This was the case in the Kasp vs. BB match and I am pretty sure it is still true today. Let the human have access to the computer's previous games, let the human study them and then we will have an even playing field.
Seeing so many posts describing this as a "man versus machine" thing compels me to mention advanced chess, a new form of chess recently proposed by Garry Kasparov. The gist of it is that instead of humans and computers working either alone or against each other, a human player and a computer player team up. Personally, I think competitions like that are great for exploring how humans and computers can achieve a better symbiosis with each other, taking advantage of the strengths of each.
From wikipedia:
Advanced Chess is a relatively new form of chess, first introduced by grandmaster Garry Kasparov, with the objective of a human player and a computer chess program joining forces and competing as a team against other such pairs. Many Advanced Chess proponents have stressed that Advanced Chess has merits in:
* increasing the level of play to heights never before seen in chess;
* producing blunder-free games with the qualities and the beauty of both perfect tactical play and highly meaningful strategic plans;
* giving the viewing audience a remarkable insight into the thought processes of strong human chess players and strong chess computers, and the combination thereof.
The worst part about AI research is that, whenever we manage to do something people consider to be AI, they say that this thing is not "real" AI, that something else a little ahead is "true" AI.
:-)
The above (paraphrased) isn't mine, but I agree fully.
Sigh