|Human Chess players||19||26%|
|Chess Computer Programs||53||74%|
My feeling is this has long been considered by most people to have been resolved that the computers are better than humans at playing chess, but I (just) wanted to see a poll for it.
Which is 'better' though? I'd still say humans. If you study a bit of computer science, you'll come to the depressing conclusion that the current computer is nothing more than glorified abacus.
It can add, subtract, multiply, and sometimes divide (you can divide with a table or via multiplication). It can compare two values, via AND, NOT, OR, and XOR. That's pretty much it, and by the way these values must be in binary.
It is only because it can do these things so fast on value after value that it is able to produce the rich world of computing we know today. It can never "think outside the box", it cannot entertain exotic abstractions or possess any intuition. So for now, I have more "respect" for the human brain than for any current computer system.
> It is only because it can do these things so fast on value after value that it is able to produce the rich world of computing we know today. It can never "think outside the box", it cannot entertain exotic abstractions or possess any intuition. So for now, I have more "respect" for the human brain than for any current computer system.
'Exotic abstractions' though are Speculative and 'Human intuition' is not 100%. Maybe silicon can beat these--at least in play. --probably NOT in Analysis!
> --probably NOT in Analysis!
THAT would take a next-level permanent-hash learning system like IDeA or something right?
> It can never "think outside the box", it cannot entertain exotic abstractions or possess any intuition.
Those aren't required to be a "chess player".
> you'll come to the depressing conclusion that the current computer is nothing more than glorified abacus.
the abacus may be a highly efficient and ingenious instrument compared to who knows what goes on in the human brain.
1. Abstract notions: Computers have to be told things explicitly by humans (e.g. that some simple endgames are draws). Humans can figure out such situations by themselves with general and abstract reasoning. [By the way, this is why computers are absolutely useless (even counterproductive) with some beautiful studies. Computers cannot understand zugzwang, domination, long-term strategy, fortress, etc.]
2. (Un)fairness in a match: computers are allowed to use anything (databases, endgame tables, faster hardware), humans are not. When equalizing conditions of play (e.g. correspondence with access to the same resources as the training of the software, and a fair time ratio), humans have more than a good chance. [The other extreme is to disable all tablebases, opening books, etc, and here too humans can do well.]
3. If I were to ask you who runs faster, an athlete or a car, what would you answer? The car cannot move an inch without the driver (or the engineer who started the automated program). In the same way, the chess engine cannot make even one move without the human who started the computer and the program.
3.5. Finally, computers do NOT play chess, they just compute evaluations, some reasonable, some wrong, some completely off.
I define it as being able to choose a legal move and make it on the board. A way is doing it like humans do, e.g. if the human says "e2e4" and e2e4 is a legal move, the human is playing chess. If the engine says "e2e4" the same standard holds.
If playing chess requires being able to pick the piece and move it, a mechanical arm can be built, no problem.
I don't understand these "the machines aren't really playing chess" argument, what are doing the humans that the machines aren't? having neurons talking to each other, or whatever, isn't in the rules of chess, computers are clearly playing the game.
play (noun): exercise or activity for amusement or recreation
to play (verb): To occupy oneself in amusement, sport, or other recreation
Now show me an engine that takes pleasure in the game, or finds it an amusement or a recreation!
This differs from your restrictive definition of "legal/allowed procedure of conducting a (chess) game within a given set of rules". Let's not forget that this restrictive definition was added later(!) to the definition of "play"; furthermore, the restrictive definition was primarily derived from the basic definition of "play" at a time when amusement and recreation were the principal objectives of games.
> This is the most basic definition of play
The definition you choose to take, to claim that computers can't play chess.
What about all the people that don't feel "stimulated" when they lose their games? Is it that they're not "playing" because they're not amused? What about their opponent that does get the enjoyment, is he playing alone?
It looks like semantics to avoid the fact that you can't look at a pgn:
1. e4 e5 2. Nf3 Nf6 3. Nxe5 d6 4. Nf3 Nxe4 5. d4 d5 6. Bd3
Nc6 7. O-O Be7 8. Re1 Nf6 9. c3 O-O 10. Qb3 Rb8 11. Qc2 Bd6
12. Bg5 h6 13. Bh4 Bf4 14. Nbd2 a6 15. b4 Ra8 16. a4 Qd6
17. b5 Na5 18. Qa2 g5 19. Bg3 Bxg3 20. hxg3 Be6 21. Ne5 c5
22. bxc6 Nxc6 23. Ndf3 Ng4 24. Nxg4 Bxg4 25. Nh2 Be6
26. Qe2 Bd7 27. Ng4 f5 28. Ne3 Rae8 29. Qh5 Ne7 30. Rab1
Bc6 31. a5 Kg7 32. Bxf5 Nxf5 33. Nxf5+ Rxf5 34. Rxe8 Bxe8
35. Qxe8 Rf7 36. Rb6 Qe7 37. Qxe7 Rxe7 38. Rd6 Rc7 39. Rxd5
g4 40. f3 Kf6 41. Rd6+ Kg5 42. f4+ Kf5 43. Rd5+ Ke4 44. Rc5
Rd7 45. Kf2 Kd3 46. d5 Rd6 47. c4 Rd7 48. Ke1 Re7+ 49. Kd1
Rd7 50. Rc8 Ke3 51. Ke1 Kd3 52. Kf2 h5 53. Ke1 Re7+ 54. Kf1
Rd7 55. Kf2 Rd6 56. Rc7 Kd4 57. Ke2 b5 58. Rc6 Rd7 59. cxb5
Rxd5 60. Rxa6 Rxb5 61. Rd6+ Kc5 62. Re6 Kd4 63. Re5 Rb2+
64. Kf1 Rb3 65. Kf2 Ra3 66. Rxh5 Ke4 67. Rg5 Ra1 68. Rxg4
Rxa5 69. Rg7 Ra2+ 70. Kg1 Rd2 71. Kh2 Rf2 72. Rf7 Kd5
73. Rf5+ Kd6 74. g4 Ke6 75. Kg3 Ra2 76. Re5+ Kf7 77. g5 Kg7
78. f5 Kf7 79. Re6 Ra4 80. g6+ Kg7 81. Re7+ Kf6 82. Rf7+
Kg5 83. g7 Rg4+ 84. Kf3 Kh4 85. Re7 Rg3+ 86. Kf2 Rg5 87. f6
Rf5+ 88. Kg1 Rg5 89. Re4+ Rg4 90. Rxg4+ Kxg4 91. g8=Q+ Kf5
92. f7 Ke6 93. f8=Q+ Ke5 94. Qg5+ Kd4 95. Qb4+ Kd3
96. Qgd2# 1-0
And tell if the game was "played", because you don't know if the players had recreation or not.
Was the game played?
You don't know.
With my definition I am absolutely sure that the game was played, so my definition is better.
As for the pgn, it could be a composition as well as a real game, from what I know. On the broader perspective, even if the moves recorded were the result of pieces moved on a board, it still cannot say whether any of the players really played the game, or they were simply making moves, striving to get an advantage of some sort (e.g. ELO points, avoid punishment by who knows who, etc.).
My definition, while not necessarily better than yours, is consistent with my views related to chess.
That is, in your universe, you can only be sure that your games were played, it sounds like solipsism.
And by the way, happy about winning a game is not the same as enjoying the game.
> happy about winning a game is not the same as enjoying the game.
Oh, wow, now someone could be happy about winning a game he didn't play?
I guess you also have special definitions for "happy", "enjoy", and maybe even "is". I give up, I'd rather discuss with someone that uses more common definitions. I'll leave with a xkcd comic.
I am never going out to buy an air conditioner with my sysadmin again.
> I think the cube on the right might be defective. It changes shape!
What if it's a feature? I'd certainly rather have a cube that changes its shape than a normal cube.
>What if it's a feature?
Maybe that's their problem? What's the value of a cube that's not a cube?
And, you broke the xkcdsw chain
> Hmm, I'm not sure... maybe they didn't went shopping for a cube in the first place,
They didn't "went" anything because that's bad grammar.
The dictionary leaves virtually no room to cheat: http://www.thefreedictionary.com/play Go to v.tr. 5.a. "a. To engage in (a game or sport): play hockey; play chess." <---I did not add the chess there...that is in the original.
I think it is silly that we keep redefining what is the most valuable aspect of human thought and human chess based on the order that machines have been able to leave us in the dust. I don't think there is any inevitable order to the way it has proceeded. Hardware could have developed very differently...software too.
If you go back a hundred years it is tactics that were the epitome of human accomplishment in chess. Is it any wonder that computer scientists went there first? And so what happens...we haughtily say "Oh that, that is calculator crap." Then it was technique...Oh look at the human isn't his technique polished converting that small advantage. Well, you know what happened. Computer technique to the yen-yang converting a microscopic edge to a win. So next it is intuition? Well, I am sure that is very doable too. It is just gathering data about lines and openings position similarity and checking the most relevant moves in changing an outcome in the right context. It will have to collect its own data of course but programming the rules should not be that hard. Training may take a few million games, less if it can learn from the same game it is playing in the calculated continuations and use the knowledge immediately.
Glorified abacus...hmm. And a million times slower neuron is so special? Now...which one is more "glorified"?
>Glorified abacus...hmm. And a million times slower neuron is so special? Now...which one is more "glorified"?
I think the 1000 firings times per second (switchings per second from a computer perspective) was from an animal...I think human ones are slightly slower. I think it was an octopus.
If memory serves, our fastest neurons are in our retinas. It has been a few years since I had Physiological Psychology.
>Neurons fire at most around 1000 times a second. A transistor, well much faster. 100Ghz is the current record for a transistor: http://physicsworld.com/cws/article/news/41643 http://jn.physiology.org/content/104/3/1625.abstract
A neuron is really too different from a transistor to make such a comparison work.
Cycles/second does not give a complete picture of performance.
At the moment computers cannot use those graphene transistors as there is no "band gap".
They are both logic parts and work very similarly. You get enough intensity going in and something comes out. If a neuron fires it does so along its axon and everything at the end gets the message. Perhaps it is the equivalent of 10-30 transistors but there is an equivalence somewhere and I doubt it is anything like 1000. In any case the shear speed of a transistor will allow one to do the work of a large number of neurons when collectively coordinated.
And we can't use all our neurons at the same time (unless you count grand mal seizures) or apply them all to chess. A CPU can be fairly fully utilized. Chess though is not hardwired into a CPU. If it were it would be vastly faster.
>Any transistor is vastly faster than a neuron firing rate. And that is not the half of it.
Or the 'any' of it.
>Most neurons have to reach a lot further and the speed of the transmission is vastly slower as well. So slower, longer distance, and slower to get there. And most of those disparities are three orders of magnitude or more.
Yes but each neuron can have over 1000 connections. A transistor is only one, when it is on.
>They are both logic parts and work very similarly.
This is a huge leap.
> You get enough intensity going in and something comes out. If a neuron fires it does so along its axon and everything at the end gets the message. Perhaps it is the equivalent of 10-30 transistors but there is an equivalence somewhere and I doubt it is anything like 1000.
There's that same comparison again.
>In any case the shear speed of a transistor will allow one to do the work of a large number of neurons when collectively coordinated.
This "concept" might be closer to something that we can say for certain. However it is still a long way away. I suspect that there is an exponential increase in "computing" power as the number of neurons increases. Whereas increasing the number of transistors would be linear assuming the same quality of transistors is used.
>And we can't use all our neurons at the same time (unless you count grand mal seizures) or apply them all to chess. A CPU can be fairly fully utilized. Chess though is not hardwired into a CPU. If it were it would be vastly faster.
It's not hardwired onto a brain either. There's also the issue of the computer having to always calculate linearly. Humans can take shortcuts and avoid the necessity of calculation. For now computers cannot come up with such calculation saving strategies on their own.
Speed is not anything? Get real.
"Yes but each neuron can have over 1000 connections. A transistor is only one, when it is on." Irrelevant, there is nothing preventing a thousand leads to go into a transistor or a thousand out of it. There realy is no advantage to it. The neuron and the transistor still do the same thing make an electrical pulse when an input threshold is reached. The neuron does not distinguish between each of the leads and there is just one output. It realy only would need one input if it was in the right location but it has to send out lots because it can't see where they are going. It is just a shotgun approach. One connection might get lucky and do something useful...it is totally random. The transistor on the other hand is intelligently placed so it does not need a thousand random connections. Oh, and obviously if a transistor is considered connected only when "on" the same would apply to the neuron.
The comparison if perfectly justifiable and outlined. Just because a comparison is repeated that in no way illegitimizes it. On the contrary showing more similarities in a comparison strengthens it.
Ascribing exponential increase to neurons and linear increase to transistors is the ludicrous epitome of vanity. What prey tell is you secret source of this preposterous notion.
Sorry but we are wired. Hardwired would suggest that changes can't be made. But changes can be and often are made but changes are not fast and on the fly. For all intents and purposes we are hardwired during any particular game. And only that small part of our brain that has been trained, wired, and connections developed for chess can be successfully employed in a chess game.
Common CPUs are linear but not entirely so in that several things can be ongoing simultaneously. There are multiple pipelines and such and "look ahead" is somewhat non-linear. However nonlinear neural net chips do exist...they are not used much because they require training which would slow down production. Just about anything that a neural net chip can come up with can be simulated successfully using linear processes.
"For now computers cannot come up with such calculation saving strategies on their own." Actually, I think there is at least one neural net based chess program. I think Smart Neural Chess 2.0 is a neural net simulator chess program. I don't know much about it...the website is gone. But I do have a copy on my machine I downloaded a few years ago...very small program 216K including GUI. If it is based on a neural net, I am sure it taught itself a lot. Of course I could be way off...it might not be a neural net at all. Anyone have any info on it? Possibly it was made in 2003?
The arrangement of two sets of different objects says little about how the objects themselves compare. The transistor is simply superior. As for whole brain to chip comparisons; you have to realize that we have a lot of neurons compared to the number of transistors on a chip. Most neurons are digital! On and off. Some are ternary (0, 1, 2) being able to make a double jolt…but it is ludicrous to blow this out of proportion.
“ Do you seriously think any transistor based circuit no matter how advanced can ever match it ?”
Absolutely! Doing the math, the current leading supercomputer is equivalent to 10-30 human brains. It is only our programming that sucks. Of course it uses a whale of a lot more power. Computer technology is not really that old and is progressing very rapidly. When we get true 3-dimensional transistor lattices using low power faster smaller transistors the sky is the limit. right now they can fit about 45,000 x 45,000 on a chip. If it could go 3-d that is 45,000 x 45,000 x 45,000. Even if they can only get say 10,000 in the third dimension it still would make today's chips garbage. And it should be even better than that because large groupings of transistors can be closer together because they can be at levels above or below rather than pushed to the side. And who says chips have to stay so small. A 3-in cube aught to be able to do some major crunching. It might need some liquid cooling tubes through it...that would be cool ;)
“ What about strict determinism of transistor circuits,”
Even Rybka is not deterministic. When more threads are added determinism vanishes.
“ limitation of processing digital information only as opposed to human's free will, adaptability, creativity, self-awareness and consciousness ?”
Whatever it is we do, it is binary and ternary as it is neurons that do it and that is what they are. They are not analogue or base 1000 or anything.
Because of the mythology surrounding each of these terms, (free will, adaptability, creativity, self-awareness and consciousness) a computer can’t fulfill the role. If it is defined behaviorally, then it should not be so difficult.
Creatively, computers have passed humans a long time ago in chess. Tactics are creative; as are ingenious defenses, and all sorts of other chess behaviors they have demonstrated.
“ Chess in my opinion is irrelevant for purpose of comparing human brain vs. transistor circuits. Obviously infinitely fast abacus could solve game of chess yet playing chess is such limited scope of what human mind can achieve that in no way would it make the abacus superior to human brain. In other words: there might be some limitations of deterministic Turing machines. Like for me it is very difficult to imagine how one could create program for creating other programs = program that would produce source code for given specification. For example specification like "create new chess engine for UCI from scratch capable of defeating no.1 player on given hardware" (with just rules of chess and UCI explained,”
The human programmer goes in with more than this. He/she knows about move generators, search, and eval and likely some previous programs. They know what alpha-beta and null move are and such.
“large database of chess theory literature and maybe hundreds of games of no.1 player). I am not sure if this can be done sometime in future with just progression of programming techniques. Somehow I doubt it. Maybe (or maybe not ?) it's similar to how you can not run Shor's algorithm efficiently on transistor chips either.”
Can you run Shor's algorithm on your brain efficiently? ;) I bet an Apple II computer could beat your brain.
Funny how we dismiss chess so easily. I think rather that it is the machine that is not given a fair break. We have already decided that computers are stupid therefore by definition anything they can do well has to be redefined as something trivial even if the computer scientists chose the things we claim are the most intelligent activities that have any kind of objective measurement of performance. I suppose you don’t think much of what Watson did on the Jeopardy! Game show either.
And you should see what genetic algorithms can do. They can invent all sorts of bizarre things that function better than any human engineer could make. It is all about defining the problem, giving it the raw materials and a reasonably fast and accurate automated testing procedure. For chess you would have to have the task broken-up I would think: move generator, search, and evaluation. Using genetic algorithms for move generator should be fairly strait forward. Not sure if it can beat the best human ones but it is worth a try. It might be a struggle at first to generate all and only the legal moves, but eventually maybe it will be the fastest. For test there could be a file with a few million positions and their legal moves. Check might be confusing ;)
Search and eval would be more challenging but not impossible. Again you would need a few million positions that you knew the objective truth about…that is probably the hardest part. But the bar does not have to be quite so high, I suppose. If say there were a million positions with a strong engine’s eval of say 3 minutes on strong hardware, perhaps those evals will be good enough to be a standard to compare to for a instant look for eval. Search is not very friendly for genetic algorithms but still think it could work. Perhaps move in .1 sec compared for that second set of position's evals.
I admit it may be a while before we find the right algorithms for meaning and context. Consciousness/self-awareness that does not sound that difficult to me. That sounds a like a program that can oversee other processes, evaluate progress, project likelihood of success and make adjustments based on those assessments. Though from our point of view we would like it to be able to sense its environment accurately and identify objects and entities. And perhaps anticipate other entities’ behavior and adjust its behavior accordingly.
True, Note that I based my claim on two assumptions: 1.) there is qualitative difference between human brain and transistor based circuits (no matter of internal organization of the circuit). 2.) you can model human brain as "network" of neurons and transistor circuit as "network" of transistors (network as a metaphor of internal organization). If those two assumptions are correct then there should be qualitative difference between neurons and transistors as well. However I am not a neurobiologist I don't know if assumption 2.) corresponds with reality. If it's too big of an oversimplification I still hold with at least my "primary" assumption of human brain vs. circuit qualitative difference (for reasons already explained). On the other hand I know a little bit about perceptron = simplified computer model of neuron and if this model corresponds at least slightly to reality than it is quite puzzling why you would say that Transistor is superior to Neuron. I mean that perceptron has weights that can be changed (e.g. by training) so it can give different outputs for equal input over time - while Transistor doesn't change. So if biological neurons works similarly then saying transistor is superior to neuron is bit like saying apples are superior to cars, isn't it ?
"Even Rybka is not deterministic. When more threads are added determinism vanishes."
As you can see I was referring to transistor circuits being deterministic not that multithreaded algorithms are. If multithreaded applications have some non-determinstic flavor due to how timeslices are allocated to threads by the OS / due to interruption handling caused by user input etc. - then this is nice. But obviously if you put entire system to the exact same state (which can be represented by vector of bits) and then run calculation with the same input vectors of bits you will get the exact same vectors of bits on output consistently and repeatedly each time (unless the circuitry is broken of course). In other words even multicore systems are nothing more than just deterministic Turing machine equivalent. While if you try to apply the same logic on human brain, e.g. like this:
1.) it is possible to store computational state of human brain into vector of bits
2.) for a given state and given input vector of bits you get always the exact same vector of bits on output
you are basically saying that free will is just an illusion or that your entire life is already predetermined or that you are not responsible for you actions and so on. As I don't think this is true this is one of the aspects of qualitative difference I have in mind when comparing brain to circuits.
"Because of the mythology surrounding each of these terms, (free will, adaptability, creativity, self-awareness and consciousness) a computer can’t fulfill the role. If it is defined behaviorally, then it should not be so difficult."
I agree that those terms don't sound too computer scientific or that their definitions are somewhat elusive. I also agree that sometime in future humanoid android satisfying behavioral definitions of (at least some of) those terms could be constructed - like a relatively good approximation of what one would expect from real human. Problem here is that it won't be the "real" thing and it will be not so due to "mythology surrounding each of these terms" but because if nothing else then "free will" is simply not possible on deterministic Turing machine which is direct consequence how deterministic Turing machine is defined. Meaning if this android commits a murder then author of his software is responsible for that, not the android its self. You can argue whether free will of humans is "real" or not. But if you have committed a murder I doubt jury would listen to claims like that your actions were completely predetermined and there was nothing you could do about it.
So this is again the qualitative difference of human brain vs. transistor circuit. By this time I think you can either accept that there has to be at least some principial difference or that free will is just an illusion or option 3) or 4). Like this:
1.) OK there is some difference since concept of free will suggest that human brain can not be simulated on deterministic Turing machine. And since deterministic Turing machine can be at the same time simulated on human brain that means that computational power of human brain (defined by class of problems that it can solve) > computational power of deterministic Turing machine
2.) Free will is indeed an illusion. All human actions are predetermined and therefore simulation of human brain doesn't seem to trivially violate any restrictions implied by definition of deterministic Turing machine. We can at least say it's not impossible that your own brain could be simulated on some of todays supercomputers. We still can not be sure about equality only that computational power of human brain (defined by class of problems that it can solve) >= computational power of deterministic Turing machine. If they were equal that would mean there is no qualitative difference between deterministic Turing machine and human brain and you could indeed download Einstein's brain and run it on your personal computer or cluster for own enjoyment.
3.) You can hardly put = between computational power of human brain and deterministic Turing machine if for nothing else then just because human's tendency of making errors. You are not guaranteed that human simulating Turing machine won't make a mistake and therefore entire simulation is not guaranteed to be deterministic. Accepting 3.) has some severe implications for 1.) and 2.)
4.) all of this is bullsh*t because author obviously doesn't take into account following well know fact xyz... etc... and in addition is very deceptive when he states zyx... etc...
"Creatively, computers have passed humans a long time ago in chess. Tactics are creative; as are ingenious defenses, and all sorts of other chess behaviors they have demonstrated."
This one you can't mean seriously, right ? It's not like deterministic game of chess requires any amount of creativity if you are able to perform full exhaustion of all possible lines to find a best move
(and by coincidence the best move happens to be a fantastic tactical shot). Granted, not all lines - just some "reasonable" guided by heuristics and only to practical depths. But the rules are always mechanical and set by engine author which together with designers of the chip is the only creative force I can see there. Similarly you won't call a fantastic couple of hundred moves long mate found in nalimov bases as a great creativity achievement of computers, right ? It simply emerged from rules of chess and today's computers just have enough power to completely solve game of chess for 6 men endings - with purely "mechanical" algorithmic steps.
"The human programmer goes in with more than this. He/she knows about move generators, search, and eval and likely some previous programs. They know what alpha-beta and null move are and such."
That is exactly why I said "from scratch". It's not like those chess specific algorithms were given to humans by some friendly chess deity. Humans had to invent them from the scratch too which is by the way the kind of creativity I am interested in (as opposed to "creativity" example provided by you). So basically I am asking the program to be able to follow path that was undergone by human computer scientists / chess programmers to invent the algorithms. Starting point should be about very similar to where humans were in this field like 60 years ago or so. If as you say one currently leading supercomputer is equivalent of 10-30 human brains you can take couple of them, equip them with whatever knowledge except that they need to have zero knowledge of chess, chess programming and let's say game theory. Then present the task to them - "create C source code of chess engine capable of winning match with world no.1 player on let's say Core 2 Duo machine" provide them with any chess literature imaginable (but not computer chess literature of course) then wait for some reasonable time like 60 years doesn't matter. At the end expect 2900+ engine on that hardware (which we already know it's possible).
"Can you run Shor's algorithm on your brain efficiently? ;) I bet an Apple II computer could beat your brain."
It doesn't matter if I can or can not. I am basically trying to say there may be other computational paradigms than those defined by deterministic Turing machines. And human brain could provide "hardware" (and software) for one of them. Not necessarily for paradigm A or B just, different. So this is just an example of well defined problem (not clouded by "mythology") -
"perform factorization of large integers in logarithmically polynomial time" which you can not perform on deterministic Turing machine - or it is at least very likely that something like that is not possible.
"I suppose you don’t think much of what Watson did on the Jeopardy! Game show either."
On the contrary, I am quite impressed by the achievement. I can imagine it must have been some incredible amount of hard work behind all of it.
This is for the first time I could accept there is a computer entity with signs of (behavioral) intelligence. Of course it's still in its infancy but I like the direction it's moving into.
I think things like automated translation of texts or interactive natural language query / answer in real time will be possible (for certain types of questions at least) in not "so distant" future.
But if you mean how much of a step forward is Watson in terms of approaching to solution of the problem of "program that creates chess program" then I would think that Watson still deosn't have much relevance. To have some tiny relevance it should be at least capable of successfully answering following question:
If game of chess is played on board of 8x8 squares that in chess notation are numbered left -> right A..H, down -> top 1..8
and chess Knight moves always one square horizontaly and two squares vertically (or vice versa) - but it can not go beyond the border and if Knight is on square B2 list all legal moves of the Knight - squares it can go to - using chess notation. (it could be even more detailed or descriptive to make sure it's clear of ambiguity etc.) I didn't see entire Jeopardy show with Watson but I think he wouldn't have a clue about question like this. Correct ? But even if he did as I said the relevance to solving entire problem of program creating chess program would be very very tiny - almost nonexistent.
"For chess you would have to have the task broken-up I would think: move generator, search, and evaluation"
So you would think ? And would also the program that is creating chess program think it is a good approach ? How exactly would he know it's good ?
And since this post is long enough as it already is I will comment just shortly.
1.) I don't think what you propose explains anything how it could be done
2.) Maybe there is a good reason why I've never heard of genetic algorithms used for chess moves generator. I not only miss the point how you would like to do that but also motivation for it
3.) From scratch = it doesn't mean you have any other engine outputs available. No you don't - see again my note about friendly chess deity. Maybe we could grant you occasional games against strong human players.
You chose 'better' not to be synonymous with 'stronger'. Which is absolutely not true.
Example: The players sit at the chessboard. If an onlooker says, "may the better player win", we know the intent of the statement intuitively and do not race for the thesaurus to attempt to change the meaning.
Well, judging from the responses - some might.
As for computer chess, you can add "faster" as well. These are all different meanings. The fastest engine is not necessarily the strongest, and if the strongest overall is full of bugs, another slightly weaker one might be the better analysis engine, etc.
If I say humans - of course meaning strong humans like GMs - are better (not stronger overall!), I also include the hardware each, into the consideration. The typical (general) computer advantages are 1. speed and 2. precision in terms of data processing and storage. That's why artificial intelligence is superior over natural intelligence now, in chess. If you remove (1) and (2) from a top chess engine, or scale these two elements down to human level, it wouldn't even be average club strength.
If you mean under standard tournament controls then computers still own of course.
> Why do you not mention under which time control ? I think if time approaches zero then computers are better, if time approaches infinity humans are.
> If you mean under standard tournament controls then computers still own of course.
That Is interesting. So where do you think the 50% mark might be?
I do think the closed game is getting better, I say 5-8 years ;) But the greatest remaining weaknesses in my opinion is in recognizing futility, inevitability, fortress positions, and the need for it to recognize what I call a "hinge"...something that has to change for progress to be made from one perspective or the other and direct search for ways to directly address that hinge and applying the progress of that assessment to the value of the position. I do think this is doable. Naum, for instance, shows progress in this area.
I think goal based search is the ticket to closed positions and to some degree the other weaknesses I mentioned. My idea is to take each piece at a time in a position and just deposit it on each square on the board and do a brief tree calculation at each. Two or more pieces can be deposited as well. Assessments that change the outcome strongly to the positive are then back calculated from the drop and forward calculated from the initial position with attempts to connect the start with the positive result position sets in a controlled sequence tree. Extensions that move pieces toward those squares (as measured in lower number of moves and higher number of paths) may accelerate a connection too. Clear barriers to achieving these positive positions are indicators of potential fortress positions and just plain drawn positions. Some deductive logic has to be applied, but it should not be too hard.
There is one more weakness, but I may make a program and don't want to give away my secret. But it is something that does not stand out because humans are weak at it too but a programming approach I have could make a machine very good at it. Humans and other engines would loose regularly without the slightest clue why
Well if they could find better or equal moves to humans in any possible position (under the same time control) than this is basically a definition of complete domination of computers.
If playing better chess means play a practical game that always starts from the same well known position - they should be at least able to completely avoid positions in
which human's are better (under the same time controls). Then it would be still more or less clear that computers are better.
But obviously none of those two conditions are met by modern engines. Today it's more like this:
In 80% of cases computers are clearly better.
In 15% of cases computers are equal
In 4% of cases humans are clearly better
in 1% of cases computers play just completely ridiculous moves or sequence of moves that not even 1500 ELO human player would play
(all numbers just illustrative)
So overall - does it mean computers are better ? I don't know. What I have the biggest problem accepting is how you can label an engine 3200+ while some sub 2000 human player
is able to achieve draw with basically just shuffling the pieces... and engine is unable to come up with anything. (if I am to believe that "father's" games you can find over chess forums are legit)
Like the final positions of one of his games (he played white)
b3r3/4r3/1b2k3/1p1pPp2/p1p2Pp1/P1P1P1Pp/1P1BK2P/1N1B1N2 w - - 0 190
Here the computer score suggest easily won position yet for last X moves both sides were just shuffling the pieces. And not only that but during previous stages of game engine goes for this position / pawn structure showing something as almost gradual improvement of score (and practically all engines of today want to play the same blundering moves) Even if this would be possible only under some specific time conditions and only in one game out of thousand I still find the game quite "disturbing" for the engine. E.g.: drawing like this is worse than losing in a normal game.
Btw: does anyone know how successful could similar anti-computer strategy be in a serious long time control game ?
It is like a football team you don't say the coach played the game even though he made some of the decisions. Or...Is it the horse or the rider that wins the Kentucky Derby.
As of now (2011) computer aided humans (centaurs) are definitely stronger than just computers at any reasonable time control, and this human "know-how" obviously can not be easily added programatically. Otherwise engine authors would have certainly done so resulting in overall much stronger engines. Hence idea of long time control match humans. vs. computer is to reduce advantage of computers ability of calculating variations while boosting importance of human's strategical and planning abilities as humans are still superior in this area (long time controls should help humans avoiding tactical blunders while added think time for engines could prove useless in avoiding positional blunders). How would match like that go I don't know. It's difficult to guess when we even don't know how machine vs. human ELO should be compared.
Edit. also the superior position understanding is only one part of humans "know how"
I was referring to postal chess. Most the players put the positions in the engines and do "analysis". They go through some set of actions involving using their engines.
I doubt centaurs just say hey I know what is best here and just do it...they feed it back into the machine for approval. A machine that has the control of say 10 other machines and just gives each of the top ten moves it sees to ten waiting machines would do the same thing only more thoroughly.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill