Kuro5hin.org: technology and culture, from the trenches
create account | help/FAQ | contact | links | search | IRC | site news
[ Everything | Diaries | Technology | Science | Culture | Politics | Media | News | Internet | Op-Ed | Fiction | Meta | MLP ]
We need your support: buy an ad | premium membership

[P]
Chess Computers On Track to Overtake Humans in 2004

By John Chamberlain in Culture
Mon Feb 23, 2004 at 02:09:51 AM EST
Tags: Software (all tags)
Software

Over the last few years the rated strength of chess computers has been creeping upwards and is at the verge of overtaking Garry Kasparov, the top-rated human. Jeff Sonas has written an article for ChessBase News detailing these statistical results from the SSDF. They show computer program Shredder at 2812, a mere 19 points away from Kasparov's 2831 FIDE rating. By linear extrapolation the programs seem to be getting stronger at a rate of about 50 points a year so if the trend holds computers will pass Kasparov sometime this year.


Only two humans, Garry Kasparov and Vladimir Kramnik, still seem capable of holding their own against the computers. Kasparov drew a match against Fritz last November and Kramnik drew a similar match against Fritz in October, 2002. As the number of humans who can play evenly against computers dwindles the question is whether computers will soon outstrip all humans.

The human-conqueror may already be here in the guise of a new program called Hydra. Two weeks ago Hydra finished ahead of both Fritz and top-rated Shredder by a comfortable margin at the 2004 Paderborn computer chess tournament. Hydra uses parallel processing and a field programmable gate array (FPGA) daughter board provided by Virtex to achieve its strength. Traditional programs lose search speed when they increase the complexities of their evaluation heuristics in contrast to Hydra's parallel hardware which allows it to use more advanced heuristics with no loss of search speed.

Whether these kinds of advances will overwhelm the last of the humans remains an open question. Jeff Sonas is not convinced they will and has written up some articles giving his reasoning. Today on Developer Diary I also wrote my analysis of the likely outcome. Currently no Man Versus Machine matches are scheduled for 2004, but after the World Computer Chess Championships to be held in July a challenge may emerge.

Will 2004 be the turning point for human versus computer chess? By autumn we may know.

Sponsors

Voxel dot net
o Managed Hosting
o VoxCAST Content Delivery
o Raw Infrastructure

Login

Poll
When computers overtake humans in chess?
o Never 3%
o Eventually 27%
o In a few years 39%
o In 2004 - this is the year of the computers 29%

Votes: 81
Results | Other Polls

Related Links
o article
o statistica l results from the SSDF
o FIDE
o match against Fritz last November
o similar match against Fritz in October, 2002
o Hydra
o 2004 Paderborn computer chess tournament
o Virtex
o some articles
o my analysis of the likely outcome
o World Computer Chess Championships
o Also by John Chamberlain


Display: Sort:
Chess Computers On Track to Overtake Humans in 2004 | 155 comments (149 topical, 6 editorial, 1 hidden)
What did I miss? (2.42 / 7) (#1)
by cestmoi on Sun Feb 22, 2004 at 03:25:53 PM EST

I thought Deep Blue beat Kasparov. Doesn't that count as "passing Kasparov?"

Computers Still Lower Rated than Humans (3.00 / 6) (#2)
by John Chamberlain on Sun Feb 22, 2004 at 03:33:31 PM EST

Yes, Deep Blue won ONE match. But we are talking about rating here, not the results of a single match . Computers are still rated below the top humans.

To put this in tennis terms, if the number 50th in the world upsets Roger Federer that doesn't make him the #1 player. Likewise, because a computer won one match against Kasparov (in which Kasparov played recklessly) does not make the computer higher-rated.

Chess ratings use statistics on ALL the games played by both the computers and humans to rate overall strength.

[ Parent ]

deep blue was a joke (2.25 / 4) (#20)
by omegadan on Sun Feb 22, 2004 at 06:44:09 PM EST

Read and be informed!

Religion is a gateway psychosis. - Dave Foley
[ Parent ]

Not quite sure what the point is. (2.75 / 4) (#35)
by cestmoi on Sun Feb 22, 2004 at 09:49:05 PM EST

Without seeing the documentary, it's hard to make any comment about the film. The article, on the other hand, is atrocious. The article reads like it was written by von Däniken. For example,
"I believe that none of the IBM scientists and chess consultants that I have interviewed and worked with were capable of cheating. They're all good people," says Jayanti. "But there were three different corners in this ring. There was Garry Kasparov, there was the IBM team of scientists and there was a third party: the corporation. And the corporation needed to win at all costs."
If the people operating Deep Blue weren't capable of cheating then how did IBM, the corporation, manage to cheat?

[ Parent ]
Behind Deep Blue, by H. Tsu (none / 1) (#42)
by GenerationY on Mon Feb 23, 2004 at 01:36:56 AM EST

Is a book worth reading for the other side of the story...he has a lot to say about Kasparov's bizarre attacks of paranoia that occured when he was losing. The case seems to center around game logs that revealed Deep Blue's weightings. Kasparov thought they should be handed over to him. Deep Blue's team thoght, I think reasonably, that Kasparov's team would use the logs to analyse directly weaknesses in Deep Blue's game. According to Tsu, Kasparov even tried to psych him out in one game. Which didn't work because of course Tsu was just moving the peieces as directed by the software.

To be honest I doubt IBM cared about losing. As far as I understand it that was the whole appeal of the matches; it was a win-win proposition for IBM who got international coverage for a relatively small sum (they did after all return to fight again after losing quite heavily first time out; I doubt they'd of returned for more punishment if failure was perceived as massively costly to prestige).

[ Parent ]

Favourite deep blue joke... (none / 1) (#56)
by GenerationY on Mon Feb 23, 2004 at 06:24:44 AM EST

http://www.theonion.com/onion3013/chess.html

[ Parent ]
What you missed... (none / 1) (#49)
by ltd on Mon Feb 23, 2004 at 03:34:49 AM EST

What you missed is that this article talks about programs running on PCs vs. dedicated supercomputers like the Deep Blue. Doesn't change the fact, that it's pure crap of course (see my other comments...)

[ Parent ]
Deep Blue was programmed to beat Kasparov (none / 0) (#106)
by damiam on Mon Feb 23, 2004 at 10:30:30 PM EST

Kasparov played badly, and Deep Blue won. Kramnik or most other GMs would have creamed Deep Blue (as would the present-day Kasparov), because Deep Blue was not a general chess computer, it was a Kasparov-beater. Only recently have computers become general enough to stand a chance against most major GMs.

[ Parent ]
An apparently common misconception... (none / 0) (#127)
by beefman on Tue Feb 24, 2004 at 11:51:46 PM EST

Deep Blue *was* a general chess computer.  Nobody knows if Kramnik could have beaten it, but certainly "most other GMs" could not have.

Playing style is important in chess, but only modestly so.  There's no such thing as a player that is stronger than Kasparov but weaker than most other GMs.

Since we don't know how the game ends, we can't rigorously define perfect play in all cases, but we can glimpse it in increasingly large sections of the game, and style is becoming correspondingly less significant.

-Carl

[ Parent ]

Chemical enhancements (2.66 / 6) (#3)
by curien on Sun Feb 22, 2004 at 03:35:02 PM EST

I wonder if the abilty of machines to surpass humans in competitive games will spur the acceptence of chemical human enhancement of competitors.

I mean, seriously, what's the real difference between growing muscle mass with a special diet and growing muscle mass with steroids? Will altering the human body through chemical means become the norm, allowing us to stay ahead of the machines for just a little longer?

Please pass the sappho juice. It is by will alone I set my mind in motion...

--
All God's critters got a place in the choir
Some sing low, some sing higher

You mean like Ritalin or something? (none / 3) (#24)
by Big Sexxy Joe on Sun Feb 22, 2004 at 07:45:43 PM EST

I wonder if the best chess players take that.

I'm like Jesus, only better.
Democracy Now! - your daily, uncensored, corporate-free grassroots news hour
[ Parent ]
what's the difference? (none / 2) (#55)
by Vellmont on Mon Feb 23, 2004 at 05:57:08 AM EST


I  mean, seriously, what's the real difference between growing muscle mass with a special diet and growing muscle mass with steroids?

Cancer, Liver failure, roid rage?

Yah, yah, you're talking about from a cometitive standpoint.  But seriously the problem with steroid use is that they have serious health effects.  If everyone has to use 'roids to be competitive, you'll have a ton of unhealthy athletes that've ruined their bodies just to compete in a sport.  It's a bit different from normal sports injuries.  Pulled muscles and torn ligaments can be healed and you can lead a normal life.  Liver failure will kill you.

The other difference is simply that the special diets don't do a hell of a lot, where the steroids build up a lot of muscle mass.

As to the original question, I think it's only chess geeks that worry about when computers will surpass humans in chess.  Seriously, who cares?  My car goes faster than any human can ever hope to  , but we still have the olympics.  

Once the chess computers can easily trounce the top human, this man vs. machine contest will be over, and no one will care.  The contest will go on to computer vs computer, or human with computer vs computer.

[ Parent ]

Hate to inform you (2.54 / 11) (#4)
by godix on Sun Feb 22, 2004 at 03:46:54 PM EST

... but when computers rank better than all but two humans then it's probably fairly safe to say the computer have already won. These computers can beat 99.99999... of humanity fer christ sake.

It's dawned on me that Zero Tolerance only seems to mean putting extra police in poor, run-down areas, and not in the Stock Exchange.
- Terry Pratchett
Well.. (3.00 / 9) (#5)
by Psychopath on Sun Feb 22, 2004 at 03:51:54 PM EST

..but only 0.0001% of all computers are able to do this ;-)
--
The only antidote to mental suffering is physical pain. -- Karl Marx
[ Parent ]
I don't know about anyone else (3.00 / 9) (#12)
by godix on Sun Feb 22, 2004 at 04:41:17 PM EST

but my home computer can whip my butt 100% of the time. But then again, I'm still trying to remember how the horse thingie moves so I guess that isn't suprising.

It's dawned on me that Zero Tolerance only seems to mean putting extra police in poor, run-down areas, and not in the Stock Exchange.
- Terry Pratchett
[ Parent ]
See below... (none / 2) (#48)
by ltd on Mon Feb 23, 2004 at 03:30:43 AM EST

Please see my other comment... The SSDF ratings and the FIDE ratings are NOT comparable!

[ Parent ]
Shows the weakness of AI and models of human brain (2.71 / 7) (#6)
by sphealey on Sun Feb 22, 2004 at 03:58:26 PM EST

To me, this news simply shows the fundamental weakness in AI research and all existing models of how the human brain works.

When you look at the processing power of a CPU vs. a human brain, computer programs should have been able to outplay humans since 1965 or so. 1975 at best. And yet it has taken 50 years (since 1945) to get to this point. It is as if John Henry were still competitive with a D12 Cat.

And that's just chess. Kasparov also writes political rhetoric and participates in management of international chess (among other things: he probably does a lot more as well). Yet these super-systems are capable of doing one and only one task. Not so impressive in my eyes.

sPh

Actually (none / 3) (#9)
by MechaA on Sun Feb 22, 2004 at 04:29:12 PM EST

Estimates seem to range around a lot but even the more conservative ones seem to agree that the brain is pretty fast, and that computers would be catching up to the memory and processing power of the brain only recently, if at all; certainly not forty years ago.

k24anson on K5: Imagine fifty, sixty year old men and women still playing with their genitals like ten year olds!

[ Parent ]
not really (3.00 / 7) (#14)
by Delirium on Sun Feb 22, 2004 at 04:51:54 PM EST

The brain is very powerful. Estimates range from 10-100 billion neurons, which is one hell of a lot of massive parallelism. It's true that their switching speed is slower than digital computers (on the order of 1ms rather than 1ns), but it's hard to overcome that massive parallelism advantage with raw switching speed.

In any case, I think it's safe to say that computers that are even remotely near the processing power of the brain have only emerged in the last 5-10 years. A computer in 1975 probably was not even computationally equivalent to a cat.

[ Parent ]

It will all be a joke (none / 2) (#111)
by MicroBerto on Tue Feb 24, 2004 at 11:40:37 AM EST

Someday, if civilization lasts that long, mankind will laugh at our feeble attempts at AI! But you gotta start somewhere. You don't just jump into the weightroom and bench 500 on your first day.

Berto
- GAIM: MicroBerto
Bertoline - My comic strip
[ Parent ]
MuaHaHa! -the True Power of your Machine Overlord (none / 0) (#117)
by snowlion on Tue Feb 24, 2004 at 05:03:40 PM EST

Sure, this computer can "only" play chess well.

Next, there will be a computer that can "only" talk well.

Next, there will be a computer that can "only" recognize objects well.

Next, there will be a computer that can "only" walk around, construct maps of places, la dee dah.

Then, someone will connect the AI, over a network, so that the one playing chess well will communicate with the ones that are walking and the ones that are recognizing objects and the ones that are talking. Then the robot will move chess pieces in person.

As people create more and more interactions between intelligent machines, there will be "AI managers." They'll cache algorithms and data locally, from over the network. Say the machine needs to know about, I don't know- how to reason about Topology, or something like that-

This machine will go over the network, look around for the code that has to do with Topology, and then copy it over locally, and setup a space of data.

Or perhaps the interactions between the chess solving computer are "too far away," causing unnecessary slowdown. It'll simply cache it locally, in some open resource space.

Then the AI managers will have AI managers. We will construct intelligences of higher and higher abstraction. The will manipulate the sub-intelligences.

The machine intelligences will be indestinguishable from super-human intelligences.

--
Map Your Thoughts
[ Parent ]

+1, chess (2.62 / 8) (#11)
by debillitatus on Sun Feb 22, 2004 at 04:38:13 PM EST

But I also have some remarks. It's good that we can design AI to beat humans at a game. This shows that our AI-fu is becoming strong.

But I don't think people should read much into this. I mean, we can build a car which goes faster than Michael Johnson, and that doesn't bother anyone. This is the sort of thing which is made for a computer: a clearly defined game with deterministic rules and a finite number of possible states.

Damn you and your daily doubles, you brigand!

on the other hand (3.00 / 6) (#13)
by Delirium on Sun Feb 22, 2004 at 04:49:05 PM EST

Some AI people somewhat cynically describe the AI-critics' definition of "intelligence" as "whatever a computer can't do yet", which I don't think is far off the mark. Every time a computer beats humans at something that used to be considered a mark of intelligence, it's redefined as "not really requiring intelligence."

Perhaps as computers are starting to outstrip doctors in medical diagnosis we'll start to see that redefined as "not requiring intelligence" too? It'll be interesting to see how that turns out.

[ Parent ]

I think the key is creativity (2.83 / 6) (#17)
by curien on Sun Feb 22, 2004 at 05:54:29 PM EST

When people say, "AI", they don't really mean "artificial intelligence", then mean artificial creativity. Some people think that creativity is simply a suitably powerful intelligence with a large enough information base. Some people think creativity is entirely different. Either way, we haven't figured it out yet.

--
All God's critters got a place in the choir
Some sing low, some sing higher
[ Parent ]
I think the key is changing how you think (none / 1) (#25)
by jongleur on Sun Feb 22, 2004 at 07:53:03 PM EST

.. about things. That's my, "what computers can't do yet".

By experience you throw away dead ends, look for predictors of future state etc, to find a thinner, more essential problem than what the various raw rules say. This, finding your own way to simplify the problem, and in the process build your own concepts and understanding, organize it yourself, that's what humans do well that AI hasn't focussed much attention on yet.

And interestingly, our human brains, with the way connection strengths underly our mental activity, we actually in a sense reprogram our own hardware with this understanding! We slowly build custom hardware for the task. Cool, no? (This how I imagine it works anyway (cough); it's plausible at least, I don't know if it's possible to confirm it or whether anyone more authoritative thinks this.)


--
"If you can't imagine a better way let silence bury you" - Midnight Oil
[ Parent ]
Yeah (none / 2) (#29)
by GenerationY on Sun Feb 22, 2004 at 08:53:56 PM EST

its how both the mind (on a cognitive level) and the brain (on a neural level) does work. A gross example of neural reprogramming follows things like stroke; the brain will forge new connections to facilitate relearning and as much as possible "patch around" the damaged tissue. Its more obvious what is going in these extreme cases, but its also the day-to-day workings as well.

On the subject of AI, one approach to the sort of issues you describe (if I understand correctly) is the SOAR architecture, which -- and this is to cut a long story short -- generates its own problems within problems and its own tools for solving them. Kind of.  

[ Parent ]

Link...in case you were interested (none / 2) (#30)
by GenerationY on Sun Feb 22, 2004 at 08:56:13 PM EST

http://sitemaker.umich.edu/soar

and one application:
http://www.soartech.com/htmlonly/technology.tas.html

Nude links. Sorry, too tired to HTML it in properly.

[ Parent ]

For html-ification (none / 1) (#31)
by curien on Sun Feb 22, 2004 at 09:04:17 PM EST

auto-format is your friend.

--
All God's critters got a place in the choir
Some sing low, some sing higher
[ Parent ]
I'll have to look at that (none / 1) (#33)
by jongleur on Sun Feb 22, 2004 at 09:39:18 PM EST

but I was thinking more along the lines of this 1995 paper

A guy (not sure it was this paper exactly) had a program randomly generate predictors (aspects of positions) and refine them with a genetic algorithm. Then it could use those predictors to choose its moves. That's not how I'd want to do it but, those features do redefine the topology of the game for a player, so to speak, and that is the idea I mean.

(BTW curien, autoformat didn't format the url for me...)


--
"If you can't imagine a better way let silence bury you" - Midnight Oil
[ Parent ]
re:autoformat (none / 1) (#61)
by curien on Mon Feb 23, 2004 at 08:08:03 AM EST

Weird, usually it turns anything starting with "http://" into a link.
Testing ... http://www.google.com

--
All God's critters got a place in the choir
Some sing low, some sing higher
[ Parent ]
Well (none / 2) (#22)
by Big Sexxy Joe on Sun Feb 22, 2004 at 07:42:42 PM EST

I think pattern recognition is heart of AI.  I don't think these chess programs really recognize patterns.

I'm like Jesus, only better.
Democracy Now! - your daily, uncensored, corporate-free grassroots news hour
[ Parent ]
There's no contradiction (3.00 / 4) (#23)
by qpt on Sun Feb 22, 2004 at 07:45:38 PM EST

Between a task being a mark of intelligence in a human being yet not requiring intelligence.

Domine Deus, creator coeli et terrae respice humilitatem nostram.
[ Parent ]

AI (none / 0) (#32)
by debillitatus on Sun Feb 22, 2004 at 09:28:51 PM EST

Interesting point. But, honestly, I think the whole debate over "what is intelligence" is doomed. It seems analogous to the debate over whether a computer, or an animal, has a soul.

On the other hand, this debate also seems somewhat pointless. I'm pretty sure that there are things humans can do which machines will never be able to do, and there are things which machines can do that humans can't. We just need to find the right tools for whatever job we need done. We can call them "intelligent", or not. I don't think it matters too much.

Damn you and your daily doubles, you brigand!
[ Parent ]

Except perhaps to understand our own minds (none / 0) (#34)
by jongleur on Sun Feb 22, 2004 at 09:42:23 PM EST

and, maybe, to make possible more efficient ways to produce computers.
--
"If you can't imagine a better way let silence bury you" - Midnight Oil
[ Parent ]
Well, (none / 0) (#39)
by debillitatus on Sun Feb 22, 2004 at 11:33:32 PM EST

The first part, about understanding our own minds, I'll buy that. In fact, the question of what intelligence is seems to be connected to the idea of what mind is in the first place.

On the other hand, I don't see how any of this can impact on the design of computers. No matter what the functional definition of intelligence ends up being, we need concern ourselves here only with what computers can do, and what we can make them do. I don't see how a debate on whether or not these tasks require "intelligence" could help...

Damn you and your daily doubles, you brigand!
[ Parent ]

AI (none / 0) (#76)
by thankyougustad on Mon Feb 23, 2004 at 01:29:04 PM EST

What about a basically simple program whose only purpose is to re-write itself, while at the same time improving itself to be more capable or re-writing itself. . . etc. The idea is that as it continues to do this, it becomes more and more 'intellegent.'

It could be argued, however, that it will only become good at re-writing itself, nothing more. I tend to agree that it's impossible for computers to ever achieve 'intelligence' as we know it, since finally, all they ever are doing is following a set of human written instructions.

No no thanks no
Je n'aime que le bourbon
no no thanks no
c'est une affaire de goût.

[ Parent ]
I've thought a bit about this (none / 1) (#84)
by JahToasted on Mon Feb 23, 2004 at 02:31:23 PM EST

If a programme would rewrite itself, would it improve over time, or would entropy cause it to degenerate?

I'm guessing if its only function were to rewrite itself it would eventually arrive at a state where it is incapable of improving itself. This would either due to its being unable to creatively come up with an improvement (it being limited) or due to no iprovement being mathematically possible (it being complete).

Now if the AI were performing some other task involving input and output, I guess everything would be dependent on that input and output.
______
"I wanna have my kicks before the whole shithouse goes up in flames" -- Jim Morrison
[ Parent ]

Chess on cusp of AI (none / 2) (#38)
by John Chamberlain on Sun Feb 22, 2004 at 10:32:40 PM EST

What is interesting about chess is that it is on the cusp of machine ability. It is simple and linear enough that computers can be good at it, but just hard enough that humans can still beat them at it.

Most of the advances in computer chess nowadays are in the heuristics, the "intelligent" part of the program as opposed to raw computation.

[ Parent ]

Not really leading towards AI (none / 1) (#57)
by zakalwe on Mon Feb 23, 2004 at 07:34:33 AM EST

Most of the advances in computer chess nowadays are in the heuristics, the "intelligent" part of the program as opposed to raw computation.
I don't really think this is leading towards AI though. The reasoning about the heuristics is done by a human - they are a domain specific end product of thought rather than being a step towards a thinking entity. They need intelligence to produce, but that intelligence is supplied by the programmer's knowledge of chess - not the machine's.

[ Parent ]
Right (none / 0) (#67)
by jongleur on Mon Feb 23, 2004 at 10:57:49 AM EST

as in my other comment, here
--
"If you can't imagine a better way let silence bury you" - Midnight Oil
[ Parent ]
Quite the reverse! (none / 0) (#128)
by beefman on Wed Feb 25, 2004 at 12:08:33 AM EST

The fact that humans are still competitive at chess means, in fact, that our AI-fu is hella weak, or that humans are hella better than we thought at doing well at things we should be poor at, or both.

One interesting thing about chess AI is that research investment has been alarmingly undiversified -- almost all effort has been focused into a very narrow selection of techniques.  Evaluation functions based on material and ad hoc expert rules are a horribly ugly way to go about the problem, and the fact that such approaches still rely on opening books shows it isn't a very good one.

The fact that chess admits to "themes" -- key moves or sequences that appear in several nearby places on a pruned tree -- seems to suggest the game might yield to prettier, information-theoretic approaches.  Another fanciful direction might be to use geometry to place bounds on the time it takes to transform bad positions into good ones...

-Carl

[ Parent ]

Unfortunately (2.76 / 17) (#15)
by mcc on Sun Feb 22, 2004 at 05:19:22 PM EST

That's just chess, which has proven to be relatively succeptable to lines of attack such as partially computed game paths and brute-force analyzing of all possible future moves. You just need a really powerful computer and rediculous amounts of memory.

Last I checked, AI still wasn't doing so well at things like, say, Go, where the above strategies don't work so well.

It seems like it isn't AI that's finally beginning to win against human chess players; it's hardware that's beginning to win the battle. There's still some pure heuristics but it seems like the tide has turned because hardware's finally reached the point where algorithms that smack of brute-forcing become not totally infeasible.

But of course this is just my perception and I could be wrong.

---
Aside from that, the absurd meta-wankery of k5er-quoting sigs probably takes the cake. Especially when the quote itself is about k5. -- tsubame

Yes, it's unsatisfying (none / 1) (#21)
by jongleur on Sun Feb 22, 2004 at 07:40:43 PM EST

in fact you could turn it around and marvel that a general purpose human brain can keep up with a parallel, 2004 computer designed with long experience in chess.

It would be more satisfying if the computer could form its own concepts & strategies & otherwise do it more like a human would. Ie, you could turn it loose on other problems and have it figure them out itself.


--
"If you can't imagine a better way let silence bury you" - Midnight Oil
[ Parent ]
No quite... (2.75 / 4) (#47)
by ltd on Mon Feb 23, 2004 at 03:28:34 AM EST

There's still some pure heuristics but it seems like the tide has turned because hardware's finally reached the point where algorithms that smack of brute-forcing become not totally infeasible.

Your perception IS somewhat wrong. It is wrong to characterize top programs as being brute force based with "some pure heuristics." First, you don't have the source code for top programs (except for one, Crafty, which is over 100 points weaker than the best programs). Second, even Crafty contains some rather sophisticated algorithm and chess knowledge found through lots of research and experimentation. Third, take a look at the performance of programs written by very good programmers who are new to the field - if you were right, they'd be even with the best humans, but in fact they are hundreds of points below them and they'd lose 99% of their games. BTW, ridiculous amounts of memory are helpful but not necessary. Lots of disk space for ending tablebases does help though.

Most techniques from chess simply don't transfer well to Go; they are domain specific, and really don't have that much to do with general AI.

[ Parent ]
AI/Hardware (none / 2) (#51)
by harryh on Mon Feb 23, 2004 at 04:42:20 AM EST

It seems like it isn't AI that's finally beginning to win against human chess players; it's hardware that's beginning to win the battle.

What, exactly, is the difference? Some people would say not much. Certainly Kurzweil wouldn't see too much of a difference. Go is just more computationally complex. Give Moore's law another 10 years or so.

[ Parent ]
"Just" (none / 1) (#54)
by warrax on Mon Feb 23, 2004 at 05:53:05 AM EST

Go is just more computationally complex. Give Moore's law another 10 years or so.
I suspect "just" is probably understating it by about a googol orders of magnitude. IIRC one of the current problems with Go is that just scoring the board requires ridiculous amounts of lookahead, whereas chess does not require all that much to arrive at an approximate value of the table. To make matters worse, it is much harder to do search path pruning in Go than it is in chess, so each extra move of lookahead adds many more potential outcomes.

Roughly speaking, the complexity of chess is a^b and complexity of Go is (probably) c^d, but c>>a and d>>b. I don't think 10 years of increases in computing power is gonna do it.

(Btw, Moore's law does not say anything about computing power. There may be correlations, but... Besides, who knows whether it will still hold 10 years from now?)

-- "Guns don't kill people. I kill people."
[ Parent ]

OT: Googol orders of magnitude (none / 1) (#58)
by Cameleon on Mon Feb 23, 2004 at 07:53:17 AM EST

...by about a googol orders of magnitude.

So go is 10^10^100 times as complex as chess? I don't think you meant that; you probably wanted 100 orders of magnitude, or googol times (which is still a lot, btw). 10^10^100 is called a googolplex, btw.

[ Parent ]

I meant it, but... (none / 1) (#94)
by warrax on Mon Feb 23, 2004 at 05:36:17 PM EST

it was sort of mostly tongue-in-cheek, as in "very very very very very [...] very much more complex". I just forgot to put in a smiley.

-- "Guns don't kill people. I kill people."
[ Parent ]
Moore's law won't help us. (3.00 / 6) (#73)
by awgsilyari on Mon Feb 23, 2004 at 12:39:40 PM EST

Give Moore's law another 10 years or so.

You're pitting Moore's law, an exponential growth function, against another exponential growth function: Go's branching factor.

Moore's law gets stomped:

2^(10/1.5) = computers about 100 times faster in 10 years.

Average branching factor in Go = 200.

Thus, in ten years, machines still won't be fast enough to search Go even a single move deeper than they can now.

We are in the realm of extremely difficult problems here. Moore's law may be exponential, but so is Go.

--------
Please direct SPAM to john@neuralnw.com
[ Parent ]

Um.. (2.80 / 5) (#72)
by awgsilyari on Mon Feb 23, 2004 at 12:24:03 PM EST

That's just chess, which has proven to be relatively succeptable to lines of attack such as partially computed game paths and brute-force analyzing of all possible future moves.

There's no game which isn't "susceptible" in that way. The logical extension of the "brute force" principle is to brute force all the way to the end game. Any zero-sum game can be "solved" in this way. It's a matter of compute time. Chess is no more or less "susceptible" to it than any other zero-sum game; it just happens to have a low enough branching factor that we can search deep enough to do well against most human opponents.

Last I checked, AI still wasn't doing so well at things like, say, Go, where the above strategies don't work so well.

The only reason the strategy doesn't work for Go is because of the enormous branching factor. It's a matter of compute time. If we suddenly had access to advanced alien computer technology, we could very simply apply minimax search to Go. There is no problem with the algorithm. It just takes too freaking long.

There's still some pure heuristics but it seems like the tide has turned because hardware's finally reached the point where algorithms that smack of brute-forcing become not totally infeasible.

Not right. The time it takes to search a minimax game tree is exponential in depth. Assume a mid game of chess has a branching factor of about 8. That means 64 possibilities per ply. Thus, in order to search only a single ply deeper in the same amount of time we have to make the machine 64 times faster. This is not going to cut it.

It's quite irritating, when writing minimax game playing programs, when you realize that if you quadruple the speed of the evaluation function, you don't get any deeper whatsoever in the game tree. The major advances in minimax-based systems come in the form of pruning gigantic subtrees from the search -- not from simply turning up the speed dial.

--------
Please direct SPAM to john@neuralnw.com
[ Parent ]

AI (2.83 / 6) (#87)
by andersjm on Mon Feb 23, 2004 at 03:21:36 PM EST

Back in the younger days of AI in the 1970's, when AI researchers first set out to build rule-based systems, people were saying that that wasn't really AI. These systems were not really intelligent, because they were just a set of fixed rules. With no creativity and no real insight, these systems couldn't do anything that requires real intelligence, like, um, to play a good game of chess.

Nice try, AI community, but come back when your precious "AI" can beat a grandmaster at chess!

Well now computers can beat a grandmaster at chess.

Time to move the goalposts: Obviously, since computers can do it, and we know how computers work, we can't really believe that playing chess requires intelligence.

Whatever computers become capable of, we will have programmed them to do it, we will understand how it's done, and people will refuse to call it 'intelligent'.

So go ahead, set your standard for what computers should be able to do before you will call them intelligent. Rest assured that in thirty years time when your goal has been reached, you will have found a new standard and you will still be saying that computers are not truly intelligent. They only appear to be.


[ Parent ]

AI (none / 2) (#97)
by gdanjo on Mon Feb 23, 2004 at 06:01:02 PM EST

Back in the younger days of AI in the 1970's, when AI researchers first set out to build rule-based systems, people were saying that that wasn't really AI. These systems were not really intelligent, because they were just a set of fixed rules. With no creativity and no real insight, these systems couldn't do anything that requires real intelligence, like, um, to play a good game of chess.

[...]

So go ahead, set your standard for what computers should be able to do before you will call them intelligent. Rest assured that in thirty years time when your goal has been reached, you will have found a new standard and you will still be saying that computers are not truly intelligent. They only appear to be.

And what's wrong with that? If we had stopped in the 70's and declared the computer intelligent, that would have helped no-one (except the researchers who would have been placed in the history books).

And your argument makes the fatal assumption that intelligence in humans somehow stays still while the computer's increases. But what if our intelligence is increasing as we build the machine's intelligence? We could have called the 70's computer intelligent, but then we'd have to make a new word for the new human intelligence which the computer should aim for - why not just call it "intelligence"?

When the computer overtakes us in the intelligence department, only then will they be considered intelligent. Then I'll proudly proclaim that computers are no good at art, and give them another challenge - now that's (cunning) intelligence.

Dan ...
"Death - oh! fair and `guiling copesmate Death!
Be not a malais'd beggar; claim this bloody jester!"
-ToT
[ Parent ]

Intelligence is a contest to you, it seems. -nt (none / 0) (#138)
by andersjm on Wed Feb 25, 2004 at 03:49:10 PM EST



[ Parent ]
Life is a contest -nt (none / 0) (#139)
by gdanjo on Wed Feb 25, 2004 at 05:46:12 PM EST


"Death - oh! fair and `guiling copesmate Death!
Be not a malais'd beggar; claim this bloody jester!"
-ToT
[ Parent ]
maybe (none / 3) (#16)
by cronian on Sun Feb 22, 2004 at 05:39:40 PM EST

I think Chessbase had an article a while back that pointed that while comptuers are improving, the World's top players are also improving. They looked at the rate of improvement, and claimed that humans were improving faster. Regardless of whether this is actually true or not, there is no reason to assume computers will definitively overtake humans. It could go back and forth between humans and computers over who is the best.

We perfect it; Congress kills it; They make it; We Import it; It must be anti-Americanism
that article series is linked above (none / 2) (#18)
by John Chamberlain on Sun Feb 22, 2004 at 06:05:04 PM EST

Exactly, that is the 3-part series of articles written by Jeff Sonas that is linked in the feature text above.

[ Parent ]
ohh (none / 1) (#19)
by cronian on Sun Feb 22, 2004 at 06:11:34 PM EST

I guess it wasn't all that clear in the write-up, and I never clicked the link. Oh well.

We perfect it; Congress kills it; They make it; We Import it; It must be anti-Americanism
[ Parent ]
-1: A few years too late (2.00 / 5) (#28)
by Lelon on Sun Feb 22, 2004 at 08:26:53 PM EST

I'm sorry but this is not only old news but it is old news that has been repeated over and over. Kasparov lost to Deep Blue fair and square. Granted, current chess computers are smarter, but they are much slower then the custom-designed chess processors that Deep Blue used. They deserve praise for being nearly as competitive as Deep Blue with much less processing power, but if you were to combine today's software with that hardware Kasparov would have even less of a chance then when he lost.


----
This sig is a work in progress.
if you had actually read the article... (none / 2) (#36)
by John Chamberlain on Sun Feb 22, 2004 at 09:49:23 PM EST

If you had actually read the article you would know that Kasparov is still higher rated than the top computer program (Shredder). The news is that there is the possibility in the near future that the programs may surpass the top human in ELO rating--something they have not done before.

The current crop of top chess programs are widely considered to be stronger than Deep Blue. There evaluation algorithms are much better and they calculate around the same magnitude of positions per second (taking into account that Deep Blue's architecture had redundancy that reduced its real positions per second from a nominal 200 million to a real <10 million). <p> Kasparov's match loss in the mid-1990s was an isolated upset in which he played recklessly and both he and Kramnik have shown since then that the top programs still have a way to go to prove that they are better.

[ Parent ]

Publicity Stunt (none / 0) (#74)
by hardburn on Mon Feb 23, 2004 at 01:11:10 PM EST

Deep Blue was an IBM publicity stunt. IBM set rules which were more favoriable to Deep Blue (like not allowing Kasparov to view Deep Blue's games, but Deep Blue got fed plenty of Kasparov's old games) which aren't typical for a sanctioned tournament match. Further, Deep Blue was built specifically to beat Kasparov's style. While it could beat Kasparov, Deep Blue may have been completely destoryed by another Grand Master with a different playing style (though we'll never know for sure, since IBM scrapped the thing after the news reports died down).

Chess programs since then have gotten much more generalized and have competed with Grand Masters using rules closer to actual tournaments.


----
while($story = K5::Story->new()) { $story->vote(-1) if($story->section() == $POLITICS); }


[ Parent ]
sort-of (none / 0) (#126)
by beefman on Tue Feb 24, 2004 at 11:40:00 PM EST

It was a publicity stunt, and beating Kasparov was the goal, but there's only so much you can do as far as programming a chess computer like Deep Blue to beat a specific opponent, which is not much.

The important thing about the 1997 match is that it was only 6 games, which is not very many.  And in those six games it can well be argued that Kasparov outplayed Deep Blue.  The match was tied until Kasparov lost by a well-known move-order blunder in the opening.  In other words, the deciding game was of very poor quality and gives us very little information about the relative playing strength of the contestants.  I was there and Kasparov recognized the blunder the instant he let go of the piece.

You might say that this reveals an important weakness in human players.  But then I would argue that Deep Blue's chess engine had an unfair advantage in that it had access to an opening database while Kasparov was not allowed the same.

-Carl

[ Parent ]

They both had access to opening databases... (none / 0) (#149)
by irrevenant on Fri Feb 27, 2004 at 05:32:41 AM EST

...Deep Blue just remembered them better than Kasparov did.

[ Parent ]
Hmm (1.42 / 7) (#41)
by ShiftyStoner on Sun Feb 22, 2004 at 11:55:30 PM EST

 I want to play this Shredder. They should allow people to play against it online.
>
( @ )'( @ ) The broad masses of a population are more amenable to the appeal of rhetoric than to any other force. - Adolf Hitler
you can play Shredder (none / 0) (#70)
by John Chamberlain on Mon Feb 23, 2004 at 11:06:11 AM EST

You can play Shredder. ChessBase sells the exact program they use competitively (of course it runs on powerful PCs when it competes). The link to their shop is: http://www.chessbase.com/shop/

Shredder costs $55

[ Parent ]

heh (none / 0) (#88)
by Battle Troll on Mon Feb 23, 2004 at 03:35:20 PM EST

Make sure you're lit.
--
Skarphedinn was carrying the axe with which he had killed Thrainn Sigfusson and which he called 'Battle Troll.'
Njal's Saga, ca 1280 AD
[ Parent ]
Is that so? (1.40 / 5) (#43)
by Tyler Durden on Mon Feb 23, 2004 at 02:06:02 AM EST

Well, Shredder may be able to defeat Gary Kasparov, but it will never defeat Splinter!  Turtle Power!!!

On a more serious note.. computers that can play chess are pretty nifty, but far from intelligent.  A world champion chess machine is no smarter than a toaster or a coffee maker.  It does exactly what it's been manufactured to do.  You can make it complicated as you want to.. adding randomness, probabilities, some form of long term memory... but it still only does what it's made to do.

The whole problem with AI is that it's all looked at as some sort of problem that a person can solve.  But it's not.  A true AI will be emergent.  Humans will have to provide the right conditions, like computational resources, sensors, etc.

A person cannot be created by taking a bunch of atoms and smashing them together... neither can AI be generated by assembling a bunch of computer instructions.

Jesus Christ, EVERYONE is a troll here at k5, even the editors, even rusty! -- LilDebbie

In all honesty (none / 1) (#75)
by mcc on Mon Feb 23, 2004 at 01:21:47 PM EST

A world champion chess machine is no smarter than a toaster or a coffee maker.  It does exactly what it's been manufactured to do.

Some of the upper-level human players don't seem to be all that different.

[ Parent ]

That's out there... (none / 1) (#101)
by Tyler Durden on Mon Feb 23, 2004 at 09:00:45 PM EST

Genius and insanity often come as a package deal though.  Sort of the Universe's crazy sense of irony -- or balance.  

Jesus Christ, EVERYONE is a troll here at k5, even the editors, even rusty! -- LilDebbie
[ Parent ]

Not so odd.... (none / 0) (#148)
by irrevenant on Fri Feb 27, 2004 at 05:02:18 AM EST

...given that a key indicator of genius is to hit upon ideas that no-one's come up with before.  Insane people, by definition, think in a different way to 'normal' people, so it's no real surprise that insane people are more innovative.

Of course, VERY insane people are often unable to function sufficiently in society (without drugs or other treatment) to implement their inspirations, so the great geniuses tend to be those on 'the edge of chaos' - insane enough to 'think outside the box', but sane enough to apply those thoughts...

[ Parent ]

What is intelligent? (none / 0) (#136)
by edremy on Wed Feb 25, 2004 at 11:56:28 AM EST

On a more serious note.. computers that can play chess are pretty nifty, but far from intelligent.

Schaffer (of Chinook fame) would rather disagree. Defining "intelligence" is tricky- everytime computers start beating humans it's all of a sudden not something that requires intelligence.

Does winning at chess, checkers, scrabble or similar games require intelligence? We used to think so, but now computers can beat any human at these games. Doing differential equations? Done. Designing electrical circuits from scratch? Done. How about seeing and moving around without human intervention? Getting there. Driving? Some of the DARPA Grand Challenge entries are pretty impressive.

At what point do you simply have to fall back to a Turing text-like system and say "It's doing something that looks intelligent. Therefore it is"? Otherwise we might find a whole lot of human experience doesn't really need intelligence after all.

[ Parent ]

Intelligence (none / 1) (#143)
by Tyler Durden on Thu Feb 26, 2004 at 08:33:27 PM EST

I've heard that argument, but very little progress has actually been made in AI over the years.  There has been a lot of work done.. we've learned a lot, but we're not really any closer to creating a true AI (IMHO).  

Chess, for all it's mystique, is ultimately a deterministic game.  There are a countable number of possible moves.  With current resources, a brute force attack is still not quite feasible, but before long it will be.  Go is the same thing.  A lot of people want to jump up and down about how Go is so much harder.. but ultimately, there are a finite number of moves there as well.  

The problem is that actually doing the things: solving differential equations, playing chess or checkers or whatever, is not where the intelligence is.  The only thing required is to follow a set of rules.  

Does driving a car require intelligence?  There are systems that can do so in a limited fashion already.  Well, it seems unlikely that dogs could drive a car (even if they were physically able).  Does this mean machines are smarter than dogs?  what about pigs, monkeys, doplhins?  

I think real intelligence is being able to reason about things.  To be able to look outside the system.  To construct new systems.  Understanding concepts like cause and effect, or infinity.  A program to play chess, plays chess.  It cannot operate outside of that tiny system.  It can't play checkers, it can't play poker.  It plays chess.  And just like we cannot grasp what may be beyond the universe, the chess program cannot grasp anything beyond its universe.  

I think the Turing test is still the best way to prove machine intelligence.  But I think there is a distinction between a program that manipulates language in an attempt to fool a person, and an artificial consciousness that actually understands the dialog.

... we might find a whole lot of human experience doesn't really need intelligence after all.

I think that's true.  Most of what we do does not require a lot of intelligence.  We humans like to wallow in our superiority, but I doubt we're as high up the ladder as we like to imagine.

Jesus Christ, EVERYONE is a troll here at k5, even the editors, even rusty! -- LilDebbie
[ Parent ]

Pattern Matching (none / 0) (#147)
by irrevenant on Fri Feb 27, 2004 at 04:58:54 AM EST

I think most people agree that knowledge is not intelligence - that intelligence is the ability to learn and interpret.

Essentially, what the human brain is very good at is pattern matching.  A lot of what we call 'intelligence' ie. the flexibility to interpret and adapt to new circumstances is actually sufficiently advanced pattern matching to find similarities to previous experiences.

[ Parent ]

+1FP, is indeed a landmark (1.80 / 5) (#44)
by rtmyers on Mon Feb 23, 2004 at 02:09:40 AM EST

Deep Blue may have beaten Kasparov in a match, but computer programs overtaking him in terms of FIDE rating is a genuine milestone. This article is short and to the point.  

Yeah, we all know go is real hard and computers suck at it. And it'd be great to have an article on why that is. But that would be a different article.  

However, I would like another sentence or paragraph about Hydra. Is it the use of FPGA's or of parallel processing that allows it to use "advanced heuristics with no loss of speed"?  Parallel processing has its own bottlenecks and limitations; is Hydra doing something interesting to get around them?

--
Bob Myers

-1 (none / 1) (#50)
by ltd on Mon Feb 23, 2004 at 03:41:49 AM EST

Deep Blue may have beaten Kasparov in a match, but computer programs overtaking him in terms of FIDE rating is a genuine milestone. This article is short and to the point.

Except they aren't. The computers don't play in the FIDE sanctioned events.

[ Parent ]
Why go is hard (none / 2) (#59)
by cca93014 on Mon Feb 23, 2004 at 07:54:40 AM EST

Go has a much larger search space. That's pretty much it.

[ Parent ]
That and... (none / 2) (#64)
by mikera7 on Mon Feb 23, 2004 at 10:50:37 AM EST

...the fact that no-one's come up with a vaguely decent position evaluation function.

Having written a simple Go playing program a while back, I can tell you that this proved to be a much bigger headache than managing the search space.

[ Parent ]

.. which, with our search-reducing strategy (none / 0) (#69)
by jongleur on Mon Feb 23, 2004 at 11:02:18 AM EST

we do better at. We reduce the original search space to one reorganized around significant features learned by experience.
--
"If you can't imagine a better way let silence bury you" - Midnight Oil
[ Parent ]
unfortunatly (none / 1) (#80)
by garlic on Mon Feb 23, 2004 at 02:01:25 PM EST

The hydra site is more interested in talking about the source of the name and that they're from the UAE then how they do what they do. I also don't believe the link to the FPGA site is correct. Virtex is an FPGA made by Xilinx, but that virtex site looks like it has nothing to do with FPGAs at all.

HUSI challenge: post 4 troll diaries on husi without being outed as a Kuron, or having the diaries deleted or moved by admins.
[ Parent ]

This is completely flawed! (2.83 / 12) (#45)
by ltd on Mon Feb 23, 2004 at 03:10:08 AM EST

"Jeff Sonas has written an article for ChessBase News detailing these statistical results from the SSDF. They show computer program Shredder at 2812, a mere 19 points away from Kasparov's 2831 FIDE rating."

No, no, no! The FIDE and the SSDF are composed of two completely non-intersecting sets of players! If you know anything about how the ELO rating system works, you know that you cannot compare ratings between them! If you don't believe me, how about this fact: a couple years ago, the SSDF arbitrarily lowered everyone's rating by 100 points because the ratings started to "appear" to be too high. The initial ratings of the participating computers were supposed to be based on their games with humans but that was many years ago (15+) with different software on completely different hardware.

Slightly related (none / 2) (#60)
by Cameleon on Mon Feb 23, 2004 at 08:01:41 AM EST

The lists are slightly related, because, as you point out, the SSDF tries to maintain a 'realistic' relation between the two ratings. As pointed out in the third part of the articles by Jeff Sonas that are linked to, the SSDF ratings were related to human ratings by comparing them to games between computers and Swedish players. The lowering by 100 points was the result of comparing the computer ratings with matches played between players and computers worldwide. See also this quote:

To summarize, before the correction, in early 2000, the SSDF ratings were still accurate in how they ranked computers against each other, but the actual rating numbers were too high, across the board. Those numbers ultimately were coming from a few hundred games played against Swedish players in 1987-1991. And it was becoming too much of a stretch to extrapolate forward from games played by the top Mephisto and Fidelity computers, on 68020 processors against lower-rated humans, a dozen years in the past. For one thing, there was no allowance for the fact that human players had gotten objectively stronger, or had learned to play better against computers, since 1991.

So, at that point, about 100 games were analyzed from events between humans and computers in 1997-2000. The humans in those games had an average FIDE rating below 2400; the only two events against really strong humans were Junior at Dortmund 2000, and Fritz at the Dutch Championships in 2000. Thoralf Karlsson also had to make some assumptions about the impact of different hardware, since the hardware used by Junior and Fritz in those events (for example) was different from that used by the SSDF. The conclusion from all of this analysis was that all SSDF ratings should be reduced by 100 points. There have been no further corrections since then.

So although both ratings are not directly comparable, there is some relation between them, and comparing them is not completely without merit.

[ Parent ]
Good point, but SSDF is relevant (none / 2) (#68)
by John Chamberlain on Mon Feb 23, 2004 at 10:59:21 AM EST

Good point, the SSDF rating is certainly not the same as a FIDE rating, but currently FIDE does not rate computer vs humans so it is left to outside organizations.

There is a concensus that the SSDF ratings are in the right ballpark for the computers and the results are consistent for their ratings. So, until FIDE rates computers the SSDF is the best we have.

[ Parent ]

Humans always win (2.50 / 6) (#46)
by smilinggoat on Mon Feb 23, 2004 at 03:20:08 AM EST

Humanity isn't winning, but it isn't losing either.

Good point from the article (although taken completely out of context). These programs are programmed by humans. It's just a different way of playing chess. Think of it like computer music vs. more traditional acoustic music. There's some pretty cool computer music out there that can play things more accurately than any human can. But computers don't know how to get drunk on whiskey and howl into the night on mandolins and guitars or trip out to some real mellow sitar music. To quote Mojo Nixon, "Drum machines don't get drunk, start up a fight, sneak in round the backdoor, try to screw your wife." It's a different kind of game.

I believe computers will be able to beat all humans in real time chess games this year. 2k4 is just getting started, we have a long way to go and a alot of numbers to crunch. Hell, software itself is still in its infancy. Programming gets cooler and more powerful every day, it's only a matter of time...


~
Pure Data, where music looks less and less like itself.
That's not quite true... (none / 0) (#121)
by skyknight on Tue Feb 24, 2004 at 09:09:50 PM EST

These programs are programmed by humans.

Well, that's only sort of true, and under what I would deem a bad definition. Their initial setup is put in place by humans, but the machines actually learn things by playing, and develop constructs that the human programmers did not directly put into them. I don't think that this is all that different from the tabula rasa that is a newborn human baby. In the case of these chess playing programs, humans are only bootstrapping them (though possibly tweaking them, too).



It's not much fun at the top. I envy the common people, their hearty meals and Bruce Springsteen and voting. --SIGNOR SPAGHETTI
[ Parent ]
Im surprised it's taken so long (1.80 / 5) (#52)
by nebbish on Mon Feb 23, 2004 at 05:36:50 AM EST

Although chess is obviuously a very complicated game, it is governed by strict rules and possibilities and can be reduced to a (very complicated) mathematical problem. If computers can't do this yet, they're much less advanced than I thought.

---------
Kicking someone in the head is like punching them in the foot - Bruce Lee

Computers are not advanced (none / 2) (#62)
by JyZude on Mon Feb 23, 2004 at 10:09:35 AM EST

Computers are very simple things. They add, subtract, multiply, and divide. They transform input to output. They're just calculators with stateful memory.

And chess is governed by strict rules, and is actually a very simple game, mathematically. But, when you compose the game of chess as a math problem, you find that it's a much bigger problem than you expected. Though the rules of each round are simple, they must be applied to each move. Plus, you have to predict up to 25 moves ahead for each move you plan to do.

You can't brute-force chess. So what can you do? Make the computer think. How do you make a calculator think? Well, there's been years of research into it...

In the 60's they predicted a computer world chess champion by the 70's, in the 70's they said it would be the 80's. In the 80's they predicted the 90's. Every time they got a bit further, they discovered that the problem was bigger than they expected, and that the human mind was doing some extremely complicated stuff to play chess.

-----
k5 is not the new Adequacy k thnx bye


[ Parent ]
It's the computational complexity (none / 3) (#65)
by Eccles on Mon Feb 23, 2004 at 10:54:36 AM EST

It is quite possible to write a program that, in theory, would "solve" chess -- that is, it would figure out all possible moves and counter-moves, and figure out the "perfect", unbeatable game.  However, there's no computer that now exists that could run it to completion, and even one that could analyze a trillion positions a second would take longer than the lifetime of the universe to run.

And for a computer, Go makes Chess seem simple.

[ Parent ]

Just try it (2.00 / 4) (#71)
by hardburn on Mon Feb 23, 2004 at 11:32:11 AM EST

I once tried to make a series of subroutines that would validate a chess move (this was for purely personal education, and I didn't look up any of the liturature on the subject).

There are a lot of special cases, especially for pawns. It has to move one direction if it's black, and another direction if it's white. It can move two spaces initially and one space for every other move. It only moves forward, except if it's attacking, in which case it must move diagonal.

For any peice, you have to make sure it's not landing on a square occupied by your own peice, and no peices can jump any other peice except knights.

In my own study (which may not line up with what is in the various litature), I eventually found the easiest solution was to compute all possible moves for a peice and determine if it's moving onto one of them.

With all these special cases, I can see why chess is such a difficult problem for computers.


----
while($story = K5::Story->new()) { $story->vote(-1) if($story->section() == $POLITICS); }


[ Parent ]
Wow. (1.60 / 5) (#77)
by tkatchev on Mon Feb 23, 2004 at 01:36:29 PM EST

You're a bad programmer. Where did you study?

   -- Signed, Lev Andropoff, cosmonaut.
[ Parent ]

no he's not (none / 0) (#81)
by muyuubyou on Mon Feb 23, 2004 at 02:02:54 PM EST

...at least for what he said. Doing even the simplest operation with a chess position takes a lot of cycles.

Getting "millions of positions per second" without special hardware is still impossible today's home PC hardware if you're doing anything interesting with each of them.

[ Parent ]

Why the hell would you bother (none / 1) (#96)
by ninja rmg on Mon Feb 23, 2004 at 05:45:28 PM EST

With something like what he is describing? You might as well just construct all the possible moves and check if a given move is on the list. If you look at chess programs, you'll see that is all they ever need to do.



[ Parent ]
Learn the big "Oh" notation. (none / 2) (#112)
by tkatchev on Tue Feb 24, 2004 at 11:48:10 AM EST

American universities love it.

The problem isn't in the number of cycles, the problem is in the fact that the number of moves you need to check grows exponentially with the number of chess turns.

The "complicatedness" of each individial check is utterly and completely irrelevant.


   -- Signed, Lev Andropoff, cosmonaut.
[ Parent ]

If you think that is why chess is hard (none / 1) (#83)
by p3d0 on Mon Feb 23, 2004 at 02:09:26 PM EST

...then you just might be a redneck.
--
Patrick Doyle
My comments do not reflect the opinions of my employer.
[ Parent ]
theres a better way... (3.00 / 4) (#85)
by zorander on Mon Feb 23, 2004 at 03:02:00 PM EST

Traditionally, chess software has used rotated bitboards (8 bytes) and bitwise ops to determine valid positions. Generate a bitboard for position of enemy pieces, of your pieces, of valid moves from positions for each type of piece, do some magic bitwise operations and voila, you've calculated all possible next positions in near constant tine. Granted there are still some special cases, like castling, but in general, those are simple to check and bitboards deal with the more seemingly complex pieces like pawns. Checking the way you did it is *very* inefficient. When I had to write a simple chess program for an assignment, the move generator (which was implemented the way you describe) was taking up about 95% of the time for thw *whole program*. Before I could even do anything interesting the move generator had to be redone using bitboards. Brian
---- Want to get into Linux? Cheap systems available now at eLinuxBox.com.
[ Parent ]
there are several viable implementations (none / 0) (#86)
by muyuubyou on Mon Feb 23, 2004 at 03:17:42 PM EST

http://www.xs4all.nl/~verhelst/chess/representations.html

Bit boards seem promising with affordable and fast 64 bit processors already available.

[ Parent ]

Ahh, I see (none / 0) (#110)
by hardburn on Tue Feb 24, 2004 at 11:14:07 AM EST

So it's a time/memory tradeoff (and probably a tiny fraction of memory on today's systems). I can certainly see why the solution I gave above would be very CPU intensive. In any case, it was a personal project intended only to see how far I could get on the problem on my own. Not sure if I even have that code saved anywhere.

Thanks for making a far more constructive reply than the others.


----
while($story = K5::Story->new()) { $story->vote(-1) if($story->section() == $POLITICS); }


[ Parent ]
Cool (none / 0) (#135)
by p3d0 on Wed Feb 25, 2004 at 11:25:28 AM EST

On a 64-bit machine, the bitboards would fit in a register.
--
Patrick Doyle
My comments do not reflect the opinions of my employer.
[ Parent ]
flawed assumption (3.00 / 4) (#79)
by muyuubyou on Mon Feb 23, 2004 at 01:44:05 PM EST

  You underestimate the power of a dedicated and talented human brain.

  Computers have been playing very good chess for a long, long time. The thing is grandmasters can beat a machine that calculates millions of positions applying complex heuristics to each of those positions - that's what I find amazing.

  I remember the face of the chinese-american manipulator (and co-programmer) in one of the last Deep Blue (or was it Deep Thought?) matches ; when Kasparov massacred Deep Blue in the first game, he couldn't believe it - Deep Blue played better than he could ever dream possible, and Kasparov was still consistently superior. When Kasparov stood and extended his hand to shake he sat still looking up to him like he was in front of God or something, nodding with his head.

  Browsing the public part of Deep Thought's algorithm (it was partly released) and it's hardware capabilities, I wonder how can a human being beat that.

[ Parent ]

It's easy to do slowly (none / 3) (#82)
by p3d0 on Mon Feb 23, 2004 at 02:07:58 PM EST

If you take the straightforward approach you suggest, you end up with an unbeatable chess computer that requires longer than the lifetime of the universe to make each move. If you want a computer to beat a human, the game must finish within his lifetime.

If computers can't do this yet, they're much less advanced than I thought.
Either that, or chess is much more complicated than you thought.
--
Patrick Doyle
My comments do not reflect the opinions of my employer.
[ Parent ]
Why? (none / 1) (#99)
by stormie on Mon Feb 23, 2004 at 07:21:11 PM EST

If you want a computer to beat a human, the game must finish within his lifetime.

Why? Surely if you die mid-game, you forfeit, and the computer wins?



[ Parent ]
Most pro-level games are played with timers (none / 1) (#105)
by damiam on Mon Feb 23, 2004 at 10:04:25 PM EST

The computer would lose before it could make its first move.

[ Parent ]
Until computers (1.05 / 20) (#53)
by moomun on Mon Feb 23, 2004 at 05:49:15 AM EST

can engage in hot gay sex then human chess players will always be in front - or behind, depending on their mood.

<M>

Chess ratings (none / 3) (#63)
by A55M0NKEY on Mon Feb 23, 2004 at 10:20:13 AM EST

One way to determine who is the 'top' chess player would be to hold a tournament, or a series of tournaments and tally up the score.

But these chess ratings don't ring true as being meaningful at the top levels. Who is qualified to test the top player in the world? No one. So any rating the top player might hold means 'wicked high' but I don't think it can be a meaningful quantatative measure.

The chess ELO rating is highly accurate (none / 1) (#66)
by John Chamberlain on Mon Feb 23, 2004 at 10:54:39 AM EST

Currently chess ratings are devised using a method developed over 50 years ago by Arpad Elo and the system is known as the ELO system. It is an accurate measure of competitive peformance.

You might be interested to know that other sports use the same rating system. For example, tennis uses an ELO system to rate top players. They do not publish the real ratings, they only publish the ranks of the players, unlike chess where the real ratings and ranks are published.

To see the top ratings and ranks go to the FIDE top 100.

For an explanation of ELO ratings and how the rating system works.

[ Parent ]

I read more about ELO (none / 0) (#133)
by A55M0NKEY on Wed Feb 25, 2004 at 09:47:11 AM EST

I read more about the Elo system so I understand it better. My confusion was from a 'chess test' I took that was included with a computer chess game I had when I was a kid. The 'chess test' was a series of game boards that gave you multiple choice move options. You made the move, and the game calculated your 'Elo' score based on whether your answer was a 'winning move' or a 'losing move'. Thinking this was how Elo scores were normally calculated, I wondered who would be qualified to make up/score such a test for the world champion since there is no way to know for certain whether a move is winning or losing except in trivial cases. ( calculating a 'perfect' game of chess would require converting the mass of Jupiter into pure energy just to power the computer to do it. )

But now I know that the Elo system requires that actual games be played against other rated players. The 'chess test' was just an attempt to approximate it, and would, indeed not have worked on a world champion since the test was probably designed by someone with much less chess skill than a world champion.

For anyone who is interested, Elo is very similar to batting average. It is an average of game scores calculated by taking your opponent's skill (Elo score) into account and raising it by 200 points if you win, using that score if you draw, and lowering it by 200 if you lose. Hence if a person who had never played chess before loses a game played against Gary Kasparov (Elo = approx 2800) then the newbie who still can't remember that a knight moves in an 'L' shape gets an Elo score of 2800 - 200 = 2600. Clearly, like a batting average ( everyone starts with a batting average of 1000 which is much better than any major league player ) Elo only becomes meaningful after lots of games. Though if Babe Ruth played only against little leaguers his Batting average would soar, his Elo score would not since the little leaguer's relatively suck.

On the same note, if Gary Kasparov played against me ( Elo approx 0 ) then even though he would win, he would have the number 200 averaged into his Elo score which would bring his Elo score down.

So if you are 'going for elo' then it is not in your best interest to play anyone with an elo score more than 200 below your own.

This brings up Rock-Paper-Scissors ( Rochambeau ). Last night I saw the world Rock Paper Scissors (RPS) champion on the Conan O'Brien show. He currently has the world title after beating out 300 contestants. No doubt his RPS Elo score is hella-high. However, I might program a computer to play RPS by always using the same strategy: Rock Rock Rock. If ( without prior knowledge ) my program beat the world RPS champion ( It would have a 50% chance of doing so ) then it would have the world's highest Elo rating for RPS even though it would have no chance of winning an RPS tournament ( since the contestants would quickly wise up to the fact that the machine never does anything except rock, and beat it soundly with paper.

[ Parent ]

ELO (none / 0) (#153)
by Rich0 on Thu Mar 11, 2004 at 10:30:49 PM EST

On the same note, if Gary Kasparov played against me ( Elo approx 0 ) then even though he would win, he would have the number 200 averaged into his Elo score which would bring his Elo score down.

I don't think that is true.  I haven't read up on ELO in a little while, but I think the way it works is that when you have a match between two rated players, a certain number of points are up for grabs.  You never lose points for winning, or gain points for losing.  You always lose points for losing, and gain points for winning.  However, the number of points gain/lost is based on relative strength.

So, based on the calculator on the FIDE website if you played against Kasparov with an established rating of 200 for yourself then in one game there could be the following outcomes:

You:
Win - you gain 10, Kasparov loses 10
Lose - your scores don't change
Draw - you gain 5, Kasparov loses 5

Kasparov, of course, has no incentive to play you since the best he can do is keep his rating the same.  He loses points even for a draw, since he is such a strong player he is expected to win.  If he draws consistently against weak players it would demonstrate that he is a weak player as well - hence his rating would fall.

If Kasparov played an identically-rated player then the winner would get 10 points, the loser would lose it, and in the event of a draw nothing would happen.

In a top-level match only 10 points of the rating are ever up for grabs.  For even beginners only 25 points are up for grabs.  You can't get one lucky draw against Kasparov and suddenly be rated second-best in the world.

[ Parent ]

Anti-computer chess (none / 3) (#78)
by Big Sexxy Joe on Mon Feb 23, 2004 at 01:39:05 PM EST

Everyone was so mad that Deep Blue was programmed specifically to beat Kasprov.  But the Grand Masters are becoming more and more relient on anti-computer strategies to even draw against computers.

Also, the thesis of the linked article is that the human grand masters are getting better faster than the computers.  Well, right now there's only two that can beat them and they're going to get old eventually.  By the time a new great human chess player emerges it's probably going to be to late and he will never catch up with the machines.

I'm like Jesus, only better.
Democracy Now! - your daily, uncensored, corporate-free grassroots news hour

Bad analogy (none / 1) (#104)
by damiam on Mon Feb 23, 2004 at 10:03:10 PM EST

Deep Blue was programmed to use anti-Kasparov strategies. It could not have beat many non-Kasparov Grand Masters. Meanwhile, Kasparov and Kramnik can beat pretty much everyone else, including computers. They may use certain strategies against them, but they are still flexible enough to win against a broad range of opponents. Deep Blue was not, and that was the issue.

[ Parent ]
Well (none / 1) (#107)
by Big Sexxy Joe on Mon Feb 23, 2004 at 11:29:29 PM EST

I'm sure that they could have programmed Deep Blue to be good against others but it would still just use the Anti-Kasprov strategy against Kasprov.  Therefore, they could have augmented the program but there was no need to.

I'm like Jesus, only better.
Democracy Now! - your daily, uncensored, corporate-free grassroots news hour
[ Parent ]
Maybe they could have (none / 0) (#119)
by damiam on Tue Feb 24, 2004 at 05:36:24 PM EST

But they didn't, so we'll never know. Kasparov was picked because, aside from being World Champion, his style was deemed vulnerbale to computer play. If IBM had felt confident in Deep Blue's ability to beat other GMs, they would have let other GMs play against it. It would have been great publicity for Deep Blue to beat the top [x] players, but IBM knew that Kasparov was all they could do at that point.

[ Parent ]
Well we still have.. (2.00 / 5) (#89)
by RandomAction on Mon Feb 23, 2004 at 03:51:57 PM EST

..GO I don't know for how long though

Damm that was Redundant Sorry <NoTxt> (none / 1) (#90)
by RandomAction on Mon Feb 23, 2004 at 03:58:29 PM EST



[ Parent ]
Correction...? (none / 2) (#91)
by skintigh on Mon Feb 23, 2004 at 04:26:52 PM EST

I assume what the author meant was the FPGA is a Virtex family FGPA, made by Xilinx. I've often wondered why FPGA's aren't used more often in Chess machines. I've also wondered why some rich person doesn't take those FPGA designs, make them into ASICs, and increase the machine's speed by 1000%

Right, Virtex is made by Xilinx (none / 0) (#95)
by John Chamberlain on Mon Feb 23, 2004 at 05:40:25 PM EST

My mistake. Virtex is not the company, it is the line name for a set of FPGAs from Xilinx.

My article in Developer Diary today mentions this. In theory you could put the OS on FPGA and make an FPGA PC that would run 1000s of times faster than an ordinary PC.

[ Parent ]

No no no (none / 0) (#98)
by gte910h on Mon Feb 23, 2004 at 07:08:43 PM EST

My article in Developer Diary today mentions this. In theory you could put the OS on FPGA and make an FPGA PC that would run 1000s of times faster than an ordinary PC.

FPGA's and ASIC's are specified with something that looks like a programming language, but as a C programmer who occasionally has done programmable logic, I assure you its NOTHING remotely alike. There is no possibility of "porting" an OS to FPGA/ASIC. You could "Port" a processor to one, but that'd be slower than the normal CPU for sure. Sure C LOOKS LIKE Verilog and VHDL looks like Ada and JHDL looks like Java, but you're doing very mutually exclusive things when you're using the specification language rather than a programming language.

[ Parent ]
if Hydra runs on FPGA why not OS? (none / 0) (#100)
by John Chamberlain on Mon Feb 23, 2004 at 08:57:49 PM EST

If a chess program like Hydra can run on an FPGA why not an operating system? Logic is logic. All programs can be reduced to just four types of instructions: set, branch, compare and add. All four of these can be coded in logic gates (obviously).

If there is some operational reason why an operating system could not be ported to an FPGA I would like to know what it is.

[ Parent ]

Hizza wha? (none / 0) (#109)
by gte910h on Tue Feb 24, 2004 at 01:33:10 AM EST

Logic is not logic ;)

Seriously, the type of highly paralized, number/mask based computing you do for chess is very apt for PLDs.

But an os? I mean the point of an OS is to be a HW intermediary, and run user programs. Are we supposed to write all our programs in Logic? Programming lanuages assume a turing machine. Ma Intel and Ma AMD have pretty freaking fast turing machines. There are some tasks that work the parallel way that PLD's do (math and filters, come to mind), but most don't seem to really fit that paradigm.

YOu can't really do many of the things a modern computer does faster than it does. Like checking memory? One at a time please. Initializing HW? same deal.

[ Parent ]

Well (none / 0) (#134)
by Lagged2Death on Wed Feb 25, 2004 at 10:14:55 AM EST

Designing and constructing an ASIC part costs big bucks, and when you're done, the design is cast in stone forever. FPGAs are pizza-money cheap, and you can change them around in just a few minutes if you find bugs or have a brainstorm. If you want 10x the performance, it's probably pretty cheap (no more than a few hundred bucks) to just throw 10x as many FPGAs at the problem, as long as it's a parallelizable one (like chess).

If you were planning to build thousands of dedicated chess-playing grand-master beater computers, then an ASIC might be cost-effective. (But with insanely fast and cheap general-purpose PCs everywhere, who would buy such a beast?) Otherwise, that money would be more effectively spent on algorithm research, the fruits of which could eventually improve chess-playing FPGAs, software, and ASICs.

Starfish automatically creates colorful abstract art for your PC desktop!
[ Parent ]

Injecting Go into computer chess articles. (2.00 / 4) (#92)
by Fen on Mon Feb 23, 2004 at 04:54:14 PM EST

This is a running slashdot joke. Some wide-eyed poster enthusiastically pipes in about Go in any story on computer Chess. Like he's discovered an ancient Asian secret bigger than sliced bread.
--Self.
Joke? (none / 1) (#131)
by Ted Briderider on Wed Feb 25, 2004 at 05:16:57 AM EST

Shows just how funny the people on slah dot are!!

[ Parent ]
Joke? (none / 1) (#146)
by irrevenant on Fri Feb 27, 2004 at 04:40:31 AM EST

AFAIK, it's not a joke.

The point usually made is that Go is more of a challenge to computers than Chess, because it's much less vulnerable to brute-force number crunching.

I dunno enough about Go to offer any more detail...

[ Parent ]

After the singularity? (1.25 / 4) (#93)
by Fen on Mon Feb 23, 2004 at 04:57:52 PM EST

Perhaps the spiked intelligence will find a way to solve chess completely. There may be some real theoretical limits to brute-force capability, but a post-singularity intelligence could happen on enough shortcuts to do it.
--Self.
Geek Religion (2.80 / 5) (#113)
by Bridge Troll on Tue Feb 24, 2004 at 11:49:30 AM EST

While many geeks pretend to be athiest, it is a sad reality that many of those "athiest" geeks are, in fact, believers in an idol -- technology.

Look at this guy! He speaks of a singularity as if it were some sort of final Armageddon, after which only the righteous will be left. I mean, really, ... this Singularity, when spoken of so reverently by its adherents -- eternal life, man reaching (the) heaven(s) -- could look like the 'story' of Revelations with some good word substitutions. The parallels with religion in general, and Christianity in particular, make this comment appear, to the trained eye, as the filthy venom of just another kind of religious extremist.

Geeks are simply fanatics of a different faith.




And besides, pounding your meat with a club is a very satisfying thing to do :) -- Sleepy
[ Parent ]
Tic-Tac-Toe (2.25 / 4) (#102)
by Zabe on Mon Feb 23, 2004 at 09:05:39 PM EST

Eventually you will be able to map all possible moves, so you'd know who will win from the beginning, or play always to a draw (like in tic-tac-toe)
Badassed Hotrod


No (2.75 / 4) (#103)
by damiam on Mon Feb 23, 2004 at 09:33:54 PM EST

Chess is a tremendously complex game. I think I've seen it calculated that even if every electron represented a bit, there's not enough storage space in the Solar System to hold a chess-move database (and current computers would take billions of years to generate it).

[ Parent ]
Hmmm.... (none / 1) (#108)
by Zabe on Mon Feb 23, 2004 at 11:51:52 PM EST

I looked it up here is what it says (Go Wikipedia! :) )

http://en.wikipedia.org/wiki/Chess

The number of legal positions in chess is estimated to be between 10^43 and 10^50, and the game-tree complexity approximately 10^123

10^123 is a lot, probably more then the # of atoms in the universe...  , so I guess full game trees won't be around for a while.  Oh well, go quantum computers! :)

Badassed Hotrod


[ Parent ]
The game is still finite (none / 2) (#114)
by Shimmer on Tue Feb 24, 2004 at 12:40:55 PM EST

And thus can be "solved", at least in theory. Correspondences with the number of electrons in the universe seem utterly irrelevant.

Wizard needs food badly.
[ Parent ]
In theory, yes, chess is solvable (none / 0) (#118)
by damiam on Tue Feb 24, 2004 at 05:34:12 PM EST

But theory is utterly irrelevent. Unless we can actually compute and store that amount of data, we will never solve chess. The number of electrons in the universe is relevent, because we cannot currently conceive of any storage scheme that could store more than a bit or two per electron.

[ Parent ]
Electrons don't matter (none / 1) (#124)
by Shimmer on Tue Feb 24, 2004 at 11:28:14 PM EST

Why do you assume that the entire move tree must be calculated and stored in order to solve chess? Think about a game like Nim. Although the move tree for Nim can be quite large, the solution is very short. There is no need to calculate the entire move tree.

Wizard needs food badly.
[ Parent ]
Well sure, for Nim... (none / 0) (#140)
by Adam Tarr on Thu Feb 26, 2004 at 03:04:27 AM EST

Well, in Nim, there is a heuristic that can operate on a simple transformation of the game board, to find the optimum move in one step.  Unless there is such a transformation in chess, or such a "1-step" heuristic, then a total solution to the game can only be found by tracing backward through the move tree.  This means finding a path to checkmate from every possible response.  Truly, a monumental task.

That said, I'm not convinced chess will never be solved.  While the branching factor is clearly too large for normal computers to completely swallow, in theory a quantum computer could handle it without difficulty.  So if large-scale quantum computers ever come to be, then one of them may be able to reduce every chess position to a certain win, lose or draw.

[ Parent ]

It's inevitable... (none / 0) (#116)
by araym on Tue Feb 24, 2004 at 04:19:06 PM EST

I took an AI class last semester where game theory was the main topic and we were told that the avg branching factor for chess was approximately 36 and that a game between skilled opponents should last about 50 moves. So to fully map a min-max tree you would need 36^50 evaluations of the heuristic function, this could possibly be reduced with techniques like alpha-beta pruning. The computation would take an enormous amount of time although the memory is inconsequential in comparison since you would not need to store the vast majority of the states.

The amazing thing is that once the tree has been mapped there will be no way to beat the computer EVER, it would never make a move that would end in a terminal loss state so the best you could hope for is a draw....

Any game that is finite eventually will be dominated by computers. Go is a particularly interesting game for AI as it's branching factor is 100!

-=-
SSM

[ Parent ]
Well... (none / 1) (#144)
by UserGoogol on Thu Feb 26, 2004 at 10:02:50 PM EST

Some games do not have a solution. Chess might turn out to be like Tic-tac-toe, or it might turn out to be more like Rock-paper-scissors. In RPS, you have to guess what strategy your player might use without any information from the game itself. As such, it is not possible to determine the outcome of two perfect players playing against each other in Rock-paper-scissors. The same might be true of Chess.

[ Parent ]
Just found this link by chance On IEEE (none / 2) (#115)
by RandomAction on Tue Feb 24, 2004 at 03:24:58 PM EST

Chess programs keep getting better, but grandmasters have learned to anticipate their game. Take a look here: Psyching Out Computer Chess Players

completely wrong (none / 1) (#120)
by beefman on Tue Feb 24, 2004 at 08:31:58 PM EST

The author is apparently aware (judging by one of his replies to a comment) of the other articles in Mr. Sonas' series, but he hasn't addressed why he has come to the completely disagree with Sonas' conclusion.

-Carl

omg Omg OMG! (none / 3) (#122)
by ShiftyStoner on Tue Feb 24, 2004 at 10:38:13 PM EST

 First chess, WHAT NEXT. I bet this is how the matrix started. They're going to TAKE OVER THE WORLD. If the lab mice don't do it first.
( @ )'( @ ) The broad masses of a population are more amenable to the appeal of rhetoric than to any other force. - Adolf Hitler
Learn from boxing (none / 2) (#123)
by swagr on Tue Feb 24, 2004 at 11:21:08 PM EST

We have different weight classes in boxing. We should have the same in Chess.
Which makes me wonder... When are we going to see humans boxing with robots?

we do (none / 0) (#125)
by beefman on Tue Feb 24, 2004 at 11:29:34 PM EST

Most tournaments have sections, and/or pairings are done by rating.

-Carl

[ Parent ]

brain weight (none / 0) (#137)
by swagr on Wed Feb 25, 2004 at 12:37:23 PM EST

Most tournaments have sections, and/or pairings are done by rating.
Yeah. I'm saying: do it by weight. Since the brain is the key component here, pairings should be done with brain weight. So a chess playing computer's "guts" should weight no more than 1.4 kg.

[ Parent ]
Nope. (none / 0) (#145)
by irrevenant on Fri Feb 27, 2004 at 04:31:05 AM EST

The brain is not the sole source of human intelligence.  There's also flow of hormones etc. that influences mental ability.

More importantly, without the human body (lungs, heart, arteries etc.) to keep it running, the brain is just a lump of meat.

If you're going to compare weight class, you need to compare the weight of the brain AND ALL THE WETWARE NEEDED TO KEEP IT RUNNING with the computer CPU AND ALL THE HARDWARE NEEDED TO KEEP IT RUNNING.

An interesting technical that would be sure to be argued about is external power - are computer chess players allowed to hook into the mains?  IMO, it probably shouldn't be allowed and batteries/generator/whatever should be factored into the weight class.

[ Parent ]

Boxing with robots. (none / 0) (#132)
by Vendor on Wed Feb 25, 2004 at 05:47:25 AM EST

Sounds like an interesting idea. Patent that.

[ Parent ]
Computers and humans together? (none / 1) (#129)
by Fen on Wed Feb 25, 2004 at 12:25:08 AM EST

This is an aspect of the singularity--improving the man-machine interface. As of now, we are interacting just at the start, and leaving the machine on its own after programming it. Imagine interacting at every step. This has been done with the Other Game with the computer generating two moves, and the player deciding where to actually put the stone.
--Self.
This means nothing about humans being beaten (none / 3) (#130)
by Attercop on Wed Feb 25, 2004 at 02:28:03 AM EST

All a computer is, is a machine that can follow instructions. That is *all* All a chess-playing program is, is a list of instructions written by a human that are designed to win a game of chess. ie "If he does this, then you do that. And if he does that, then you do this" etc I just don't see it as a machine vs human competition. If a computer wins, it is actually a testament to human intelligence.

Sort of... (none / 1) (#141)
by Adam Tarr on Thu Feb 26, 2004 at 03:16:45 AM EST

All a computer is, is a machine that can follow instructions. That is *all* All a chess-playing program is, is a list of instructions written by a human that are designed to win a game of chess. ie "If he does this, then you do that. And if he does that, then you do this"
That's a bit of an oversimplification. Certainly, the opening libraries amount to exactly that. But as soon as the opening libraries are abandoned, the computer has to make a decision based on looking ahead at a huge number of potential chains of moves and counter-moves, and evaluating those positions. That still involves following instructions written by a human, but they are no longer the sort of trivial instructions you describe.
I just don't see it as a machine vs human competition. If a computer wins, it is actually a testament to human intelligence.
That is a perfectly reasonable viewpoint. It is a triumph of the engineers that design the customized parallel processors, and it is a triumph of the chess experts that prepare the opening libraries and train the program, and it is a triumph of the software designers that design and fine-tune the program's approach to looking ahead and evaluating positions. The computer didn't build itself, after all. But saying "this is a triumph for artificial intelligence" is really just shorthand for all of the above.

[ Parent ]
on the same pitch (none / 0) (#151)
by relief on Sun Feb 29, 2004 at 08:13:05 PM EST

if a group of madmen create machines that destroy humanity, that would also be a testament to human intelligence. or human stupidity. i think the problem is that intelligence is a relative concept.

----------------------------
If you're afraid of eating chicken wings with my dick cheese as a condiment, you're a wuss.
[ Parent ]
Win a world championship! (none / 2) (#142)
by pelliott on Thu Feb 26, 2004 at 10:30:15 AM EST

If computers want to be considered the best, they should win a world championship, qualifiying THE SAME WAY HUMANS HAVE TO! That is, they should have to play in, and win the necessary tournaments to be considered a challeger. Their games in qualifying should be published, just as humans games are, so that opponets can study their style for flaws! Everytime the programmers change the "hardcoding" of their program, the computer should have to requalify, so that masters will have a chance to study the new style. The computer should be allowed have games read into their database without requalifying just as masters can read the chess literature. I believe that the rules these computer vs. human expositions unfairly benefit the computer by allowing it to "come out of nowhere" as it were, playing the GM, without giving the GM a chance to study its style, while in contrast the GM's previous games have all been published. If a computer can win the world championship this way, then it will have something to crow about.
---- There is no Religion Higher than Truth.
ego subroutines (none / 0) (#152)
by avidya on Wed Mar 10, 2004 at 09:10:11 PM EST

"If computers want to be considered the best"

If computers ever develop egos like that we're all screwed. It's bad enough that we can develop machines to do our work for us, and decide to let them, but what happens when the computers decide that they're more capable and take over!

Humans are becoming the tools that run computers, when it should be the computers that are the tools of humans.

[ Parent ]
Bah! (none / 1) (#150)
by Dirty Liberal Scumbag on Sat Feb 28, 2004 at 04:27:58 AM EST

Wake me up when a computer can do one of the following:

1. Laugh at a dirty joke.

2. Love pop music "just because."

3. Fool me into thinking Dosteovsky is still alive.

Until then it's all just equations and probabilities.

Cheers
DLS
---

I am now whatever you wish me to be to excuse your awesomeness.

Why are humans still winning? (none / 0) (#154)
by StangDriver on Tue Apr 06, 2004 at 02:13:20 PM EST

What are humans doing that the computer cant do? Currently, it seems computers are still unable to completely *solve* chess, but surely they are able to analyze more possible moves than a human opponent can. Why aren't humans being crushed by computers?

yes (none / 0) (#155)
by soart on Wed Jun 29, 2005 at 12:09:50 PM EST

Will 2004 be the turning point for human versus computer chess? By autumn we may know.
机票打折机票
Chess Computers On Track to Overtake Humans in 2004 | 155 comments (149 topical, 6 editorial, 1 hidden)
Display: Sort:

kuro5hin.org

[XML]
All trademarks and copyrights on this page are owned by their respective companies. The Rest © 2000 - Present Kuro5hin.org Inc.
See our legalese page for copyright policies. Please also read our Privacy Policy.
Kuro5hin.org is powered by Free Software, including Apache, Perl, and Linux, The Scoop Engine that runs this site is freely available, under the terms of the GPL.
Need some help? Email help@kuro5hin.org.
My heart's the long stairs.

Powered by Scoop create account | help/FAQ | mission | links | search | IRC | YOU choose the stories!