Chess Championship: Humans vs. Computer

An anonymous reader writes "X-bit labs has posted very interesting editorial called "Chess Championship: Humans vs. Computer". During the last 10 years computers penetrated into various spheres of human life. In this article guys try to find out how well computers can play chess and if it would be correct to say that artificial intelligence is superior to human mind. Interesting read."

Comments

Interesting, but...

[TopShelf]

I don't know if chess playing really qualifies as AI. The game gets broken down numerically such that the computer's job is just to crunch through the myriad possible moves and select the best one. All the intelligence goes into the algorithm that rates various positions, and the calculation scheme by which possibilities are evaluated, which are the human inputs. It just sounds like too narrowly focused a task to be considered AI.

Re:Interesting, but...

[beders]

The best chess players look for patterns, but through experience and insight can discard millions of possible moves and concentrate on the most effective and bring about a squence that forces their oppenent in a direction they don't want to go. Saying that I have no idea how to get a computer to do the same :)

Re:Interesting, but...

[Afrosheen]

Mad geniuses, IMHO, are just like the 'AI computers' that play chess. They tend to be very deficient at everyday tasks (hygiene, social skills, excercise, etc.) yet extremely proficient at a specific task. Once you strip away everything that makes you human, you can focus on one thing and become superhuman.

There are, of course, anomalies; people that are genii but continue to lead somewhat normal lives, but these people are rare. True genius comes at a cost, and that cost is high for most.

Re:Interesting, but...

[manonthespoon]

AI != Thinking like a human being, imitating a human being, etc.

AI has many branches, several of which are applied AI and have to do with having a computer do some complicated task. Whether or not the computer is thinking should be seperated from whether or not we're talking about AI.

If this article is about how computers may be becoming more intelligent then human beings, then they weren't paying attention when Deep Blue first beat Kasparov. No one at IBM ever said: "Behold, the first intelligent computer! It can think for itself!"

However, they did very clearly say that chess is sufficently well understood that a fairly basic AI algorithm and heuristics (and a lot of specialized "Chess Processors" could essentially turn the game into a huge search problem.

I think that people should not mix terminology here. If they are talking about computers having human intelligence, then that is what you should talk about, not AI. Because AI includes all of the things like playing chess, and emulating human intelligence, and learning algorithms, and...

Re:Interesting, but...

[fd]

I think we are agreeing. :-) Searching itself is not AI in it's own right. But there seems to be a hangup that people think all a chess engine does is search. In fact that's only half of the equation. A chess engine also EVALUATEs every position it reaches. And because those positions are not at the end of the game it has to make a quantitative and qualitative evaluation (e.g. I have more pawns but two of those pawns aren't as "good" because of X). Chess positions are more than the sum of their parts. The player with the most pieces isn't always in a better position. That second part, the evaluation, is where I think AI research could benefit chess engines.

Humans are much better than computers at evaluating static positions on the board. Imagine a player with the brute force searching capability of a computer combined with the chess knowledge and intelligence of a grandmaster evaluating each position.

chess != AI

[tigress]

The game of Chess is not a measure of intelligence. It's a measure of mathematics and memory. A computer is no smarter if it's able to play chess than one that isn't.

The reason for this is that Chess is a game where the rules are strictly defined. For each move, there can only be a limited - and known - number of outcomes. This reduces the entire game to a matter of mathematics and statistics.

No, the real test of intelligence would be for a computer to react to and handle a situation where the rules are NOT predefined - such as a real world scenario.

When a computer is able to take a limited number of inputs and make a judgement based on the (possibly) inaccurate and (definitely) insufficient data available, you can start talking about intelligence. Still, even then you're not talking about true intelligence. AND, for that matter, such programs do exist - they're called expert systems.

No, what I'm prepared to call intelligence is a program that not only is able to make a judgement based on possibly bad data, but is also prepared to admit that it made a mistake and learn from those mistakes. That would, in my opinion, be a truly intelligent program.

After all, assuming it's able to do that, it'd certainly be a lot more intelligent than a lot of humans I know. =)

Re:chess != AI

[MnO-Raphael]

The game of Chess is not a measure of intelligence. It's a measure of mathematics and memory

True, but as Turing pointed out: if you can't tell the difference in a certain context, does it really matter if it's *really* intelligent or not?
AI is a misplaced term - "adaptive systems" would fit much better. I too have a problem with calling something that doesn't even know it's playing chess for intelligent.

The problem with your argument.

[nigel.selke]

Every time technology advances, the definition of what tasks can be said to require "true intelligence" changes. For example, before the advent of calculating machines, arithmetic was considered something that was uniquely human. Now, people dismiss computers' abilities to do lightning fast arithmetic, and, in fact, use it as a basis for putting down other abilities of computers/other high technology ("But that all comes down to number crunching. It's not true intelligence.").

Of course, you (and they) could be right about it. But it's interesting to note that chess is another prime example of this. Computers became extremely good at number crunching and large-scale analysis, and people shrugged it off. "A computer would never be able to compete with competent chess player, and could certainly never compete with a Grandmaster. Chess requires true human intelligence." 20 years later, a computer tied with the reigning Chess Champion. Now - Chess doesn't really require true intelligence, it all boils down to number crunching.

The problem is, where do we draw the line? As computers start adding more and more to their lists of abilities, especially in areas such as pattern recognition and expert systems, are we going to claim that those things don't require intelligence, and can also all be brought down to number crunching? To me, it seems like a form of denial. Instead of clinging to the old ways, why not recognize that computers might just be better at a lot of things that we previously thought were "human-only" areas of skill, and adapt accordingly.

Re:The problem with your argument.

[Jugalator]

You make a good point -- in the end, regardless how advanced the AI is, it might all boil down to number crunching, just like it all boils down to chemical reactions in a brain.

Re:The problem with your argument.

[zmotula]

But chess *really* does not need any intelligence if you have enough computing power. You can be either (a) intelligent with low computing skills --- human or (b) dumb, but with excellent computing skills --- computer.

Only our insufficient computing power makes chess the nice game that requires intelligence.

Computers don't enjoy playing chess, it's a routine <g>

Re:The problem with your argument.

[Kynde]

1) Make a computer with true free will. Let's see AI do something it wasn't originally designed to do because it wants to.

How would you measure that? Especially if you knew that in the it boiled down to number crunching with some entropy input. You do remember that the concept of free will is meaningful only subjectively, i.e. from one's own point of view.
Although it is widely held that among human kind if one has it then all do, but that does not apply to AI.


2) "True intelligence", at least on par with us, will happen when a computer does everything we do mentally, while having full articulate motor skills, and then takes it upon itself to create an AI that crunches numbers better than it does, beats itself at chess, etc.

Bollocks, the earlier poster said it well. We're just drawing the line further and further, mostly because what we're after is that "well, err, when they're like us" while all along we're not quite sure what that means.

Moreover, the planes is infested with actual human beings that would fail on either of those.

Besides both of your points there are unscientific, neither of which can be measured in any way. That's all there really is to it though. Milestones. Wether a person qualifies that as AI is subjective to the definition of AI, for which here in /. I'm guessing are a myriad of different interpretations.

There are a number of good well defined tests that we can put the AI through, every one of those passed is significant. Especially the forementioned arithmetics and chess should NOT be forgotten, because they indeed were once held high.

Re:The problem with your argument.

[Anonymous Coward]

"A computer would never be able to compete with competent chess player, and could certainly never compete with a Grandmaster. Chess requires true human intelligence."

Well, I used to play chess in the good old days, and have never come across this point of view. I have heard people say computers never going to be able to do the numberchrunching FAST enough to be able to compete with a human, but that is a totally different claim.

One of the most important developments in AI is how the definition of intelligence changes. As times go by, we see more and more advanced concepts become fragmented into strictly rule based and knowledgebased procedures; easely solved with computers.

I think in the future, the people that study AI, will spend more time concentrating on where humans clearly break rules in order to solve problems. At this stage we might see the start of Artificial Creativity, which I personally think is one of the most important aspect of AI.

Re:The problem with your argument.

[platypus]

To add another example. Richard Feynman writes in one of his books about how he listened to a conversation between to students of mathematics(at MIT or so) where one tried to explain some mathematical concept to the other.
Feynman described that after a long while and much intense explaining, finally the other student "got it", and said something along the lines of "Oh! YES, THAT'S TRIVIAL!"
Feynman goes on to make fun of mathematicians by proposing that mathematicians only understand trivial problems, because anything they have already understood is declared trivial by them.

This is a bit extreme, but it decribes exactly the notion some AI critics seem to have when judging AI advances.

Re:The problem with your argument.

[Dahlgil]

What you have here is really the crux of the definition problem. AI will not be considered AI until a machine is able perform an act of mental dexterity for which we do not have an explanation (not very likely). You see, as soon as we have a perfect understanding of how a mental process was carried out, we no longer consider that process to be an act of intelligence but simply a mechanical routine. The changes we have witnessed in the definition of intelligence really point out what the very definition of intelligence is. "True" intelligence makes "sense" to us, but is never perfectly understood. When we, as humans, make a decision or reach a conclusion, there is always some element of mystery about it. We don't know, for example, exactly how the thoughts leading to the "act of intelligence" are represented in the brain, or exactly what neurons fired, or what sensory or memorized inputs contributed to it. We just "feel" and "sense" an intelligent rationality about it. On the other hand, if we *did* know all of the physical mechanics about how the "thought" was carried out, we could readily model it in a computer, and step through the entire process of the thought in a debugger. But would we still regard the thought as an act of intelligence any more, or as just a routine? Isn't this exactly what has happened in the case of chess computers? I do not think that there will ever be machines that will be regarded by humans who live contemporarily with such machines as "intelligent", for precisely the reason of definition. Yes, we will develop machines that manipulate and process information better and better, and in ways that more accurately reflect the kinds of decision making humans perform. But as long as it is possible to go back to the machine and retrace the exact algorithim, storage mechanics, and logic flow that are being followed in the machine, it is unlikely that people will ever accept that as real intelligence.

Re:The problem with your argument.

[platypus]

OTOH, there's an interesting short story in one of Hofstaedter's (spelling right? to lazy to look up) books where one guy build mechanical "bugs" which could express very simple "emotions" - i.e. making some pet like noises, making noises of fear, crawinling around etc. - and he asks a visitor to smash one bug with a hammer. The story describes how this man wasn't emotionally able to do that, because he developed feelings for this bug.
(IIRC the real story is somewhat more involved, but you get the idea).

I bet if you decorated an "intelligent" AI with some emotional dressing, you could significantly lower the barrier to accept it as "intelligent".

Shows how deeply involved the human perception of not only intelligence, but life in general is.

Human V AI...

[bazmonkey]

...if it would be correct to say that artificial intelligence is superior to human mind. Seeing as how it was our mind that created AI, somehow I just don't think so.

You show me AI that takes it upon itself to create it's own AI that outperforms itself, then I'll concede. That's the mark of intelligence: having the capacity to create something more capable than yourself, and not only make it, but think it up.

I really wish I could believe stuff like this-

[Ieshan]

I hate it when people compare AI and human chess players and say the following three things:

a) The computer cheats because it can evaluate more moves
b) The computer cheats because it has "traps" and "100% win situations" programmed in
c) The computer cheats because it has access to previous human games and can "guess" a player's strategy

This might be true, but most grandmaster chess players have played thousands upon thousands of hours of chess. They can immediately rule out half the moves on the board as "stupid" or "unhelpful", and they themselves come with the special knowledge of having seen many, many board situations and having worked out their solutions.

Chess is an interesting game because it is on the scale of infiniately complex.

Computers also have a serious disadvantage: the players they play against are not computers, and therefore do not evaluate moves with the same algorithms. For instance, when Deep X makes his check he says, "I'm going to do this... and then... Kasparov might do that... and I might do this... and Kasparov might do that..." - all the while substituting in what he believes are probable moves for Kasparov based on his own algorithm. This may be disadvantageous because Kasparov may analyze a situation from a different perspective - and while this is a factor in EVERY chess game (human vs. computer or human vs. human) - it is important to note that the computer does not have the priviledge of analyzing the situation from these distinctly human perspectives.

Expert System, not AI

[MickyJ]

Chess playing software is an example of an expert system, not a true AI system.

Ho-hum

[Malfourmed]

Nobody got outraged when that new-fangled mechanical auto-mobile contraption started to outpace the world's fastest human runners. Yet a computer beating a grandmaster in chess was an apocalpytic event. As others have pointed out chess can be won by using a fairly unsophisticated brute force mathematical approach at which computers excel. It's really no big deal.

I'm much more intrigued by developments in artificial creativity - poems, musical compositions, jokes, stories; where the rules governing the construction of these works are much more elusive. When a computer-generated novel wins the Booker Prize we'll have passed a signficant threshold.

Or to come back to the chess comparison - if a computer programme which adopted a human approach to chess playing, eg calculating no more than three or four moves ahead rather than nine or ten, evaluating a dozen potential decision branches rather than thousands, beat a human grand master - that would be a more significant advance in AI.

It would be like building a human-shaped robot which was able to out-run (not just outpace) a person, rather building a mechanical device which gets there by adopting an entirely different paradigm: wheels, not legs; brute force chess move evaluation, not (largely) intuitive leaps.

Don't test with trivial games

[Anonymous Coward]

Chess is a trivial game, a problem practically solved by simple algorithms. If you want a better test, see how badly computers play GO, a game with much simpler rules.

If you think computers should play better go (and know what you talk of), consider helping Gnu Go

So what is considered AI?

[Bombula]

It's one thing to know what something isn't; it's quite another to know what it is. I think it's clear that running a search and performing arithmetic are functions to simple to have emergent properties resembling intelligence. So then what properties would have to emerge? What are the properties of intelligence?

In my admittedly ignorant view, intelligence largely boils down to three closely related things:

1. Noise filtration.

Humans and animals - even simple ones - can prioritize what sensory input to process. This is how we pick objects out of the background visually, sonically, and - in humans - abstractly from conceptual landscapes.

2. Pattern recognition.

Correctly identifying patterns within chaotic data streams are where biological computers (brains) excel, thanks probably to massively parallel processing and phenomenally well designed algorithms courtesy of natural selection. Listening to one person's voice in a crowd requires both (a) ignoring all other sound, and (b)correctly identifying and processing the relevant data coming in, including information about context. Current Voice Recognition technology, for example, is poor despite massive number crunching because algorithms for noise filtering and pattern recognition are crude. Note also that pattern recognition is 4-dimensional: we recognize things in motion, not just standing still (read "behavior").

3. Information inference.

Current software doens't allow computers to handle a lack of data very well. If information is missing, brains fill in the gaps and make inferences efficiently and effectively. Sometimes this goes wrong, as when you mistakenly think you see something out of the corner of your eye. But mostly we get this right, hence the brain's accurate and effortless construction of motion from still frames flashed 24 times per second on a movie screen.

A simple test of these qualifiers is anticipation. When software can filter noise, recognize patterns, and infer information well enough to demonstrate the faculty of anticipation, then we will be making steps towards genuine AI.

The flaw of being perfect

[csritchie]

Chess programs have always been limited by the fact they try to find the most logical move; that leads to the most logical sequence for the current board position.

They are hardly cheaters.

True they capitalize on mistakes, but if you play Fritz, or Chessmaster on the most diffuclt setting, even a relative novice can make it to move 20. The computer will try to read your opening and play "book" against it.

Whereas if you were to play Kasparov as a relative
novice, I would wager the game would be over, or at the very least you would be in a position that could not be won, by move 15 or so.

If a human sees you make a move that isn't the best possible move, they can switch their whole strategy to be more aggressive. Computers play the board not the person.

So far programs treat Kasparov and a relative novice the same. Knowing no difference aside from how the game develops.

A perfect thing can only make the perfect choice.
Luckily we aren't limited by such trivialites ;)

Re:To test a powerful computer, play an ancient ga

[mav[LAG]]

That's the thing, eventually the computer will be able to go through every combination, and be the best Go player in the world.

This is not true - Go has too much depth to be effectively searched beyond just a few moves. The first 14 moves of Go have more than 200^14 possibilities. Go games take many many more moves than that to complete.
The second problem is that an effective searching algorithm is only the first step. The really hard part is trying to come up with an analysis function based on pattern matching. There are no weights for different pieces, some more important than others. Each stone is worth the same. It's the arrangement of stones which counts - something really hard to describe as a heuristic.

Read the grandparent's linked article - it explains all this a lot better than I can...