Turing test applied to chess? I would like to present an exercise in detecting “intelligence” via the Turing test. Without observing the actual entities behind a series of actions and/or responses, can one determine whether a computer or a human was the entity responsible? Though applying this question to a game of chess, by definition, is not a strict application of the Turing test, the general idea is present.
Why have I chosen chess when we all know that computers can outperform humans in solving complex mathematical systems of equations? Simple – equations are pure math: chess is not. Chess has always carried with it a sense of needing a [i]different kind[/i] of intelligence to play it well. Sure, one must have intelligence to solve complex mathematical equations also, but chess’s intelligence relies more on intuition – not straight number crunching. In math, in general, the “rules of the game” give no leeway. Presented with a math problem, all computers had better come up with the same solution. There is no individuality involved. In chess, the rules allow for much greater leeway, and hence, much more opportunity for individuality, and error. In fact, it is often stated that one cannot win a chess game, the opponent must lose it. This is true (though of course, in order to win, one still must be capable of capitalizing on the accumulated a!
dv!
antages).
Thus, it seems to me that a computer crunching numbers is conceptually much different from a computer playing chess, and that playing chess well involves some form of intelligence (more on this below).
The following portion of a game was played today (05/29/2000) – I am presenting it in both the newer algebraic and classic chess notations, so that more people will be able to follow along (hopefully).
[b]Algebraic notation[/b]
1) e3 e5 2) a3 Nf6 3)c4 Qe7 4)Nc3 Nc6 5)d3 d6 6)Be2 Be6 7)Nf3 0-0-0 8)Qa4 Kb8 9)Bd2 Qe8 10)Rc1 Nd7 11)b4 Nb6 12)Qc2 Be7 13)e4 f5 14)Be3 Nc8 15)Nd5 f4 16)Bd2 Qg6 17)g3 fxg3 18)fxg3 Rhf8
[b]”Classic” notation[/b]
1)PK3 PK4 2)PQR3 NKB3 3)PQB4 QK2 4)NQB3 NQB3 5)PQ3 PQ6 6)BK2 BK3 7)NKB3 0-0-0 8)QR4 KN1 9)BQ2 QK1 10)RQB1 NQ2 11)PQN4 NN3 12)QB2 BK2 13)PK4 PB4 14)BK3 NB1 15)NQ5 PB5 16)BQ2 QN3 17)PN3 PxP 18)BPxP KRB1
My question is, were computers playing:
(1) white
(2) black
(3) neither
(4) both
This question addresses whether one can determine, without observing the actual entities behind a series of actions and/or responses, if a computer or a human was the entity responsible.
Chess is all about actions and reactions. In fact, I contend that anything that can play chess [i]well[/i] possesses “intelligence”. Note that when I state “can play chess [i]well[/i]”, I mean more than simply following the rules of the game (not committing any illegal moves, like castling into check, for example). That is implicit in simply playing chess – well or not. To play chess “well”, an entity must also:
(1) initiate, and defend against, immediate threats (i.e., understand and use tactics)
(2) develop long-term goals (i.e., be able to formulate and carry out far-reaching strategies)
(3) demonstrate an understanding and application of the various theories associated with each phase of the game.
Concerning theories, each of the three phases of a typical game of chess has concepts particular to it. In the opening, important concepts include development, control of the center, when to exchange central pawns and when not to (maintaining or relieving tension in the center), taking the initiative, time/tempo, standard sacrifices, etc.. In the middle game, important concepts include pawn formation, king safety, mobility, control of half-open files by rooks, attack and counter attack, sacrifice leading to king hunt, etc.. In the endgame, important concepts include proper use of good bishop vs bad bishop, advantage of bishops over knights when pawns are on both flanks, advantages of knights over bishops when pawns are on one flank only and the position is blocked, “square of the pawn”, and standard drawing principles, such as the basic king and pawn stalemates and drawing when being a bishop and pawn down if the opponent has a rook pawn and the wrong colored !
bi!
shop, etc.
Chess computers meet all of these criteria (including all of the above listed concepts of the three stages of the game), and therefore, must possess [i]some form[/i] of “intelligence”. After all, it took studying the game fervently for many years in order for me to reach the level of correspondence chess candidate master, yet, I can no longer beat a chess computer program that costs $50 (I have a couple based on the Fritz chess engine - not on the lower-level ChessMaster 3000, 4000, 5000 series).
One of the criteria for recognizing "intelligence" is contingency, and it seems to me that it is easily satisfied. For each move a “player” makes, there are typically 20 to 40 live possibilities from which to choose. Each time “player” A makes a move, “player” B then has numerous live possibilities, and from those, makes a choice (in fact, it has been calculated that there are more possible ways of making the first 10 moves in chess than there are stars in the observable universe). For example, in the ending position of the above game (after black’s 18th move), white has 41 legal moves from which to choose. When white made his 19th move (not listed), black then had 30 legal responses.
Complexity (in the form of small probability) also seems to be satisfied. As a rough estimate, let us suppose that each side had 30 alternatives for each move. The probability of following the same series of moves, at random, to arrive at the last position given (after move 18), is then 30^36, which is a little better than 1 in 10^53.
The final position could have been reached by transposition of random moves also, but still the probability of the final position being obtained is very small. For example, an alternative method would be estimate the probability of each piece being on the particular square it happens to be on after move 18. The probability of the white knight being on the d5 square instead of any other is 1/64. With 30 pieces remaining on the board (and neglecting conditional probabilities – the white knight and a black pawn cannot both occupy d5 at the same time), a rough [i]maximum[/i] probability estimate would be (1/64)^30, or a little better than 1 in 10^54 (very close to the other estimate).
So going through Dembski’s EF, we pass nodes one and two – the chess position is not attributable to law (there is no chess law that says this position must be reached after 18 moves) or chance (if we use a “local” as opposed to a “universal” probability bound: keep in mind, though, that chance can still be the “winner” if the event is not specified).
Unfortunately, I cannot state with certainty that the final position is specified in any sense. This seems a bit odd to me as each of the 18 moves by each player was specified in that of the 30 or so alternatives, a single choice was made – but not just [I]any[/I] choice, an [I]“optimal”[/I] choice - one with the best interest of the “player” in mind. This selection of the good and rejection of the bad was made even though there are no rules in the game of chess that by themselves specify which move to select – only knowledge of the underlying theory can do that.
Ahah! I just remembered that I find specification based on the following:
[quote]"Not only do we need to observe that a choice has been made, but also we ourselves need to be able to specify that choice. It is not enough that one possibility has been chosen and others have been ruled out. We ourselves need to be able to make the same choice. ... In hearing a Chinese utterance, someone who understands Chinese not only recognizes that a choice was made from the range of all possible utterances but also is able to specify the utterance that was made as coherent Chinese speech. Contrast this with someone who does not understand Chinese.
In hearing a Chinese utterance, someone who does not understand Chinese also recognized that a choice was made from the range of all possible utterances. But this time, because the person lacks the ability to understand Chinese, he or she is unable to specify the utterance as coherent speech. To someone who does not understand Chinese, the utterance is gibberish. ... This choosing of one among several competing possibilities, ruling out the rest and specifying the one that was chosen encapsulates how we recognize intelligent causes, or equivalently, how we detect
design." (William A. Dembski, Mere Creation, InterVarsity Press, 1998, p109-110)[/quote]
Since I do speak "chess-ese", I can see the specification (as could other chess "geeks"). I just can't communicate it to “non-chess speaking" individuals.
One might argue that the computer itself does not possess intelligence, but rather merely takes the place of the people who programmed; that a chess computer is no more than a ‘puppet’ and has no intelligence of its own. I disagree with this as being totally accurate in that computer chess programs “learn” from their mistakes: they adapt to situations based on their past “experiences” (and in neural nets, a related topic, the computers do “learn from scratch”). And even if computers were merely “puppets”, they would still be “intelligent puppets”. Can one teach a dog, or a cat, or a dolphin, or a tree, or a bacterium, or a virus, or a rock to play chess well? No. These listed entities don’t have the ability to retain the vast amounts of information needed, or the ability to process that information properly, to play chess well. But computers do have those abilities, which sets them apart from!
, !
and above, the animate and inani
mate objects I listed.
So, after all this, what are others’ views on whether a chess computer possesses some form of intelligence? Have human intelligent agents already created a new form of intelligence that is not biological?
This archive was generated by hypermail 2b29 : Mon Oct 09 2000 - 12:57:41 EDT