From Philadelphia To Pallakkad
Garry Kasparov would concur with W C Fields in that he would also, on the whole, rather be in Philadelphia. The latter preferred it to being dead, and Kasparov would be nostalgic about his win against Deep Blue Version I there, just one year and one era ago. Of course, Kasparov is also raving to see the colour of Deep Blue IIs silicon insides. If IBM allows him a rematch, he is likely to beat Deep Blue Version III quite convincingly. Which would retrieve his honour, but never, ever, erase the impression created by Deep Blue IIs matchplay win last week in the Big Apple.
One cannot recall the last time a mathematical thought-experiment made it to primetime but Conan OBrien conducted a fudged Turing Test on his late-night show and the audience cracked up as the computer out-talked the humorist. In fact, that particular version of the Turing Test is unlikely to be passed yet, but in the limited area of strategic games, computers have been passing Turing Tests for years they routinely outperform the best humans at backgammon, go, Chinese checkers and draughts.
Alan Turing was the genius who conceived maximally powerful computers, now called Universal Turing Machines (UTMs), out of imaginary, infinitely long rolls of paper and an I/O device to read and manipulate binary code. His definition of intelligence was unaba-shedly behaviourist, based on the logic that intelligence is only jud-ged via observation, not through telepathic transmission of intent.
In 1950, he suggested that any computer that fooled human observers into believing it to be human would be intelligent, regardless of whether it thought in the same manner as humans. The Boston Computer Museum formally tried the Turing Test in 1991, with eight smart programmes involved in a chat with eight humans. PC Therapist III, a silicon psychoanalyst, fooled the humans for the longest period. Its authenticity was increased by its mildly sexist Freudian patter, which confirms that it was written by a standard computer geek.
Humour, paradox, music, art, mathematics, are all grey functions dependent on right-brain creative impulses. These are all activities where talented human beings outperform Boolean behemoths. But till last Sunday, chess was also included in that range of creative activity. Although the better programmes outperformed 99 per cent of humanity, the very best grandmasters used to consistently outplay machines using superior intuition to counter superior number-crunching.
Philosophical speculations about the murder of chess with the blunt instrument of number-crunching appear somewhat premature. Chess has 10 to the power of 120 possible positions, far more than the total number of atoms. At Deep Blues speed, it would need roughly 10 to the power of 95 years to see them all. Thats googols cubed more than the universes estimated age of 18 billion years (less than 10 to the power of 11).
Human beings play chess by a mystic combination of pattern matching and visuo-spatial projection that compensates for superior calculation. In effect, top GMs filter critical paths of possible tactics out of the labyrinth. According to a study conducted on former world champion Max Eawe, even the best players dont look faster than one position per second. So, top-end human chess intuition is equivalent in thinking power to an amalgamation of an idiot program with a calculating ability like God (200 million times faster than a human being), as Viswanathan Anand described Deep Blue.
Deep Blue and other chessplaying programmes look at all moves and countermoves (known as ply) to a certain specified depth of moves or a time cutoff. They value the positions analysed according to weightages assigned by a set of strategic rules an expert system constructed by their programmers.
One obvious hole was the horizon effect which short-circuited earlier programmes. Wily human beings figured out themes which became apparent only over the programmes calculation horizon. Deep Blue, however, looks 40 ply deep, far deeper than human calculation can reach.
The other problem is inherent in the nature of chess. There are six pieces with varying abilities to control space and the king is of infinite value since checkmate is the objective. All evaluations of better placed pieces, control of territory and others are secondary to king safety which fluctuates unpredictably.
So general strategic rules often break down and its a fuzzy situation. When strategic rules are invariant as in go, draughts, computers are magnificent since there are no exceptions to their expert systems. When they are even fuzzier as in contract bridge, which involves communicating with a partner, machines are abysmal. Chess lies in between.
The fuzziness is why chess programmes rapidly decline in strength. Human beings, with their self-correcting ability, rapidly work out weaknesses and head for specific positions that the programme always mishandles. By denying Kasparov a record of Deep Blues weightages, IBM did weigh the odds heavily, since he had only six games to crack the logic. Thats why I suspect that not just Kasparov, but an elite of 50-odd players will consistently beat Deep Blue once its weaknesses are apparent. Even tinkering by the software team aided by GMs Benjamin and Fedorowicz will not necessarily help.
The major gains from this match come in unquantifiables. First of all, IBM and Kasparov have created a street level awareness about computer technology that didnt exist before. The programmers have gained insights into parallel processing and pattern matching that could lead to huge numbers of repetitive and difficult tasks becoming easier.
Computers still dont pass the Seshan variation on the Turing Test. They cant impersonate citizens of Pallakkad who, according to Seshan, can be classified into cooks, crooks and violinists. Machines cant cook, but they prevent food burning in the microwave and they run chemical tests which differentiate billions of very similar compounds for drug companies. They can match fingerprints, signatures, payrolls and DNA records even though they cant raid the till. They cant compose good music but they can test for possible plagiarism. They cannot write love letters for some lovelorn nerd like an electronic Cyrano De Bergerac, though they can write some of the most (unintentionally) funny poetry since Lewis Carroll.
They can simulate nuclear explosions and aid research into fusion, thereby obviating the need for further tests. They can handle weather-prediction with greater degrees of accuracy than human beings. They are pretty good at trading financial markets too. These are applications that require the same sort of pattern matching and parallel processing skills Deep Blues team would have learnt. In addition, their inability to get bored makes them perfect as Air Traffic Controllers and Maglev Train Drivers, both jobs where human error arising from concentration lapses can be disastrous. Maybe this match and its technological fallout will take machine intelligence one megaflop closer to successful ration cards applications in Pallakkad.
Top-end human chess intuition is equivalent in thinking power to an amalgamation of an idiot program with a calculating ability like God (200 million times faster than a human being)