The strange tale of the Mechanical Turk
The Mechanical Turk was a chess-playing automaton built in the 1760’s by the Hungarian inventor Hans von Kempelen, reputedly to impress the Empress Maria Theresa of Austria.
It consisted of a large, box-like chess table, with an ornately costumed mannequin sitting behind it, and intricate mechanical systems for moving the pieces on the table-top chessboard.
The claim made by von Kempelen was that this was a completely automated machine, which could play a decent game of chess against a human player.
An LLM (large language model) is simply a vast collection of textual and numerical data, which is used to generate answers based on statistical probability and laborious comparison.
“The Turk” enjoyed a long and successful career, during which it toured Europe and North America and beat many human opponents—including Benjamin Franklin and the Emperor Napoleon Bonaparte.
After various changes of ownership, the Turk ended up in Philadelphia’s Chinese Museum, where it was sadly destroyed in an 1854 fire.
The mystery of how the Turk could play chess so well had long been a source of fascinated speculation, and theories abounded.
However, it was not until three years after the Turk’s unfortunate demise that the secret was finally revealed in a 1857 article published in The Chess Monthly magazine by Silas Mitchell, the son of the automaton’s final owner, where he explained that the table had been constructed to conceal a small human operator who would come up with the moves.
Throughout the Turk’s storied career, a succession of diminutive chess experts had secretly been employed to supply the brain power from within the contraption.
Thus, the Turk was in reality a brilliant and highly intricate confidence trick, designed to convince audiences that a machine could playing winning chess unaided.
Nowadays, of course, it is universally accepted that the strongest computerized chess engines can easily defeat any human player, but it was only fairly recently that this tipping point was reached—when IBM’s Deep Blue narrowly defeated the then-World Champion Garry Kasparov in 1997, in a six-game re-match after Kasparov had comfortably beaten it the year before. Ironically, this first match had also been held in Philadelphia.
Now, 200 years after the heyday of the Turk, the situation is reversed, as AI tries to trick us into believing that its responses to our prompting questions are “human”—that is, that they are informed by genuine analysis, experience, judgement and nuance.
This is not the case, however. An LLM (large language model) is simply a vast collection of textual and numerical data, which is used to generate answers based on statistical probability and laborious comparison.
There is no real “intelligence” involved—just as it says on the tin, in fact.
The poor old Turk must be spinning in his grave.