Turochamp EmulatorChess · History · Artificial intelligence
An idea from 1948.
A different way to play.

02 / The Turochamp story

Before the machine could play.

A set of instructions, a chessboard and a human doing the calculating.

Two people, one paper machine.

Alan Turing and David Champernowne devised Turochamp in 1948. Their challenge was not simply to encode how pieces move. It was to make a procedure choose a reasonably useful move from many legal possibilities.

Material was one clue. Activity and safety were others. Looking ahead could reveal that an attractive capture would lose a more valuable piece in reply. These are recognisable chess ideas, but expressing them as instructions made them available to calculation.

Doing the work by hand

Turing manually executed chess instructions rather than running an original Turochamp executable. The recorded game against Alick Glennie, usually dated 1952, makes that experiment tangible: the procedure had to cope with a real opponent, not just a carefully selected position.

Glennie won. The game is interesting precisely because the machine’s play was imperfect. It shows a practical attempt to turn judgement into a sequence of operations.

Replay the Glennie game

What survives?

The Turing Digital Archive holds a typescript, with handwritten corrections, of Turing’s contribution to Digital Computers Applied to Games, published in B. V. Bowden’s Faster Than Thought in 1953. It is an important source for studying his chess procedure.

That is different from possessing a finished program and the precise environment in which it ran. A reconstruction must decide how to translate descriptions into data structures, legal-move calculations and stopping rules.

Why reconstructions can disagree

Kasparov and Friedel’s account describes modern reconstructions disagreeing with parts of the Glennie score, including the first move. Their programs preferred e3, while the recorded game began with e4. Manual arithmetic and interpretation are both part of the historical problem.

Try the opening yourself.

On the play page, choose Black for a new game. Watch the reconstruction’s first move, then compare the candidate scores. The site exposes its calculation rather than silently forcing the recorded opening.

A useful distinction

Replaying a historical game reproduces a sequence already known. Running a reconstructed algorithm asks a fresh question at every position. Both belong here, but the replay must not be mistaken for evidence that the engine generated those moves.