Baffled Computers

The preposterous NASA scheme to repair the Hubble Space Telescope with robots - at twice the cost of a visit by humans and probably less than a tenth of the effectiveness - shows the strength of the lingering myth of super-intelligent space robots. And it will always be a myth. Space robots have their uses, but they cannot handle unexpected situations. They can only do what they have been programmed to do. They are mere computer-driven machines, and computers can only compute; they understand nothing. Here is a simple demonstration of this idea. In the following chess position it is White’s turn to play:
	When a human player is White, his course of action is obvious. While heavily outnumbered - Black has three officers and White has none - White’s King is protected by an impenetrable array of pawns. White has therefore only to move his King up and down the back rows, and he can force a draw. But a computer will never see it this way. In a famous experiment, the all-powerful chess-playing machine Deep Thought (that once defeated Kasparov) made the incredibly stupid mistake of taking the Castle with his pawn! This exposed White’s pawns to attack by Black’s remaining Castle, and he soon lost the game. All chess machines that I know of will make the same blunder. In the words of Roger Penrose: ``The human player sees the barrier of pawns and understands that it is impenetrable. The computer did not have that understanding - it simply computed move after move. That is the difference between mere computation and the quality of understanding.’’
It will be the same in space. People who believe that space exploration should be left entirely to robots–including the British Government - imagine that machines are going to become ever more sophisticated until they rival humans. This is the fable of ``hard AI’’, of super-intelligent machinery. And it is likely to remain a fable, for ever outside the realm of science. Tasks in space beyond a certain complexity are for astronauts only.