The Question

Can a machine think? The question sounds clear until you try to test it. We cannot open a person and point to a thought. We usually infer thinking from behaviour, language, memory, judgement, and the way someone responds to the world.

So the question has a hole in it, and the hole is not really about machines. It is that we have no way of checking the claim for anyone, including the person sitting next to us, and we get by on the assumption anyway.

Alan Turing's move was to stop trying to fix that. Rather than settle what thinking is, he proposed something you could actually run and watch, and let the answer to the harder question take care of itself. It is a piece of intellectual sleight of hand, done in the open, and people have been arguing about whether it is brilliant or a dodge ever since.

The argument got sharper when something passed for human without deserving to. A very simple program, simple enough that its entire method fits on one page, held conversations that people found genuinely affecting. Some of them did not want to be told how it worked. That reaction, more than the program, is the thing worth understanding.

What it leaves behind is a problem that has not aged: fluent language is real evidence of something, and deciding what that something amounts to takes an argument the evidence cannot supply.