The Imitation Game
Turing · Computing Machinery and Intelligence · 1950
Turing finds the question "can a machine think?" too vague, and proposes to replace it with a test. A judge converses in writing, without seeing them, with a human being and a machine; the judge must guess which is which. If, over the exchange, the judge can no longer unmask the machine, on what grounds deny it thought? Turing sees in it less a proof than an operational criterion, and wagers that machines will one day succeed at it. What we now call the "Turing test" is a loose version of it: programs sustain a conversation well enough to fool us, but is the test really passed, or did it never measure the right thing?
Does being unable to tell a machine from a person by conversation prove that it thinks, or does behaviour never prove the presence of a mind?
What it does in the debate
Supports The substrate does not matter
We never grant thought to others except on their words and gestures: to demand more of the machine is to apply a criterion we apply to no one.
- Of another's mind I have no direct access: I observe only conduct, words, reactions, and it is on that basis alone that I grant them an inner life.
- If, over a sustained exchange, nothing distinguishes a machine's answers from a person's, the criterion I use everywhere else is met.
- To deny it thought nonetheless presupposes a further criterion, one I require of none of my fellows and would be hard put to state.
Therefore So whatever realizes the conduct of a thinking being must be held to think: the substrate is no part of the criterion.