Wednesday, May 28, 2008

Turing test 2.0

We all know about the Turing test (an excellent philosophical overview of which can be found here). It's intended to be one measure of progress of artificial intelligence - at least of the weak variety. What I'd like to see is a variant of the Turing test, which I suspect might be even more illuminating than the original.

Basically it would involve three individuals: two having a conversation (A & B); and a third human observer (C). A and B could be human, or they could be AI, and C doesn't know one way or the other. The trick would be for the human observer, C, to pick which, if any, are AIs.

I'd imagine the AIs would have to be capable of passing the first Turing test to qualify for Turing 2.0, and other conditions of the Turing test would apply, such as passing only typed messages etc.

I wonder whether two AIs conversing would follow a conversational different path compared to talking to a human? For one, this could minimise the impact of dirty tricks played by AIs to encourage the human to steer the conversation. It might also reduce - or at least change the character of - unexpected turns in the conversation that might be a result of speaking to a n unpredictable human (and a human who knows they're trying to spot an AI).

Besides, wouldn't it be fascinating just to see what two AIs would chat about? Do they have any opinions on Big Brother? Who do they pick in the upcoming election? What do they think of the other AI's position? Would they become friends? Could they become outraged? (If so - that would be tremendous: an AI becoming 'outraged' at another AI.) Could it come to blows?

In some ways this parallels a thought I had while studying the philosophy of mind in my undergrad years. Sure grandmaster-clobbering chess programs are impressive, but they're surely not much sign of true artificial intelligence - even of the weak variety. However, if the chess program got frustrated at losing its queen, knocked the pieces off the table and stormed out to watch TV - THAT would be a sure sign of intelligence, if you can call whatever causes such strangely typical behaviour in us 'intelligence'.

I think the day someone builds an Artificial Idiot, then we're getting close to building a machine in the image of man.


Post a Comment

Subscribe to Post Comments [Atom]

<< Home