No we haven't passed the Turing Test

The New IT Age brigade will be all in a tizzy over the announcement that a computer has passed the Turing Test. We haven't, not really. We may have passed a silly little game, but we're not even close to an Artificial Intelligence that is indistinguishable from human.

As the BBC article says

The 65-year-old Turing Test is successfully passed if a computer is mistaken for a human more than 30% of the time during a series of five-minute keyboard conversations.

First, the test is pretty silly when it is limited to a 5 minute conversation. Even though Turing originally proposed this duration, it is clearly a low threshold. You might only begin to get suspicious in 5 minutes.

Second, the competitors used a clever tactic (some might say they cheated): they claimed to be a 13-year-old Ukrainian boy (an unfortunate choice right now), thus limiting topics and vocabulary.

Third, the terms of the "Turing Test" are not what a reasonable person-in-the-street would expect them to be, nor are they in the spirit of what Turing intended. To fool judges 30% of the time in 5 minutes is hardly definitive. A true Turing Test is when you can't tell it is a machine i.e. the judges' call should be random, i.e. 50% of the time.

Finally, all bots use the same tactics: they obfuscate, they appear confused, they echo back to the interrogator, they stall as long as they can. None of them actually understand the conversation. in this sense this Turing Test is bollocks: a bot that passed the test would still be useless at being your butler. Turing clearly intended that the test should be passes without trickery. We are so far away from that it is absurd to even predict when it might happen.

Note that we aren't able to even see the winning transcripts. I bet they make pathetic reading.

So chill. The real Turing test is when a computer is indistinguishable from a human without any silly constraints. AI has been twenty years off for fifty years now, and still is.

It reminds me of this Turing Test:

See also an excellent post from TechDirt

Syndicate content