There are two reasons to be a little sceptical about reports that a computer has finally passed the Turing Test. A little background: this is the one set out by the computer pioneer Alan Turing to decide mathematically if computers could ‘think’.
The reports suggest that 33 per cent of judges were convinced by a programme which persuaded them that the computer they were communicating with was a Ukrainian teenager called Eugene. That was above the threshold set by Turing back in 1950 so – if it really has been passed – they will be breaking open the champagne in Artificial Intelligence circles.
One reason to wonder is that this is actually ‘Turing Lite’, a version of the test designed to be more entertaining to meet the requirements of the American inventor Hugh Loebner. Loebner made the first really concerted attempt to encourage engineers to take the Turing Test, offering a prize of $100,000 to anyone who could build a computer that could pass it.
The Loebner Prize Competition in Artificial Intelligence took place for the first time in 1991 in Boston, with the philosopher Daniel Dennett chairing the judging panel.
Dennett pointed out the immediate problem, which was that Turing’s test was extremely tough and there was absolutely no chance that any existing computer was going to pass it. So he lowered the bar. There were ten judges and ten terminals and they would all spend fifteen minutes with each terminal. Six would be computers and four would be human and the judges would have to work out which. Even then, Dennett did not expect to have to hand out any prizes. This latest experiment only exposed the judges to each terminal for five minutes each.
Back in 1991, Dennett lost his nerve during the actual judging and went to the office to knock-up a certificate just in case. In the event, he gave out three. “The gullibility of the judges was simply astonishing to me. How could they have misjudged so badly?” he wrote. Often, as it turned out, they were simply doing what Turing suggested and giving the machines the benefit of the doubt.
You can read more about this and the background to the test in my e-biography of Turing.
But the issue for Eugene is how much the judges gave the contestants the benefit of the doubt, especially if they are Artificial Intelligence enthusiasts. How much did they ask themselves, not ‘is this a human being?’ but ‘is it possible that I could be convinced that this is a human being?’ Not the same at all.
But there is another reason to take a look at this, which is the implication that the media draws from the experiment – about robots overtaking human beings.
The truth is that, in many areas, that has long since happened. The victory of IBM’s Big Blue computer over chess champion Garry Kasparov in 1997 was the crucial moment. There are some things that computers will always do far better than humans.
So there seems to be a broader debate emerging about what it means to be human in the light of the Turing Test. The Turing Test never claimed to be able to verify anything metaphysical, but that is where the debate is going. It is really a debate about what it means to be human and whether machines can be in any sense human too.
It isn’t about conventional intelligence, but love, care and generosity. Turing believed that intuition was computable. That remains the heart of the test.
So we need to update the test in the light of this. So forget the Turing Test, for a moment. What we need is a Boyle Test (let’s call it that for the time being). It needs to show whether those elements that seem uniquely human can be displayed by a computer. In other words, can you develop a relationship with one?
Before you say you can’t exactly marry a computer to find out – there seem to be huge swathes of the population who appear to be able to fall in love online or over the phone. Well then, let’s test the proposition.
So here is the Boyle test. Is it possible to fall in love with a computer? Could you forma relationship with one? Would you want to stay in touch? That will need more than a random collection of mildly autistic responses from a pretend teenager.
Find out more about the Turing Test in my ebook Alan Turing: Unlocking the Enigma.