Saturday, 20 May 2017

The HAL test

I’m sure you’re all familiar with the famous “Turing test”. It is named after Alan Turing, one of the greatest ever Englishman, a pioneer of computer and a man who Winston Churchill himself said did more to defeat the Nazis than any other single person. Turing devised his test as a thought experiment to see whether a person could tell whether they were talking to a computer or not. The idea is that if the computer passes the test, then it’s intelligent.

I think about the Turing test all the time, especially when I’m using the chatbots to talk to my mobile phone provider or my bank. On the one hand I don’t really care whether I’m talking to a person or to a bot so long as they can tell me whether my data package will work in the Ivory Coast or what a SWIFT code is in general and what my account’s SWIFT code is in particular (although they were unable to explain why someone needed a SWIFT code to send money to and couldn’t just use my e-mail address). Anyway, I think about the Turing test when I’m doing that sort of thing. But I do find myself wondering from time to time whether I am talking to a person or not since in most cases the scripts that they have to follow are so narrow that they might as well be a machine.

I wonder if it is going to work the other way round in the future? I wonder if the chatbots will set up a HAL test to try and find out if they are talking to a robot or a human being? Whereas in order to pass the Turing test computers have to answer questions like “how are you feeling” and “what’s the weather been like where you are recently” and “isn’t always the same with Arsenal, trying to walk the ball into the net”, to pass the HAL test people will have to answer questions like “how many different ways will this protein molecule fold given that catalyst” and “if Elon Musk leaves for Mars tomorrow and it takes him nine Earth months to get there then how much older will he be when he lands”?

Frankly, we have no hope of passing this test whatsoever. So while computers will be able to fool us that they are human, there’s no way that humans will be able to fool their computers. I anticipate anti-human discrimination just around the corner: we’ll get shunted into the infobahn’s slow lane will the chatbots accelerate away to infinity in the overtaking lane. Next time you want to find out how much it costs to post a Christmas card to Malta by second class mail, you will undoubtedly be better off getting your chatbot to talk to the Post Office’s chatbot than trying to talk to it yourself.

No comments:

Post a Comment