‘Nanniebots’ to catch paedophiles
Thursday March 25, 2004
Talk bad science
Â· As I sit here, quietly shedding the weight off my fat arse in my Dr Norbert Wurgler caffeine-impregnated SlimFit tights, I find myself bitterly regretting the title of the column. Ok. So here’s one I’m not sure of. Artificial intelligence is being used to catch paedophiles in the form of “Nanniebots”. These are AI programs which hang out in internet chatrooms, allegedly spotting the signs of grooming. They have done “such a good job of passing themselves off as young people that they have proved indistinguishable from them,” according to New Scientist. That’s the Turing test – where a computer program is indistinguishable from a real person – almost passed then; and who’d have thought it, in a program written by a lone IT consultant from Wolverhampton with no AI background. So I call him.
Â· Here’s the problem. Reading New Scientist’s chat with Nanniebot at www.tinyurl.com/2y55h, the excellent www.ntk.net/ (Private Eye for geeks) points out that Nanniebot “seems to be able to make logical deductions, parse colloquial English, correctly choose the correct moment to scan a database of UK national holidays, comment on the relative qualities of the Robocop series, and divine the nature of pancakes and pancake day.” Jabberwock, the winner of last year’s Loebner prize for the Turing test, is rubbish in comparison: try talking to it at www.tinyurl.com/2osgo. But Jim Wightman, the Nanniebot inventor – whose site claims they’ve passed the Turing test – isn’t entering the Loebner prize this year: maybe next year … it’s too buggy. But it’s live on the internet already? Can I test it? Sure. But I want to see with my own eyes that there’s not a real human being connected somewhere tapping out the answers. Jim offers network monitoring software on my computer, to prove it’s connected to the one server. But what about that server? I want to see it working on it’s own without a human, too. Can I come round to Jim’s place? He chuckles … Jim doesn’t keep the conversation datasets on site in Wolverhampton. “I know it sounds a bit Mission Impossible but … ” He’s worried they might get stolen. They’re in a secure facility “with an iron lid under a mountain!” He has no copies. It’s 18 terabytes of data, to be fair. There are copies in the hosting facilities, one in London. I offer to go there. “There might be security issues with them letting us in … ” So here it is. I’m going if I can. I’d love to see it work. If there is an AI academic who wants to come, email me: it could be the biggest ever breakthrough in AI. Or it could be a lot of fun.