Down the rabbit hole again

There’s finally more from the Jim Wightman story, and it’s more of the bizarre same. Wightman responds to a new piece in New Scientist, which says, inter alia:

While analysing the transcripts of their conversations, both Webb and Pryke noticed similarities with a program, called Alice, that is free to download over the internet. Alice is a previous winner of the Loebner Prize, which is awarded to machines that do good conversational impressions of humans.

Not only were many of the nanniebot’s responses identical to Alice’s but Wightman’s software appeared to make the same grammatical errors. Wightman’s explanation for this was that he had been forced to “grab and generate as much knowledge” as possible when he realised that he was not going to be able to demonstrate his AI database. This included borrowing parts of the knowledge bases from Alice and other programs.

In other words – though NS is too polite to say this – when forced, finally, to demonstrate something under anything approaching controlled conditions, he was forced to plagiarise a freely available chat-bot, present it as his own, and then bluster about its shortcomings. Plus ça bloody change. Though this is, at least, something closer to the article that NS should have written when first presented with Wightman’s claims.

They did miss a very useful, pragmatic first step in their little investigation, which is something James Randi typically employs as a first-cut bullshit detector. The step is to have anyone making great claims present their evidence for those claims on their own terms first, before bringing any scientific technique like double-blindness to bear. Partly it’s just an efficiency: if the claims don’t appear to be backed up even if the claimant has complete control of the environment, then it’s pointless going any further. But it’s also a bit of typical Randi wiliness. Watching someone (for example) do a bit of ostensibly-psychokinetic spoon-bending can give an observer – like an experienced magician, for example – plenty of insight into the tricks involved, and then inform any subsequent more rigorous trials.

This story still has nothing to do AI, whatever Wightman actually has, if anything. It’s a story about one man’s desire for attention (in which case it’s a wonder he doesn’t just write a blog, like the rest of us – aha, ahahahahaha) and, maybe, acceptance by a technical and scientific community. It’s a story of internet-enabled kookiness. Wightman’s response to the new NS piece is a study in self-delusion:

The New Scientist and anyone else requiring a more complete proof is welcome at anytime!

The keyword here is ‘more’, because no proof so far has been provided. Not a thing.