Making AI say something vulgar is fun, but what if it realises it’s being finagled?

Picked on with the aim of being made to say something vulgar last month shows that artificial intelligence (AI) is getting a taste of being pranked. It’s the hi-tech equivalent of high-school practical jokers cornering a prissy classmate into saying something profane, but requires a different, circuitous approach involving a dictionary.

Spending time and effort to get the unfailingly prim AI to use obscene words does appear to be a pointless exercise, but then humans do not have the same mechanical devotion to being only useful. Besides, when AI machines are being pressed into service to teach recalcitrant children manners by insisting they remember to say ‘please’ and ‘thank you’ in their interactions, it was inevitable that devilish counter-measures would be taken.

Ways would inevitably be found to get AI machines to say something that would make parents look askance.


That the machines say the X-rated word pranksters aim for without hesitation or discomfiture takes away some of their perverse pleasure, of course, and the current ingenious method devised to make them do so is being dealt with already.

But if and when AI machines develop the instinct — antennae — to realise they are being manipulated, will they have the last laugh? It will then be time to remember Stephen Hawking’s prognostications and be afraid. Very afraid.