Re:AI is going to supercharge surveillance
by notahacker on Monday January 29, 2018 @02:26PM
(#16259964)
The other issue is overestimated AI being generally a lot more dangerous than underestimated AI for the foreseeable future because it's a supposedly unimpeachably impartial intelligence source for unnecessary intervention, even if all the actual force at the end is being wielded by humans.
AI that's super-accurate at identifying potential terrorists might be a threat to me if I'm living in an authoritarian police state, but the police state is a threat to me without AI, and probably somewhat more preventable than AI surveillance. On the whole, I'd probably rather local police and intelligence services used a perfect profiling system, in the extremely unlikely event of such a thing ever existing, to using the guesswork they do now. Much more threatening to me and much more likely to be actively being used is AI that's actually pretty weak at identifying potential terrorists but trusted because it's pretty darn efficient at finding something that's difficult to explain on targets' computers. Same goes for AI databases purporting to be super-accurate crime prediction or credit granting or job selection models that are too intractable for anyone to realise they're basically doing the same job as a guy building a race and class based regression model. And of course where it comes to cyber-battles, the guys on offence tend to have a lot less need to worry about the accuracy of their AI's targeting and whether it's vulnerable to being gamed or just plain rubbish than the guys on defence
AI that's super-accurate at identifying potential terrorists might be a threat to me if I'm living in an authoritarian police state, but the police state is a threat to me without AI, and probably somewhat more preventable than AI surveillance. On the whole, I'd probably rather local police and intelligence services used a perfect profiling system, in the extremely unlikely event of such a thing ever existing, to using the guesswork they do now. Much more threatening to me and much more likely to be actively being used is AI that's actually pretty weak at identifying potential terrorists but trusted because it's pretty darn efficient at finding something that's difficult to explain on targets' computers. Same goes for AI databases purporting to be super-accurate crime prediction or credit granting or job selection models that are too intractable for anyone to realise they're basically doing the same job as a guy building a race and class based regression model. And of course where it comes to cyber-battles, the guys on offence tend to have a lot less need to worry about the accuracy of their AI's targeting and whether it's vulnerable to being gamed or just plain rubbish than the guys on defence