Google’s AI thinks this turtle looks like a gun
theverge.com
theverge.com
It's a problem because similar AI techniques are being used to classify YouTube videos as advertiser friendly, or not. It's a problem because similar AI techniques are used to classify your emails as spam or ham. To shut down "abusive" Google accounts. To direct and classify search results (and present "facts" on those same). To censor documents stored in Google Docs. To classify people as homosexual or dangerous.
Even if you're lucky enough to get an appeal through to an actual human, it can still take days, weeks, or months to get those AI classifications fixed.
From the article:
"In fact, on two separate occasions, police shot at trucks misidentified as belonging to Dorner, injuring three civilians. “We said [to Palantir], ‘We need an application that can span multiple units within an agency…multiple agencies within a county... and multiple counties within a state,’” says Jackson. “[They developed an application] based on lessons learned from Dorner.” That application, called ClueMan, short for Clue Manager, has just gone live at JRIC."
Palantir or any AI/ML software seller will tout it's system as infallible, and mistakes, as low as they might turn out to be, will have real chilling effects.