This was a big week for AI-generated deep fakes. Sultan Meghji, who's got a new AI startup of his own, walked us through four stories that illustrate how AI will lead to more confusion about what's real and what's not. First, a fake Biden robocall urged people not to vote in the New Hampshire primary. Second, a bot purporting to offer Dean Phillips's views on the issues was penalized by OpenAI because it didn't have Phillips's consent. Third, fake nudes of Taylor Swift led to a ban on Twitter searches for her image. And, finally, podcasters used AI to resurrect George Carlin and got sued by his family for violating copyrightish law. The moral panic over AI fakery meant that all of these stories were too long on "end of the world" and too short on "we'll live through this."
Regulators of AI are not doing a much better job of maintaining perspective. Mark MacCarthy reports that New York City's AI hiring law, which has punitive disparate-impact disclosure requirements for automated hiring decision engines, seems to have persuaded NYC employers, conveniently, that none of them are using automated hiring decision enginess, so they don't have to do any disclosures. Not to be outdone, the European Court of Justice has decided that pretty much any tool to aid in decisions is an automated decision making technology subject to special (and mostly nonsensical) data protection rules.
Is AI regulation beginning to suffer from backlash? Could be. Sultan and I report on a very plausible Republican plan to attack the Biden AI executive order on the ground that its main enforcement mechanism, the Defense Production Act, simply doesn't authorize the measures the order calls for.
