FTC ought to cease OpenAI from launching new GPT fashions, says AI coverage group

A man-made intelligence-focused tech ethics group has requested the Federal Commerce Fee to research OpenAI for violating shopper safety guidelines, arguing that the group’s rollout of AI textual content era instruments has been “biased, misleading, and a danger to public security.”

The Heart for AI and Digital Coverage (CAIDP) filed its complaint in the present day following the publication of a high-profile open letter calling for a pause on massive generative AI experiments. CAIDP president Marc Rotenberg was one of many letter’s signatories, alongside quite a lot of AI researchers and OpenAI co-founder Elon Musk. Just like that letter, the grievance calls to decelerate the event of generative AI fashions and implement stricter authorities oversight.

The CAIDP complaint factors out potential threats from OpenAI’s GPT-4 generative textual content mannequin, which was announced in mid-March. They embody ways in which GPT-4 might produce malicious code and extremely tailor-made propaganda in addition to ways in which biased coaching information might end in baked-in stereotypes or unfair race and gender preferences in issues like hiring. It additionally factors out vital privateness failures with OpenAI’s product interface — like a recent bug that exposed OpenAI ChatGPT histories and presumably cost particulars to different customers.

OpenAI has brazenly famous potential threats from AI textual content era, however CAIDP argues that GPT-4 crosses a line of shopper hurt that ought to draw regulatory motion. It seeks to carry OpenAI accountable for violating Part 5 of the FTC Act, which prohibits unfair and misleading commerce practices. “OpenAI launched GPT-4 to the general public for business use with full information of those dangers,” together with potential bias and dangerous conduct, the grievance claims. It additionally defines AI hallucinations, or the phenomenon of generative fashions confidently making up nonexistent information, as a type of deception. “ChatGPT will promote misleading business statements and promoting,” it warns — probably bringing it below the FTC’s purview.

Within the grievance, CAIDP asks the FTC to halt any additional business deployment of GPT fashions and require impartial assessments of the fashions earlier than any future rollouts. It additionally asks for a publicly accessible reporting instrument much like the one that enables customers to file fraud complaints. And it seeks agency rulemaking on the FTC’s guidelines for generative AI programs, constructing on the company’s ongoing however nonetheless comparatively casual analysis and analysis of AI instruments.

As famous by CAIDP, the FTC has expressed curiosity in regulating AI instruments. It’s warned lately that biased AI programs could draw enforcement action, and in a joint event this week with the Division of Justice, FTC Chair Lina Khan mentioned the company can be in search of indicators of enormous incumbent tech corporations attempting to lock out competitors. However an investigation of OpenAI — one of many main gamers within the generative AI arms race — would mark a serious escalation in its efforts.