FTC Policy Statement Presents AI Companies with a Catch-22

The FTC proposed a policy statement earlier this month expanding the definition of deceptive practices related to AI.  The statement asserts that AI users have an expectation that AI systems return factually accurate results. In the FTC’s view, companies that prioritize anything other than factual accuracy in their AI outputs may be engaging in a deceptive practice. The FTC believes that companies complying with state law frameworks like Colorado’s AI Act may be doing just that. Colorado’s law requires AI companies to avoid disparate impacts resulting from the use of their products. The FTC views this as wrongfully superseding pure factual outputs. So how do AI companies choose which authority to comply with? A recent Steptoe memo discussing the statement notes that companies may be able to disclaim their way out of a catch-22:

“The brief discussion at the end of the Statement provides at least one pathway for complying with both the FTC Act and state laws such as Colorado’s Artificial Intelligence Act: clear and conspicuous disclosures, provided persistently and prominently, that state that the AI system in question is designed to produce outputs prioritizing, e.g., compliance with state law over consumer expectations of accuracy.”

The FTC’s proposed policy statement echoes previous sentiments expressed by the administration. The high profile falling out between the Pentagon and Anthropic hinged on allegations that the company’s AI products were “woke.” While prioritizing “truth” might sound easy, the FTC’s policy may prove more difficult to enforce and comply with in practice. Differences in perspective, opinion, and preferences may largely sway what an AI user considers “true.” It is difficult to engineer outputs which will not diverge from what some users “reasonably expect” the system to provide.