US states take a page from EU’s AI Act, but biometrics impact likely minimal

America is grappling with how to regulate AI. A divide has opened up between the approaches of the federal and state governments, leading some like the Hyperdimensional Substack to suggest that the EU AI Act is being replicated across the Atlantic. But for biometrics providers, there are some important differences between state legislative efforts and the EU’s regulation.

At the federal level, the key consideration may be how to avoid constraining the economic potential of AI. U.S. states, however, are taking a more pragmatic approach.

Colorado is the first state off the mark, with SB 205 passed last year. But the Act has not yet been implemented, and does not know how to do so, according to Hyperdimensional. Proposals have also been brought forward, or at least mooted, in California, Connecticut, Iowa, Illinois, Maryland, Massachusetts, Nebraska, New Mexico, New York, Oklahoma, Texas, Virginia.

These bills have common ground in attempting to limit the potential harms of AI, including algorithmic discrimination. Those harms would largely be identified through impact assessments.

Hyperdimensional Author Dean Ball, a research fellow at George Mason University’s Mercatus Center, worries about the influence of the EU on American policy, “The Brussels Effect,” noting that “it is not a conspiracy theory that these algorithmic discrimination bills are importations of major parts of the AI Act.”

He also suggests that the bills have the potential to impose high compliance costs and excessive litigation on American businesses.

Limited implications for facial recognition

One of the most important outcomes of the AI Act for biometric technology suppliers is the prohibition on the use of real-time remote biometric identification in public spaces.

Such measures are notably absent from U.S. state proposals. Instead, they tend to focus on moving the responsibility for ensuring compliance with existing laws to before a deployment is carried out, or what Ball calls a “preemptive approach.”

Facial recognition developers have been taking on demographic differentials even earlier in the process, addressing imbalances in the datasets they use to train their models.

Ball points out that no statistical comparison has been made between the proportion of wrongful arrests that involved facial recognition (eight out of an unknown number) and those that did not, so the problem may be less severe than some suggest.

But as represented in state-level proposals, that is America’s main concern when it comes to regulating biometrics through rules for AI, even if it is not universally held.

Article Topics

AI  |  AI Act  |  algorithmic discrimination  |  biometric bias  |  biometric identification  |  Europe  |  facial recognition  |  legislation  |  regulation  |  U.S. AI policy  |  United States