SB 1546 wrong path to AI concerns

Originally published in Portland Tribune.

As a born and raised Oregonian, a proud champion of the “Oregon Way” of solving problems, and, now, the Director of the AI Innovation and Law Program at the University of Texas School of Law, I’m worried that my home state is about to join a fear-based approach to regulating AI.

Governor Tina Kotek has a decision to make: sign SB 1546 and join states that have rushed to pass AI laws or veto the legislation and encourage lawmakers to pursue a more prudent, innovation-friendly path.

This bill requires four things. First, that operators of “AI companions” provide “clear and conspicuous notice” that a user is engaging with AI. Second, that operators prevent Oregonians from using an AI companion unless the operator has implemented “evidence-based methods” for detecting if a user’s chats implicate self-harm or suicidal ideation and prevents the companion from generating any content in those domains. Third, that operators “use clinical best practices and expertise” to assess if a user is pursuing information in those domains. And, fourth and finally, that an operator prevents an AI companion from producing any content that may “simulate emotional dependence on the user,” if the operator is aware of the possibility that the user is a minor.

No doubt that SB 1546 is a well-intentioned bill. Many Oregonians are rightfully concerned about reports of problematic relationships that AI users–especially minor users–have formed with AI tools. Yet, the aforementioned requirements are a combination of poorly defined, technically infeasible, and, unconstitutional.

For starters, the bill’s expansive definition of AI companion is likely to cover AI companies that have no intention of their tools being used for problematic ends. This over-inclusive definition risks chilling innovation and contributing to a growing patchwork of state laws that not only conflict with one another but risk foreclosing technological advances that could lead to safer, more reliable AI.

Next, it is unclear whether the “evidence-based methods” and “clinical best practices” mentioned in the bill even exist. If such proven mechanisms were well-known and popularly available, then AI companies seeking to develop tools for minor users and users experiencing mental health issues would surely be touting such a discovery and broadly sharing them. The fact of the matter is that AI is a highly complex, evolving field. What’s more, it’s highly likely that such requirements run afoul of the First Amendment. The Supreme Court has rightfully struck down laws that compel private actors to share certain messages.

Finally, the broad limitations on which models can be made available to younger Oregonians is shortsighted and, again, possibly unconstitutional. All Americans–including young ones–have a right to access information. AI companies uncertain about how to comply with the vague limitations in SB 1546 may err on the side of preventing young users from using such tools or offer an unhelpful version. That’s a poor way to help young Oregonians learn how to thrive and safely navigate the AI future.

The biggest problem with this bill is that it does not adhere to the Oregon Way of governing, which ChatGPT defines as “a political and civic culture in Oregon characterized by consensus-oriented, pragmatic, and collaborative approaches to solving complex public problem.”

We’re a state that does not run from innovation (or at least we can be). Rather than try to put poorly thought through guardrails on AI, the state legislature should explore means to empower Oregonians to be more savvy AI users. Let’s take a page from Oklahoma and launch an AI literacy initiative. Let’s learn from the three moms who launched “Raising AI” and work to inform parents about how to introduce and monitor their child’s use of AI. And, let’s copy the several states that have created regulatory sandboxes that allow AI tools to be deployed but subject to close oversight and iterative, adaptive regulations.

The temptation to “do something” on AI is understandable. Oregonians should call Governor Kotek and let her know that they’d prefer to do something that is durable, actually empirically-driven, and unlikely to be struck down in court.