Are Existing American Regulators and Laws Enough to Govern AI
by Michelle Czyz
My glasses are sitting here on my desk. If someone told me that I had to evaluate them based on the criteria used to examine apples, then I suppose I could technically do it. They're both solids that I can hold in my hand. They both serve a beneficial purpose to me as a human. They both have carbon in them. But is that optimal? Maybe forcing a square peg into a round hole isn't the ideal in this situation. Perhaps a Glasses Review Board (GRB has a nice ring to it) would produce better results. Trying to regulate AI activities and tools using the regulations and governing bodies that already exist because, well, this is all we have on hand assumes a premise that AI is legally ordinary and, indeed, CAN be effectively regulated in this way. One problem is the patchwork nature of it in the United States.
If a company uses a large language model (LLM) or other AI tool to generate content without disclosing that fact to consumers, the government has concluded that it is a violation of the Endorsement Guides (16 CFR Part 255) requiring disclosures of material connections between an endorser and an advertiser; as well as the Rule of Consumer Reviews and Testimonials, 2024, both overseen by the Federal Trade Commission (FTC).
If a different company uses that same LLM or other AI tool to screen out applicants based on race, sex, national origin, age, disability or religion in their hiring practices, then it violates statutes overseen by the Equal Employment Opportunity Commission (EEOC)
Finally, if yet a different company (a bank, for instance) uses the LLM or other AI tool to produce misleading monetary collection statements or otherwise misrepresents consumers' legal rights or balances, then those violations fall under the purview of the Consumer Financial Protection Bureau (CFPB).
All of these violations occurred after the damage had been done and each of them was regulated by a separate entity. While one model can touch a wide range of topics at once—elections, employment, healthcare, consumer credit...etc.—no single governing body or structure regulates the model itself. What is a consumer to do? Quit their job and embark on a massive research project to determine which of the hundreds of federal agencies can provide them relief when they have been wronged in some way by an AI model? Hire attorneys and begin a lengthy and costly court case?
Alternatively, when we consider another governing body, the Food and Drug Administration (FDA), they do provide ex ante review of products (assessing potential outcomes and impacts of a policy, project or product before its implementation or introduction to the market). Additionally, when you have a drug or a medical device, it is clear that those products fall primarily and squarely under the purview of the FDA. The same cannot be said about AI models in the U.S. No single regulatory body, no method to take on harmful actions until they actually cause harm, and no consensus on how it should be regulated.
The quilt that is the American legal system also includes state and even local bodies and regulations. Texas, Illinois and California, in addition to having their own FTC, EEOC and CFPB-adjacent entities have introduced AI-specific laws designed to address emerging challenges associated with the new and rapidly growing technology, but there is no standardization.
On the other hand, the EEOC has a robust knowledge of employment law. We can say the same about the CFPB vis-a-vis consumer protections and the FTC vis-a-vis commerce. If the U.S. created a new AI governing body, would the tradeoff in sector depth for overall AI model coverage be a foregone conclusion? Perhaps a single governing body with emissaries from each of the major regulatory bodies in the federal government is in order as a means to ensure standardization in regulations and enforcement? Maybe this structure would allow us to keep up with emerging capabilities of AI and protect consumers and companies before it's too late.
