New Mexico House committee hears experts on AI risks, transparency and possible rules

Chair Chandler convened a House committee meeting to hear a presentation on artificial intelligence and possible state policy responses and said the briefing was intended to create a baseline understanding for lawmakers.

The presentation, given by Steve Wimmer of the Transparency Coalition and Chris Moore, a former University of New Mexico computer-science instructor, explained what AI is, how models are trained, differences between predictive and generative systems, and areas of regulatory focus such as transparency, labeling, impact assessment and duty of care.

Wimmer framed the policy problem and the group he represents. "We are a group of independent concerned citizens, essentially," Steve Wimmer, Transparency Coalition, told the committee, describing the coalition as a small nonprofit that briefs legislators and drafts model bills.

Moore summarized the central regulatory aim in simple terms: "If you walk out of here remembering 2 phrases or words, 1 is transparency is good, Black boxes, maybe not so good," he said. Moore reiterated that the technology can be beneficial but that opaque systems must be avoided where outcomes have real consequences for people.

The presenters explained key technical points and policy concerns. They said most modern systems learn from very large example sets ("training data") and repeatedly refine models via feedback loops. The briefing distinguished predictive AI (making a choice from known categories) from generative AI (recombining training data to create new text, images or audio). Wimmer and Moore emphasized that the data used to train models and the way outputs are produced are central policy levers.

Committee members pressed on harms, verification and local impacts. Representative Hochman Behel, an attorney, asked about faulty training data and downstream effects: "What assurances do we have that whatever is generated from that initial false premise doesn't just continue to build upon a faulty premise," she said, citing recent courtroom examples of fabricated citations produced by generative systems.

Moore and Wimmer answered with examples of biased or misleading outputs that had been diagnosed and corrected when companies or researchers had access to the underlying data or methodology. Moore described hiring and credit examples in which models replicated historical discrimination unless their inputs and weights were examined and adjusted. He also referenced a recent New Jersey attorney-general memo noting existing anti-discrimination law can attach liability to deployers of automated decision systems.

Both presenters and members discussed specific policy tools that states are using or considering: transparency and data-provenance requirements, labeling of AI-generated content, definitions of "consequential decision" that would trigger heightened duties, impact assessments for high‑risk systems and roles for state attorneys general and enforcement mechanisms. Wimmer said model bills and playbooks have been drawn from other states' work and that New Mexico's proposals draw on models from California, Washington and Colorado.

Committee members noted New Mexico's demographic distinctiveness and asked how vendors and deployers would show that models were tested on representative populations. "It behooves any deployer of such technology to make sure that the underlying datasets are representative of the population that they're serving," Wimmer said, adding that transparency language should include information on the populations used to train or validate models.

Presenters and legislators stressed that technical fixes and oversight must balance innovation and risk. Moore said some uses—movie recommendations or trivial search summaries—may not require the same transparency as tools used in hiring, lending, health care or housing. He urged human-in-the-loop controls in consequential contexts and independent testing where companies market decision-making products.

No formal votes or ordinances were taken at the session; the meeting was a briefing and question-and-answer period. Committee members directed staff to continue work on legislation and scheduled follow-up briefings: Chair Chandler said the committee plans a Friday session with the Department of Justice, the Department of Finance and Administration and Legislative Finance Committee directors to discuss federal funding mechanisms and related legal context.

The presentation flagged several concrete issues for lawmakers to consider as bills are drafted: how to define and regulate "high-risk" or "consequential" AI; what transparency and provenance disclosures should require (including dataset descriptions and weighting); whether labeling and provenance requirements must accompany AI-generated political advertising (a disclosure law was cited as an existing example); and how to ensure vendor testing reflects New Mexico's population.

The committee will continue work on draft language that members said should aim to enable useful innovation while guarding consumers, workers and patients from opaque automated decisions.