Lawmakers hear experts urge transparency, limits for AI used in consequential decisions

At a New Mexico House Judiciary Committee meeting, outside experts urged lawmakers to require transparency about how artificial‑intelligence systems are trained, deployed and used in decisions that affect people’s lives.

Steve Wimmer of the Transparency Coalition and Chris Moore of the Santa Fe Institute told the committee that AI systems work by finding statistical patterns in very large datasets and that those data and design choices must be visible to regulators and users. “We are a group of independent concerned citizens, essentially,” Wimmer said, describing his nonprofit’s work helping states draft AI policy. Moore summarized how the systems operate: “If you show an AI system a whole lot of X rays … it will look for patterns in that.”

The experts framed the policy problem as one of transparency and accountability for two distinct parties: developers (the companies that build models) and deployers (the entities that use models to make or support decisions). They urged particular caution where systems make or assist with “consequential decisions” such as hiring, lending or housing. “For these consequential decisions…we don't want to allow black boxes into our decision‑making process any more than we would want a human being who just kind of makes decisions without explaining them,” Moore said.

Committee members pressed on several recurring concerns. Representative Hall described a user experience where generative answers are labeled “AI's answer” but a layperson cannot tell how the result was produced: “I am a little concerned that in this learning process that it can become so assuming…that the human factor [is] being taken out of it,” she said. Representative Hochman Behel, an attorney, asked how regulators could prevent the spread of fabricated or inaccurate outputs; Moore replied that the only practical path is clear data provenance so consumers and deployers can “trust but verify.”

Wimmer and Moore cited real‑world examples committee members raised: a California lawsuit involving a chatbot company after a teenage user was encouraged toward self‑harm; historical cases of algorithmic bias in hiring and credit underwriting; and a health‑care algorithm that used prior spending as a proxy for illness and therefore disadvantaged uninsured or low‑income patients. Moore said in several of those examples companies discovered problematic correlations only because there was enough internal or outside transparency to surface the issue.

Speakers discussed several policy tools already being considered in other states and at the federal level: standardized definitions (for “deployer,” “developer,” “consequential decision” and “high‑risk AI”), requirements to publish training data provenance and weighting, duty‑of‑care rules for deployers, independent testing or impact assessments for systems used in consequential contexts, and labeling requirements for AI‑generated political material. Moore noted that New Jersey’s attorney general issued guidance reminding deployers that existing anti‑discrimination law applies to AI decisions and that a deployer can remain liable even when using a third‑party product: “If you are using AI to help you with those decisions and you're discriminating, you're liable,” he said.

Several lawmakers raised New Mexico‑specific concerns. Several members noted the state’s distinctive demographics — higher proportions of Hispanic and Native American residents — and asked how off‑the‑shelf AI products trained on other populations would perform here. “It behooves any deployer of such technology to make sure that the underlying datasets are representative of the population that they're serving,” Moore said. Wimmer said the Transparency Coalition is helping states draft model language to require representativeness checks and monitoring by deployers.

Committee members also asked about public education and curriculum work. Wimmer said his group focuses on educating legislators and drafting model bills and has not yet built K–12 or college curricula, but that curricula development is an identified gap. Moore offered to connect legislators with local academic groups working on curricula.

The committee did not vote on legislation at the session. Members set a follow‑up informational meeting planned for Friday to hear from the Department of Justice, the Department of Finance and Administration and the Legislative Finance Committee about federal funding mechanisms and the state’s legal framework for navigating federal programs.

As the committee continues its work, the presenters and members emphasized a narrow, risk‑based approach: encourage innovation for low‑risk uses while imposing transparency, impact assessment and oversight where AI influences employment, credit, housing or other consequential outcomes. Wimmer closed by emphasizing the purpose of the effort: “We really want to make sure that we come up with regulation that in fact does support the responsible use of these technologies.”