NM needs to better regulate AI, computer science professor says

May 8—Are people wrongly being denied jobs because of artificial intelligence? Are they being kept in jail for reasons that can't be explained?

Computer science professor Melanie Moses says it's happening.

Moses, a professor of computer science at UNM and external faculty member at the Santa Fe Institute, presented to the interim legislative Science, Technology and Telecommunications Committee on Wednesday about the role of AI in today's society. She urged lawmakers to create more transparency and oversight of the emerging technology.

Moses said many companies are using AI and don't even understand how the technology works.

She said that's leading to things like false arrests and people being denied jobs. People of color and rural populations are more likely to be affected by the AI tools in negative ways, she said.

"This is particularly problematic, I think, in employment hiring software," she said. "You might submit a video interview and it's analyzed by AI that has its own history of bias, what they are looking for in employees. And it gives an answer with no explanation to the employer about whether or not to hire this person or that person."

She said state government doesn't really use the software, but private companies do and it's completely unregulated.

Large language models — a type of AI that can mimic human intelligence — just aren't explainable, she said.

"If the model is not transparent and can't be explained, it shouldn't be used in this kind of high-stakes decision-making in our state," she said. "I think that's a line we can draw."

She said she's particularly concerned about incentives for AI to conduct surveillance and gather data on people.

"There's just tremendous incentive for the data that we enter to be associated with us," she said.

She said the federal government isn't moving quickly to address this problem, so it's up to the states.

"If there's no economic incentive or no regulatory incentive for people to do better, often they don't," she said.

She said a lot of states have passed laws that have loopholes, so they're not really effective. She said it's particularly relevant with trade secrets, allowing companies not to explain why someone is denied a job.

Moses said transparency should be a priority for the New Mexico Legislature. She said maybe the first step is government transparency in state decision-making, but lawmakers also need to keep an eye out for private companies.

"In many of the cases that matter to people, it's a private employer that's using this tool, and there's nothing in place that would say that they need any kind of accountability or transparency there," she said.

She also said AI literacy should be a top priority for lawmakers — "finding ways to really engage the public."

The elections AI bill legislators passed in the most recent session, which requires political campaigns to disclose deepfakes, was a commendable move and will help New Mexico stand out as a proactive state, she said, but more needs to be done.

"It's a small step on what's going to be a very long journey," she said.

The players

Moses said there are just a handful of large companies that have the computational power "to drive the next generation of AI."

"It's very hard for smaller players actually to produce a competitive, vibrant, economic ecosystem," she said.

AI technology is rapidly advancing, she said, but not necessarily in the right direction. She said AI can get a lot of things wrong that are hard to detect.

"This is actually a decline in what we want the tools to do because they are getting better at being plausible faster than they are getting better at being right," she said.

Moses said companies that have frontier models — technology that exceeds present capabilities — are the only ones that can make models bigger and faster.

"So if they can define the game as 'get bigger, faster' they automatically win because nobody else can play," she said.

She said alternatives exist, but when companies like Microsoft announce a $100 billion supercomputer in the works, that's where investments go.

"This is sort of the direction the industry is incentivized to go," she said. "And I think it's really important to remember that the rest of us don't necessarily have to be incentivized to go only in this direction."

Moses said there's an opportunity for universities or national labs to build AI models in a way that advances science for the public good.

"In terms of fostering collaboration and economic growth, I really think we're at a point where New Mexico as a whole can be a high performance and AI technology hub," she said.

She said there's also market and economic incentives to do so.

"There are lots of us who are thinking about how do we build a secure infrastructure so we can have trustworthy AI, whether it's for medicine or law and policy, education, national security," she said.

At the same time, huge centers for AI suck up a lot of energy and water, she said, and there's not enough power in the state, or potentially the country, to drive the machines.

She said it's expected that 10% of the electricity in the U.S. will soon go to modeling AI models.

"In an era of climate change, that is astonishing and really not a future we want to go to," she said.

She said there's progress in the green energy movement, but there needs to be a balance between AI using up all that energy compared to advancing it. Or, she said, experts could figure out approaches that don't require that massive energy intake.

"It is something absolutely we should keep an eye on," Moses said.