Santa Fe Institute Professor Cris Moore will testify to lawmakers on July 29 about responsible AI. (Photo by Cressandra Thibodeaux, with the background artificially expanded by Source for publication purposes).
On July 29, New Mexico legislators on the interim Science, Technology & Telecommunications Committee are slated to meet in Los Alamos and hear testimony on “responsible artificial intelligence” from several experts, including Santa Fe Institute Professor Cris Moore.
A computer scientist, Moore’s interest in AI dates to at least 2016, when he noted an uptick in discussion about its rising use. Computer scientists, civil rights advocates, legal scholars and others were concerned about using algorithms to make decisions in the criminal justice system, he says, but also about their rising use for myriad “consequential decisions, things that really affect people’s fundamental rights, their lives, their livelihoods, their opportunities: hiring, loans, leases, home-buying and so-on.”
Moore had initially thought he’d contribute to that discourse as an academic, but conversations with State Sen. Peter Wirth (D-Santa Fe) connected Moore to the late state Supreme Court Judge Charles Daniels, who in turn put him in touch with New Mexico’s Administrative Office of the Courts. From there, Moore worked with the University of New Mexico’s Institute for Social Research to evaluate an algorithmic criminal justice assessment tool used widely in the state. That led to data-driven work published in 2022 assessing the potential outcome of ongoing proposals to expand pretrial detention. That research found those proposals would lead to a very high percentage of defendants being unnecessarily and unfairly detained.
In an interview with Source, Moore said decision makers and society need to consider both sides when thinking about the use of algorithmic tools in criminal justice.
“We need to think about the false positives — where we think that someone is dangerous even though they’re not — and we detain them, and the false negatives, where we release somebody, and then they do something really serious.”
As he’s worked on these issues, Moore said, he’s learned “just how complicated the system is, and the stakes of the debate…This started out as a question about mathematics and statistics, and it’s turned much more into a question about human beings and human systems, including the judicial system. How should the systems be run, and what does it mean for those systems to be fair? What does it mean for those systems to be transparent?
This week’s AI legislative hearing comes amid ongoing government discussion of AI, near and far. In May, New Mexico’s Public Education Department published its AI guidance for K-12 education. Last week, the Trump administration released its AI “action plan.”
The following interview has been edited for concision and clarity.
You delivered a presentation to lawmakers five years ago and listed four questions to help evaluate the fairness and transparency of AI. Does that list still hold? [The list, summarized: How does the algorithm work? Can we validate its performance independently? When should humans be in that process? And what does the data really mean?]
Cristopher Moore: Often now, what I lead with is: ‘What data does the algorithm use?’ Because that already raises a lot of issues. Is it using my social media? Does it know I’ve applied for a job? For an apartment? That gets into lots of questions, including privacy. I think I would like to know something about what it does with this data, which, in some cases, gets into more technical questions about its inner workings. And yes, I would like to know, ‘Hey, does this thing actually work?’ Has any independent entity tested the vendors’ claims?’ There are a lot of AI systems out there being used to make important decisions, or to advise human decision-makers about important decisions, that have never been independently tested by anybody— we’re just relying on the vendors’ claims, and I think that’s a pretty dangerous situation.
The good thing about AI is we can measure it and observe it and get a clear sense of what mistakes it makes. But again, that’s something vendors are not incentivized to advertise.
What do you plan to tell state lawmakers?
There are things that state legislatures can do to try to ensure that AI is used properly, that it is accountable, and that it has some degree of transparency. I’m not against AI, but I am against opaque systems that their users and the people affected by them don’t understand; [whose] vendors are not willing to open up and tell you how they work; and where they haven’t even been independently tested by anybody.
I think it’s appropriate for state legislatures — and a lot of them are doing this — to draw some guardrails and to say, ‘Well, you know, if you’re going to make these so called consequential decisions about people, then you should be open to having some independent studies done.’ Or at the very least: You should make people aware that they’re interacting with an AI. One of the things that it would make sense to ask for is just disclosure. If you’re going to apply for a job, and the company you’re applying to is going to use an AI to screen the job applicants — maybe you’re applying for a job with particular skills and maybe they will use an AI to read your resume to try to certify you have those skills — at least let the applicant know that’s happening. That’s the least, to me, that’s the floor: Don’t hide the fact that an AI is being used to make a decision about you.
President Trump’s new AI ‘action plan’ intimates that states with burdensome AI regulation could lose out on federal funding. Is it worth it for New Mexico to lose funding to create those guardrails?
Well, of course, I’m strongly against that policy. You probably know that in the Big Beautiful Bill, [Republican Texas U.S. Sen. Ted Cruz] had proposed a 10-year moratorium on state AI laws and, after some back and forth with [Republican Tennessee U.S. Sen. Marsha Blackburn], that moratorium was removed by a 99-to-one vote. So I think the good news is: This is not a partisan issue, at least not yet. I’m worried that it’s becoming partisan, but it shouldn’t be partisan. In fact, some of the model laws we’re looking at come from Utah. Utah put together an AI policy office. A friend of mine works there, and one of the things that they were tasked to do is develop guidelines for mental health chat bots. So, it’s like: ‘We have a hideous shortage of mental health providers; Is this maybe a good kind of frontline thing to help people?’ Some people are like, ‘Yeah, that’s a great idea.’ And some people are like, ‘Oh, my god no.’ But it seems clear that if these things are going to be used, they should make it absolutely clear to the patient that they were talking to an AI and not to a human. And moreover, the AI should look out for things where [it will say]: ‘OK, this is above my pay grade. You really need to talk to a human about this.’
Colorado’s law is considered one of the most comprehensive and requires developers to disclose information about their AI systems as well as its potential risks of discrimination. Would New Mexico benefit from a similar law?
Our House Bill 60 in the last session was inspired by the Colorado law, and the Colorado law is still on the books, although some people tried to get rid of it in their last legislative session. I think there are things we could do to make that kind of bill better. One of the objections to the Colorado law is that it puts some responsibility on the developers of these tools, but it also puts a lot of responsibility on the deployers — the users of these tools. I think there was some legitimate concern about, ‘What if I’m a small school district using some AI, or what if I’m a rural clinic using AI? Are you saying I need to do some big, complicated study that I don’t have the expertise to do?’ I think putting more of the responsibility on the developers is probably a good idea.
New Mexico’s Public Education Department recently published its K-12 AI guidance, which includes ethical awareness as one of its literacy points, but also notes the use of AI to help with building lesson plans and such. Is this potentially problematic given the degree to which some AI systems tend to fabricate information?
It’s vitally important for people to understand that these chat bots are not trying to tell the truth. They’re also not trying to lie. They’re not trying to be deceptive. They’re also not trying to say accurate things about the world. They are simply emitting strings of words that look statistically like the strings of words in their training data.
There are basic skills of writing a book report; or reading something and synthesizing it; or reading two opposing essays on a topic and synthesizing an understanding of the arguments. If humans lose those skills, I think we’re in real trouble. I don’t mind outsourcing long division to a calculator — I’m not quite so old fashioned as to think that everybody needs to know long division. But as we outsource more and more of what it means to learn, to explore, to distinguish truth from falsehood…that certainly makes me nervous.
This post was originally authored and published by Julia Goldberg from via RSS Feed. to get your news feed on Nationwide Report®.
