For the new Congress, an AI proposal from the last Congress
Members of the bipartisan House Task Force on AI want to ensure U.S. leadership while mitigating the threats to privacy and safety.
The last Congress spent some solid effort to create a set of principles and recommendations for dealing with artificial intelligence. Members of the bipartisan House Task Force on AI want to ensure U.S. leadership while mitigating the threats to privacy and safety. Rep. Don Beyer (D-Va.), who’s a member of the task force, joined the Federal Drive with Tom Temin to discuss.
Interview transcript:
Tom Temin: And is this really I mean, it’s called a policy analysis. Is it full of legislative recommendations or more guidance should Congress decide to act on AI?
Don Beyer: Well, it’s more than 250 pages long. So it’s both. I think from the beginning, what our goal has been is to try to react in real time to the policy needs that artificial intelligence is creating on privacy on this misuse, hallucinations. And so we have many, many, many different pieces of legislation in the report that we recommended. There are 24 members of the task force — 12 Democrats, 12 Republicans — very bipartisan. We worked together really well last year and they gave a chance to each one of us to say what are the legislative priorities we believe should be pursued right now. So they’re almost all of them reflected in the report.
Tom Temin: And they apply to the governments own use of AI as well as to the industry itself? These proposals?
Don Beyer: Very much, but both ways. It’s really interesting that one of the recommendations was that we continue to invest in this — the National Institute for Standards and Technology, who have created the blueprint for good AI use. A responsible, ethical, moral, authentic AI use. But then another piece of legislation says, ‘Well, if we’re going to do that and we need to insist that every federal agency that’s contracting out for AI, apply or insist that ethical framework is used.’
Tom Temin: Got it. And is the sense of the committee in this task force that something fundamentally has changed with the emergence of the large language model versus the older forms of AI that were not all that controversial or even not all that well known?
Don Beyer: In fact, one of the throwaway lines from the authors of AI Snake Oil is that once we start using AI in our everyday life, we stop calling it AI. For example, spell-check is just AI, but we don’t think of it that way anymore.
Tom Temin: Sure.
Don Beyer: Same with the way you manage inventory. Just machine learning, computer science. But again, we don’t think about it. What’s really resonated with all of us are two things. One, as you point out, Tom, is the large language models. The fact that you can use ChatGPT or Gemini or Claude and write these beautiful papers or even do computer coding. That’s a sea change difference. The other piece is how medicine and science specifically, but businesses writ large are using these huge neural networks to solve problems that we’ve never been able to solve before.
Tom Temin: And in talking about Congress itself and considering new policies and here I’m quoting from the early parts of the report, ‘Congress should adopt an agile approach that allows us to respond appropriately and in a targeted, achievable manner.’ That’s not Congress’ strong suit, being agile and fast to address big issues.
Don Beyer: You’re such a cynic of.
Tom Temin: Just skeptic, I would say.
Don Beyer: You’re absolutely right. But also it’s not often that Congress or maybe ever the Congress gets ahead of the American people. Tom, one of the things we’re trying to do is the opposite of what we did with social media. Where for 25 years, we basically just let it laissez-faire roll ahead. The only piece of legislation we did on social media was to say that citizens couldn’t sue them for getting the information wrong. And so they sort of had carte blanche to do whatever they like. We want to make sure that there are actually frameworks and guidelines for what’s permissible and what’s not because we know that with artificial intelligence, some bad things are happening. The most obvious is child sexual assault material where you can use all the image generation of AI to generate way more pornography and way more uglier and more dangerous pornography than ever before. So we need laws around that to protect people.
Tom Temin: Yes. If you look at any social media feed these days, it seems like half of it looks like a graphic production done with AI and imagery and not something really that was taken picture of. So there’s sort of a convergence here.
Don Beyer: I know. One of the many meetings we had last year was with photographers and illustrators, who pointed out that once you’ve scanned all of their work into a large language model, which includes not just language, but also all the visual stuff, nobody needs them because you can go on and say, ‘Give me a picture of Tom Temin doing his radio show in pajamas with big boots and a cigar in his hand.’ And it will come out with that in 30 seconds.
Tom Temin: But they’ll probably get the cigar brand wrong. We’re speaking with a Virginia Democratic Congressman Don Beyer, who’s a member of the bipartisan Task Force on Artificial Intelligence. And now that there is a different administration here, but we should recall that the Trump administration had executive orders on artificial intelligence and the Biden administration had two of them. And if you read them in principle, they’re not all that different. And so do you feel that you can get some cooperative feelings from the Trump administration, toward this idea?
Don Beyer: I very much hope so. It’s interesting. We Democrats expect that Trump will put out his own AI executive order soon. It could be as soon as Monday afternoon and withdraw the Biden one. But we think that there will be an enormous amount of overlap because there is consensus largely. One of the main areas of difference and it’s not profound is how much regulation should there be. The European Union has passed its own EU AI Act, which is widely seen around the world as overly prescriptive and restrictive. Lots of licensing requirements and at least the business community feels strongly that this will dampen any creativity, any real initiative. Our committee has said we want regulation, but we want a light touch. We want to make sure that American entrepreneurs and the creative young people are doing their best without government getting in their way. But we also don’t want the Wild West. We don’t want terrorists, organized crime, child sexual assault, pornographers, etc., Just have free range without any regulation. That’s not acceptable.
Tom Temin: And looking at the federal use of AI for governmental operations. Right now, they’re pretty much operating without statute and on what administrations say in the executive orders. What do you envision could be maybe the first round of legislative proposals for federal government use for its own purposes of AI?
Don Beyer: I think the very first piece, Tom, and this is a piece of legislation that was endorsed by all 24 members of the task force is to insist that government agencies, both internally and externally, to the extent that they contracted out, use whatever the current best AI standards are at the time and this, it was in Gaithersburg, Maryland. They’ve always been the gold standard in terms of what whether it’s the length of a second or the length of a foot or an inch. But they have the big challenge right now of defining what’s responsible use of artificial intelligence. There’s a very good framework in place right now, but it’s one that they will ever be improving. One of the other things, though, that the Biden AI order did and that our task force has unanimously agreed to do is to put the Safety Institute at NIST into law. The Safety Institute is they’re looking at the potential downsides of artificial intelligence and how to prevent bad things happening to good people.
Tom Temin: Sure. And out of curiosity, how did Congress get up to speed on all of this? Because as you point out, there were hearings a couple of years ago when people didn’t know what Facebook exactly was, precisely. You have studied computer science. You’ve studied personally robotic process automation. But you’re kind of the rare bird there in Congress.
Don Beyer: Yeah, we’re lucky. The two chairs, the Republican Jay Obernolte got a master’s in machine learning from Caltech. The Democrat, Ted Lieu, was a computer science major at Stanford. So we have some smart people with good backgrounds. And the people, that 24 on the task force, are all people who wave their arms saying, ‘Please appoint me. I’m really excited and interested in this.’ We don’t expect that all 535 members of Congress have the same level of interest or background, but our job is to educate them as best we can and also put bills together that they can support.
Tom Temin: And as Congress convenes a new now as the 119th, there’s still no federal budget for 2025, no year appropriation and quite a few other big issues really hanging. Can this fight its way to the top of the agenda, do you think, or near the top of the agenda in the next few months or at least in the next year?
Don Beyer: Tom, I hope so. I pray so. We actually passed out about eight of these AI bills in the last three months. They came out of the Science and Technology Committee unanimously, but there wasn’t time to get them on the House floor. My hope is that with a relative peace in the U.S. House, that we’ll be able to produce this legislation much more quickly, get it over to the Senate and then on to the new president’s desk.
Tom Temin: And it looks like you’ll have a speaker in place for a reasonable amount of time since it takes nine to depose them now instead of one, right?
Don Beyer: Exactly. And we only had one round of voting, not 15. Tom, one of the most interesting pieces of the AI Task Force report is the recommendation to democratize artificial intelligence. Now, we don’t have all the power in the hands of the four big large language models: OpenAI, Anthropic, Microsoft, Google. And so there’s the CREATE AI Act, again, wildly bipartisan. We’d have the National Science Institute create a large language model that’s available to every American citizen.
Tom Temin: Sort of an open-source free version, you might say?
Don Beyer: Exactly. With curated databases, one of the great concerns is that as with any database, if you put garbage in, you’re getting garbage out. When they built ChatGPT-4, they scanned the entire internet, 6 trillion words. So a lot of them are not valid words, are not valid ideas. So the notion of a curated database that does its best to get rid of bias would also be helpful to be part of this.
Tom Temin: And by the way, in the task force work was the ascendancy of China in this whole area part of the background thinking?
Don Beyer: Yes, our intel chiefs, people like Mike Turner in the House and Mark Warner in the Senate, deeply believe that it’s important for us to be there first with artificial intelligence. The second Joe Biden AI executive order just recently put out was mostly about racing to build data centers and large language models in order to beat China, that there is not just so much commerce involved, but also so much national security involved in being able to be the dominant force there.
Copyright © 2025 Federal News Network. All rights reserved. This website is not intended for users located within the European Economic Area.
Tom Temin
Tom Temin is host of the Federal Drive and has been providing insight on federal technology and management issues for more than 30 years.
Follow @tteminWFED
