AI in the U.S. Congress

0:05Welcome to Practical AI. If you work in artificial intelligence, aspire to, or are curious how AI-related tech is changing the world, this is the show for you. Thank you to our partners at Fly.io, the home of changelog.com. Fly transforms containers into micro VMs that run on their hardware in 30 plus regions on six continents. So you can launch your app near your users. Learn more at Fly.io.

0:43Welcome to another episode of the Practical AI Podcast. This is Daniel Whitenack. I'm the founder and CEO at Prediction Guard, where we're safeguarding private AI models. And I'm joined, as always, by my co-host, Chris Benson, who is a principal AI research engineer at Lockheed Martin. How are you doing, Chris? Doing great today, Daniel. Excited about today's guest. I'm thankful in this particular weekend that the government has given us a holiday. So we're about to have a long weekend here in the U.S. And to kick in the long weekend, we've got with us the current congressman from Northern Virginia, who is Don Beyer.

1:25Welcome, Don. Daniel and Chris, thanks so much for inviting me on the show. Yeah, well, we're super excited to have you join us. I think Chris and I were talking before the show. It's just encouraging to see how you're engaging with the subject of AI and encouraging that there are people like yourself in our government and I'm sure governments around the world as well thinking very deeply about this topic. Could you give us a little bit of your background in terms of particularly how you started getting more and more interested in kind of science and AI policy? Well, not to go too deeply, but when I was in high school and grade school, I just loved math and math puzzles.

2:11When I was a kid, Scientific American, there was a guy named Martin Gardner who had a big puzzle at the back of Scientific American every month that I loved doing. And when I was in high school, I thought I was going to be a physicist. My dream was to work at the Niels Bohr Institute of Theoretical Physicists in Copenhagen. And then I figured out in college I wasn't smart enough to do that. So I became a car dealer instead and spent most of my professional career, although my mom didn't think it was a profession, using math that I knew in fourth grade. Once I had division, Dan, I was okay for the business world.

2:45But I was always interested in it. And when I got to Congress, I was privileged enough to be appointed to the science committee. You know, the science committee is not everyone's first choice. Like nobody gives you money because you're on the science committee. But I found it fascinating. Not only because of climate change, because we got to do the oversight of NASA. We were the first ones to see the pictures of the black hole. When they figured out gravitational waves, we had those scientists come talk to us. It was just really fun and interesting. So I got to do a lot more science than I'd ever done before.

3:20Some years ago, well, actually like 40 years ago, I'd heard a graduation speech by one of those 60-minute guys, Eric Severide. And this is like late 1970s. And he talked about how much information we were generating every year and our inability to see the patterns in that information. It was just too much. And it's always been in the back of my mind that we were being overwhelmed by information. So maybe eight, ten years ago, when AI was really starting to catch gear, there was a Coursera course. I'd taken a couple of Coursera courses. My first one was on gamification, which I thought was pretty cool.

4:00And so I took an AI course on AI. And the first three weeks, I just found totally fascinating. The idea that you could do signal amplification, that you could use mathematical formulas and linear algebra to progressively get closer to actual connections and relationships. But then I got to the first exam and literally I say I got a zero on it. It was worse than that. I didn't turn it in because I couldn't answer a single question because I didn't know Python. I didn't know Java. I had never taken linear algebra. And I just put it aside. And then two and a half years ago, our local university, George Mason University, you know, one of the things you do in politics is you tour all the new sites, any place you can cut a ribbon, right?

4:44So we were cutting the ribbon on their new innovation center in Arlington. And I was just really intrigued and actually jealous. And so as a throwaway, I said, could I ever take courses here? And they looked at me sort of funny, didn't really answer me, but I got the lady's card. And I wrote her later that night and said, I'd really like to sign up. And I was late for the filing deadline, the application deadline, but they waived it for me. And I ended up taking pre-calculus in that spring. And again, I'd taken all that 50 years ago, 56 years ago, but it was like taking it all over again. And so in the meantime, I just finished my sixth course on object-oriented programming.

5:23and I'm coding now for the first time in both Python and Java and getting ready for the other 11 courses, which as long as I live long enough will give me a master's in machine learning. That's fantastic. Ironically, before I move on to a question for you, I took that same Coursera certificate course and it was hard. It was a hard course. You did much better than I, I'm sure. It was tough. It was a tough course. and I also had to bone up on some skills that I had long since lost or not touched on. It's a fantastic story. Just to back up for one second, I actually first saw this story that you're telling us here when Ian Bremmer had interviewed you and I was following his social media releases and was very inspired by that, especially because it's so common for the public just to assume Congress doesn't understand science, doesn't understand technology and AI.

6:19And there you were doing it. And, you know, especially leaping unabashed at the age you're at, which in my view is a plus here, because we and we have a lot of folks, I'm in my 50s, that are my age and older that follow the show. And I really wanted to bring this out. I'm kind of curious, were you nervous about it? And, you know, in terms of diving into this, you know, there's this perception that even I experience at my age, that AI ML is a young man, young woman's game. And we're all kind of a little bit on the older side for it and stuff. Did you have any fear of diving into this topic? How was your head at in terms of making that step?

6:57Chris, I should have been more afraid than I was. You know, I originally graduated in 1972, a degree in economics. And I worked for a year or so. And then I decided that I didn't know what I wanted to do. I was wandering around. So I thought, well, I'll go to med school, but I didn't have any of the prereqs. So I went back and did the whole pre-med thing in about 12 months. And at the time, I had figured out that it was just sailing. I was competing against 18, 19, 20-year-olds, and I was 23. And I'd worked for a year, I was married, and I crushed, you know. I just felt like I dominated. So I thought maybe I would again, that all these kids were, you know, they're too busy trying to figure out who their next romantic liaison might be and how much they could drink.

7:44Boy, was I wrong. The kids that I've been in school with are so serious, and they have these great technical backgrounds from high school, and they were the ones crushing me. So it was really a great exercise in humility for me. And I honestly say, I did well in college and taken courses here and there over the years. I've never worked harder in a course than I did on this last Java course. I typed 193 pages of notes. Wow. It's funny you say that. I'm trying to teach my daughter right now to start taking notes in class and instruct. She's, oh, dad, and stuff like that. But I'm like the good students.

8:19That's what they do. Well, I have two packed notebooks from the Python course. I look around the classroom with the other 80 kids. I'm the only one with a notebook. A lot have laptops open, and some have their phones, and some are asleep, the whole range. And as you've been diving into this subject at a more hands-on level, I'm wondering, you know, on the one side, you're a part of and participating in conversations about AI and increasing conversations about AI on the government side and policymaker side. And then on this side, you're kind of in the weeds, so to speak, of, you know, creating literally programming and in Python or whatever it is.

9:06What's been maybe surprising to you about like how AI or machine learning is developed at a hands on level, at a practitioner level versus the perception at the policymaker level? Is there anything that's stood out to you or things that have surprised you? Before answering that, though, you touched on something that has always fascinated me. I don't know if you've read Kim Stanley Robinson's Mars Trilogy. It's a wonderful book. It got me through a gubernatorial campaign, 15 pages a night. And one of the things in it that I really took away from it was the idea of both leadership from the balcony and leadership from the field.

9:49That one of the lead characters in it would spend a year or two managing the planet Mars from like a general manager level, a presidential level. And then he'd go out in the field for two years and live in a tent by himself looking at Mars biology. And when I was in the car business, just to be mundane once again, I'd always spend two weeks every summer working as a technician in the shop, as a mechanic. By the end of the third day, I was cursing as bad as they were, and I'd hate management by then, by the third day. And, you know, but I'd come away from it at the end of those hot summer days with cut hands and a real appreciation for what it was like to work on cars all day long, replacing and water pumps and trying to figure out where that obnoxious knocking noise is.

10:34And now I'm finding the exact same parallel with artificial intelligence and serving on the AI task force and the AI caucus, the various committees that we have. That is really fun to see. And I'm still very much a rookie, a baby in this field. But I can imagine what two or three years later a senior software engineer is doing, the people that are building these wonderful models right now that understand how the neural networks come together. On the other hand, trying to figure out what do we do about deep fakes? What do we do about hallucinations? What do we do to protect our electoral systems?

11:10You know, to look at the policy side also. And I'm blessed to have people like, I have a wonderful tech fellow, which means some other foundation is paying for him for a year, who's an MIT computer scientist who worked in AI. He actually knows the math and the computer science and the hardware of it to help with me on the policy side. That's great. I'm looking forward to diving into the policy side. But before we go there, how have other members of Congress received your dive into this topic with such an intensity? Do they tend to come to you? Do they look to you? Are you seen that way? and what hopes might you have of members of Congress really digging in, whether they're in the House or the Senate, digging in to this topic with some sense of willingness to recognize that it is a huge topic for our future.

12:01Do you have any hope for that or do you think it's going to kind of stay this way? Chris, I'm embarrassed to say this, but I think the primary reaction has been amusement. Especially when I was doing single variable, multivariable calculus, I'd bring the homework to the floor because sometimes we'd have long vote sessions, especially during COVID when we had to vote by proxy. So people would come on over and say, what are you doing? And then they'd look at it and get a little anxious and walk away from me. So no, I don't expect many other people to be going back and taking undergraduate and graduate courses in it.

12:36But I do think many, many members of Congress are trying to read everything they can about it. There's an abundance of AI books out. You see them all over the place. And people are trying to read every article they can. We have had just a myriad visits on the Hill from people that are experts in the field. Just this last week, we had a number of people on the AI safety side, people from NIST game, for example. That's very encouraging to hear. And we've had people from industry and people from academia. And, you know, Stuart Russell was there a week or two ago from Berkeley. He wrote apparently the classic textbook on it.

13:15Mark Andreessen came a couple of weeks ago to give us his techno-optimist manifesto. So we've been listening to as many people as we can in order to try to develop good policies. Well, Don, I think some of us who are, you know, observing our U.S. government with things coming out like the executive order and, you know, you mentioned NIST. I know I've read a bit of some of the guidelines, best practices, some of what they're digging into. Could you help those of us who aren't well-versed in the things that policymakers are involved with around the topic of AI? How would you categorize those? What are the main focuses and what are the main activities that policymakers, congressmen like yourself, what are those main activities that are going on?

14:06And how should we kind of be expecting some of that to trickle down to us as practitioners or kind of hit our desk, so to speak? I think there's a tidal wave coming at us right now, because it seems like every group, every committee, every caucus wants to have a little AI specialty. So, for example, I'm on the AI caucus, which is pretty large right now, bipartisan, almost all of it, which is really encouraging. And so we're doing the education piece for other members and their staffs, especially their staffs. And for those maybe in an international context, because we do have international listeners, what is a caucus in terms of the AI caucus that you just mentioned?

14:51Well, the first AI caucus is open to all members of the House and Senate. And literally in earlier years, we'd have 10 or 12 members and six would show up for lunch. Now, 150 show up. Wow. A lot of them staff, but really interested in the speakers and what they can learn. And then the individual, we have smaller groups, like half of the Democrats belong to the new Democrat caucus, which sort of defines itself as being pro-innovation, pro-business, pro-trade. And so they have their own AI caucus. The progressives, I don't think have one yet, but they're really interested in it. And then probably the most important one is the Speaker, Mike Johnson, and the Democratic leader, Hakeem Jeffries, appointed a 24-person task force for this year to try to look at the 200 different pieces of legislation that have been introduced on AI and focus it down to the handful that we should pass this year that would be building blocks for the years to come.

15:46You can probably elegantly put it in four or five buckets, but clearly the deepfake problem, not just deepfake, but the whole copyright plus problem from music and illustrations and photographer and text and voice, obviously, Scarlett Johansson, that piece. Then there's the whole piece about generative AI and what can we expect and what can we rely on from the large language models that are springing up. The whole safety concern, which one of the, I think, most encouraging things. By the way, all of this is on the background of social media and the fact that we've done nothing in the 25 plus years of social media except make it impossible to sue them, Section 230.

16:29And no one's been able to come up with an agreement on how to modify 230 to allow people to be held accountable without crashing the whole internet, making it just endless lawsuits. So we're trying to get ahead of that. First of all, humbly, Congress will never be ahead of the American people. But we want to get ahead of where Congress typically is by looking at significant legislation. It's really good to hear some of this. And I don't think, you know, I'm sure that that's accessible to people if they know where to look. But, you know, I think Daniel and I are learning a lot from you here today in terms of how this works.

17:04Compared to previous technologies that we've seen over the decades, this is a bit of a different beast, AI. It's going much faster. It is likely to have a much more profound impact on, you know, work. certainly in the industry i'm in warfare across the board even you know what it means to be a human as you go forward in time to some degree with these changes happening you know we're getting big news in the ai space every week you know a couple of weeks ago as we were less than two weeks ago as we record this open ai announced uh chat gpt for omni and that alone relative to its previous version of the model changed how people are using AI in day-to-day life.

17:47You have the entire open source arena with Hugging Face. There are over a million models there. This thing is happening so fast. How do you envision Congress trying to get an appropriate handle on that, whatever that means to you, on such a fast-moving, expansive topic? I struggle as a citizen to envision how that even happens. And I've been waiting to ask you that question ever since we agreed to do this. Chris, you're not suggesting the Congress act slowly, are you? I would never do that to a congressman. That is a really hard question because, as I've discovered in my nine and a half years there, it moves glacially.

18:29Yes. You know, that our founding mothers and fathers by building in a Senate and a House, the competing chambers, and you add a filibuster in and a one person hold and a Senate that I respect as part of the founding compromise, but a very small fraction of Americans elect a big fraction of the senators. Like 30 % of Americans elect 7 % of the senators. And it's amazing we get anything done. And then you almost sometimes need what they call a trifecta, where the one party controls the House, the Senate, and the presidency to do any major legislation, like the Affordable Care Act or the Inflation Reduction Act or Donald Trump's Tax Cut and Jobs Act.

19:08Those happen under trifectas. So all we can do is keep this as bipartisan as possible and then probably deal with whatever emerges as the largest downsides. We're not really having any big downsides yet. We talk about the threats. I mean, there are very obvious things. The whole CSAM issue, making an undressed Taylor Swift or sex videos for underage teenage girls that never participated in them, that those are very real threats. And we struggle to know who to hold accountable for them. But in general, the hope is that by talking about it, looking at these 200 plus bills, looking at the plethora of bills that have been introduced at state levels, I understand more than 50 just in California, that we figure out the handful that actually make a difference and actually protect people.

19:58And by the way, I'm very impressed by the president's executive order, Biden's. Turns out it's the largest executive order in American history. They've hit all of their benchmarks so far, their timelines. And maybe the most important thing is they set up the Safety Institute at NIST, now led by Elizabeth Kelly that's staffing up. So finally, at the federal government level, there is a group that is specifically charged with dealing with the safety and trust issues. So we've seen that executive order. I think we had a previous show where we talked through certain pieces of that. Really encouraging to see some leadership there.

20:34How would you view kind of the more international side of this in terms of how the U.S. and our policymakers are proceeding forward versus policymakers across the world? How do you view that from your perspective and what conversations are going on as related to kind of our positioning within that? and the role that AI plays globally. Daniel, I think it's a really important issue, and I think there's lots and lots of conversations. I think I've had no fewer than eight meetings with actual European parliamentarians who have been putting together their EU AI Act. And just in the last three weeks, one big dinner and a long session during the day with those same people and their lead technical staff explaining how the EUA, their act is working and how it differs from ours.

21:31You know, the shorthand that's an oversimplification is they describe themselves as a regulatory superpower. And we are all committed to innovation. And so we're not licensing algorithms or giving permission to do certain things. But we also know that we have to be there in the UK when they talk about their Bletchley Doctrine. We're going to Japan for their stuff. So ultimately, in the middle run, we need to have something like a Geneva Convention on AI. This is especially true to the extent that we can engage China in it. We know China is concerned. Obviously, they're investing hugely in it, but they're also concerned about the safety parts of it.

22:11And we all have to come together, right? That's exactly what I was about to ask next is, and we often bring it up on the podcast, is kind of the safety. You have all these things pulling against each other with tension. You have the innovation just driving forward constantly, as we've talked about. The understandable safety concerns that we all have, which we have increasingly over the last few years been talking about, and which our audience is also demanding. There's a lot of concern out there. With looking at the international balance and tensions, we have Russia doing what Russia is doing with Ukraine.

22:50We have China and Taiwan and all of these things. Ukraine is the first war that is becoming increasingly AI-driven in terms of the technologies being used. China is an AI superpower along with the U.S. And we're all talking about this need for us and the Europeans to work together. But there's always the concern about bringing everybody on board. I love the idea of the Geneva Convention that you just mentioned. if all the major powers can get on board. How do you envision all these different tensions pulling against each other, possibly working out and getting the motivations of all of the Western countries led by the US with Russia and its sphere of influence in China and its sphere of influence coming together?

23:37Do you have any either aspiration or expectation on how you see that coming together over the next few years? Chris, it's probably more aspiration than expectation. And with you coming from the defense industrial base, you know how important our warfighters think this is. Indeed, I do. The chairs of our intelligence committees, the chairs of our armed services committees, they very much do not want us to be behind China. There's a lot of debate about how much human agency should there be in the use of kinetic force. Can you have machines deciding who to target rather than an Air Force pilot, even if he's sitting in a room with levers in Colorado Springs?

24:17At least that's a human being saying, let's take out that car, take out that building, rather than letting a drone decide who to attack and the like. And then there's space and the weaponization of space and the role that AI plays in that. My hope is that we will have some renewed arms control in the days ahead. It's been pretty sad the last couple of years as Russia has progressively withdrawn from arms control agreements and China has been unwilling even to sit down and talk to us. But sooner or later, if we're going to make the world a safer place, we need to talk about all the nuclear weapons on the planet and at the same time talk about artificial intelligence, too.

24:54And one of the very first pieces of legislation introduced was Ted Lieu, Ken Buck and I on prohibiting letting an artificial intelligence algorithm make the decision to launch a nuclear attack on another country. That has to be the president of the United States and the chairman of the Joint Chiefs of Staff and the secretary of state, human beings making decisions of that magnitude. And yes, they can use all the data they want, but a machine can't decide. I totally get that, 100%. But I'm often kind of shocked at how many people don't understand that, the idea of nuclear weapons. A few years ago, I was with the CTO of Lockheed Martin at the time, who's since left, and I was doing an event in London, and I was on the stage.

25:38and I actually had an audience member say, could you talk to us about, with an assumption in the question, could you talk to us about the fact that the U.S. has AI controlling its nuclear weapons and what you think the implications are? And I laughed it off and said that's not the case, obviously. But that was one of those first moments where I realized how much misinformation about these topics was out there. And obviously, since then, that's just gotten more and more. There's so many variations, deep fakes, constant misinformation that AI enables and nefarious intent. Could you talk a little bit about kind of the safety of how to approach that from regulation?

26:24And you're obviously deeply into the topic. You kind of give us a little bit of guidance. Things I can tell my family who are not into AI. Because when we sit around Christmas time and holidays in the extended family, they don't know either. And I'm always shocked that my own family doesn't know. And so we'd love to hear your thoughts on just kind of how to approach some of these incredible misconceptions that people have. Number one is safety issue is how about job elimination? We know that it's going to replace many, many jobs. One of the exciting things is what they call ambient clinical documentation.

26:58You know, doctors and nurses say 25, 30, 50 % of their time is filling out data on the clinical visit they've just finished or in the middle of it. Now there's software, hardware that listens to the conversation between Chris and his doctor and writes it all down. And by the time the doctor or the patient leaves, the doctor can read it and check it. Yep, that's what we talked about, saving an immense amount of time. But then so job elimination and what do we do? We know that that's happened in every revolution, agricultural, industrial information. But we also know that it will probably happen much more quickly now than it ever has before.

27:34So much less time to react and for people to adapt to that change. You know, second level is all the misinformation, whether coming out of a large language model, it's unintentional or the intentional stuff. How do we protect against that? Then there's the whole notion. And some people take this seriously. other people will say, nah, that with desktop ability to generate DNA, to synthesize DNA, and the ability to look up what's a smallpox vaccine, what's the DNA of that, or let's go from COVID-19, let's make COVID-27, the whole notion of bioweapons or other things that can be used based on the information that comes from large language models and AI.