How the Trump administration is likely to deal with artificial intelligence

Tom Temin: So what can we expect? Because these executive orders, two of them from the Biden administration, are comprehensive. I think they’re literally the longest executive orders ever written in the history of the presidency. Will they just get tossed out?

Justin Chiarodo: Indeed. No, it’s a great question, Tom. And I think, as with all predictions relating to the incoming administration, we want to be measured in how focused we can be and really understanding exactly where it’s going to land. But there’s no question that artificial intelligence is going to be a really critical priority for the national security and defense community and our government going forward. So I very much anticipate that the principles and foundations that have been laid out in the Biden executive orders are going to get implemented in ways to try and again, speed the acquisition and deployment of these technologies, which are really central to a lot of the pacing threats that the United States is facing today.

Tom Temin: Because we do see administrations take essentially the same ideas from opposite party priors and rewrite them in their own language, but yet they’re in the same domain and often in these kind of not-so partisan issues have similar objectives, you might say.

Justin Chiarodo: Absolutely right. I think a common theme you’ll see here. I would also draw a parallel to domestic sourcing, right. How can we bring back manufacturing capacity to the United States and to allied nations as we see this evolution in the geopolitical landscape? AI absolutely, squarely in that bucket. A real common challenge here is going to be this technology is advancing incredibly quickly, right. Much faster than one would say the acquisition cycle can keep up with it. So how can I put together a framework to speed the deployment of these technologies, which we all agree are important to our defense, while at the same time preserving critical controls around the deployment of these technologies, safety, protecting human rights or individual rights in our in our nation, which we really value. That’s going to be the balance. That’s going to be the challenge that the new administration is going to need to handle.

Tom Temin: And industry, of course, is making a big show. Many cases, contractors of how they’re using artificial intelligence in their standard product offerings. Almost every integrator now has a artificial intelligence-based type of service that they’re offering the government. So what are contractors concerned with in terms of policy and what laws might come with respect to AI do you think? What are you hearing?

Justin Chiarodo: Yeah, there’s certainly a lot of lobbyists that are very busy right now trying to help, I think, shape and communicate positions from industry on this. And how do you do this in a way that is going to facilitate competition on a level playing field and also, I think position the acquisition system to balance the need for really sophisticated technical solutions and quickly with an acquisition system that is historically had a slower lead time that’s been focused more on long analysis and deployment of requirements, very long acquisition cycle. How can the procuring agencies use new acquisition methods, use innovative acquisition methods to speed this up? How do we get those ideas to market? So I think everybody is still trying to figure out exactly what that looks like, making significant investments and trying to make that happen. And a lot of capital to I would say, I think this is an important thing. There’s a recognition that defense, technology and innovative solutions in these domains is where the money is going to go. It’s where the need is going to be. So there’s going to be a lot of interest and commitment to make it happen.

Tom Temin: We’re speaking with Justin Chiarodo. He’s chair of the government contracts practice at the law firm Blank Rome. And a lot of contractors, too, are talking about using AI for themselves in the development of responses to requests for information or requests for proposal from the government. And what is it you feed into your large language model besides the solicitation? Or is it only the solicitation that can generate something that might actually be submittable to the government?

Justin Chiarodo: Yeah, it’s a great question. We as a law firm are piloting AI technologies and how do you balance the protection of information and the security of your system, right. In evaluating these things. So I think there’s still a lot to be worked through as far as systems access and what kind of data is going into tools. Certainly putting in a solicitation in a closed system and evaluating that, let’s say, against existing solicitations that you may have records of, that kind of functionality I think is going to be really helpful to both procuring activities and contractors in figuring out what’s important, right, what matters and what’s really going to drive the requirements and the solutions going forward.

Tom Temin: And you have to be really careful. I’m thinking of what it is your large language model actually produces for you and really need to have people go over it pretty carefully. I myself tested one of the online free experimental generation sites that’s out there, and I won’t say which one, but they had dinner at Mar-a-Lago and it produced something strikingly realistic. But at the same time, really naive sounding if you know the domain in which you created it, which I do. So I was both impressed and unimpressed simultaneously. I think that’s the case, right. These tools are nowhere near something that what comes out of the oven you can slice and eat.

Justin Chiarodo: No, it’s a great analogy. Borrow the oven. Using the oven. Here is a model. It’s not fully baked yet. I think industry and the government are still really figuring out how to do this. You see, and for example, in the NDAA efforts to pilot these technologies to look at how they might work in particular applications. So let’s say operations of the DIR-One maintenance facilities. This is something that’s called out in the NDAA as a place to pilot this technology, take away lessons learned, help that really inform how you’re going to structure security controls, procurement, right. Balancing these policy interests of having a safe use of AI that preserves what we value, right. Innovation, freedom, security against the need to, again, deploy these technologies, which, as you note, are incredibly powerful tools.

Tom Temin: And we also value other transaction authority as kind of a God-given right to industry’s relying more and more on these days. And speaking of the NDAA, what of the hundreds of provisions under Title 8 actually strike you the most? What are your top two or three picks?

Justin Chiarodo: Yeah, I mean, I certainly, looking at AI and software-enabled technology and some of the efforts around funding that, I really think where where’s money going to go when I know we have a, is it the DOGE? Is that the acronym for this effort to promote government efficiency? How do they potentially look at actually appropriating some of the money under the NDAA towards these technologies? So you have DoD, for example, the Office of Strategic Capital, which has about a billion dollars to put to work with loans and loan guarantee programs to deploy advanced technologies. Standing out, I mentioned the pilot program for piloting AI in some domains like contract administration and depot maintenance. Looking at where that money is going, I think to me is really interesting. Security patrols, too, right? Coming around out there, we don’t want this technology to fall into the hands of our adversaries. China, not surprisingly, a lot of provisions directed at preventing them from getting their hands on advanced chips, which you need to run AI, addressing other sophisticated technologies and keeping them out of the hands of our adversaries.

Tom Temin: And getting back to the AI question and using models to generate responses or whatever it is you want to make, there’s also an intellectual property consideration here too. The tool that I used generates all of the sources and citations, and you should probably make sure that if you are feeding something into an LLM for purposes of, say, solicitation or any other kind of public purpose, you need to cite your sources like you would with any other type of document unless you run afoul of intellectual property challenges.

Justin Chiarodo: Absolutely. Want to avoid having this information, lose control of proprietary information. That’s going to be a big one. Obviously, China has been very aggressively coming after IP in the U.S. For decades. You see that with the Salt Typhoon hack that’s come out in the telecom system. So the security controls around how AI is used is going to be a really significant part of, I think, the forthcoming rulemaking that we can expect, right. So that’s protecting both IP of the community that’s involved in populating these models, but also in making sure that this information doesn’t get into the wrong hands.