Inside America’s AI Action Plan

0:28Welcome to the practical AI podcast. behind-the-scenes content, and AI insights. You can learn more at practicalai.fm. Now, on to the show.

0:48Welcome to another fully connected episode of the Practical AI podcast. I'm Daniel Leitnack. I'm CEO at Prediction Guard and joined as always by my co-host Chris Benson, who is a principal AI research engineer at Lockheed Martin. And in these episodes where it's just the two of us, we try to analyze some of the things that have come out in recent AI news and maybe touch on a few practical elements that will help you level up your machine learning an AI game. And really excited today, Chris, to talk about something that is very intriguing, somewhat maybe confusing for some people, exciting for some people, maybe producing a little bit of uncertainty, some, yeah, just a lot of mixed feelings maybe around this.

1:43We're talking about America's AI action plan, which was just released by the White House here in the US. So yeah, this just came out not too long ago. Obviously, we're just kind of reading through the initial release and some reactions. But I say there's been some mixed feelings. It's kind of generally rolled back some things that have maybe been happening in the US around AI and regulation. maybe introduce some other things that like there are elements of the action plan that I find really inspiring and encouraging. And so, yeah, it's maybe an interesting mixed bag for many folks. But yeah, I'm sure it's come up in your circles and in discussions, maybe even, you know, with family or around town that AI has an or America has an AI action plan.

2:40What are your initial thoughts. Yeah, since the White House released it, I've definitely had people asking what I thought of it and stuff. And I know both you and I had read it a while back as it was released. And we're just getting to a point where we had a chance to talk about on the show. And leading into it, I'm, of course, going to do my usual disclaimer. Of course. That because we're talking about stuff that can potentially touch government affairs and military, I am only speaking for myself as always and never ever speaking for any other organization and most especially not my employer Lockheed Martin.

3:14So I wanted to be very clear about that because I do want to keep my job. So this is just Chris Benson on the podcast. Good to know. Yeah. So this is an interesting mixture of a lot of things as you were alluding to. And like you, there were some things I was inspired about. There's a fair number of things in there that are inspiring, but there's also a lot of stuff that either rolls back other aspects that we probably, that I think a lot of people would agree in a nonpartisan context would be advisable, some of the protections in certain areas. And it is, I think, you know, my single biggest gripe, It is a starkly partisan document in many respects.

4:05It feels like it was written by some folks that had some good ideas in there, but were being very careful to make sure it would fit in with the current political climate. And as I read it, I felt like I kept bumping into that filter that the authors would have had on it. So that's my first impression, a highly filtered document with good and bad in it. Yeah, yeah. Well, I'm interested, I mean, kind of to set the stage a little bit. Now, we've talked about other things that have rolled out from the government directly from the White House on the show. I'm just looking back at 2019, if you remember those days.

4:532019, there was a White House executive order on AI. This was during the Biden administration. And we talked about that on the show. It's interesting in that case. So if we kind of roll back the clock a little bit for quite a while, there has been this effort at NIST, the National Institute for Standard and Technology, around kind of AI and AI risk management framework. So that kind of has some hints of regulation. It's a government thing, but it's not official on any kind of regulatory front. And then there was this White House executive order, which I think I might have said Biden. But I think this one that I referenced from 2019 was a Trump executive order.

5:44Now I'm getting all the sequence confused. But that's episode, what, 34. It's back a while. I think I was thinking because I think we did talk about another executive order. We did. Anyway, there's been this kind of rolling, you know, series of executive orders back and forth on both sides, like you say, of the partisan line. What, from your perspective, maybe before we dive into the details, some of the executive orders that have come into place do have some type of regulatory implication in the sense that I remember one of them that we talked about, for example, around model builders and the requirement of certain sizes of models to be regulated in a certain way.

6:37In terms of this AI action plan, what is your understanding, if you do have one or don't have one, of what does it mean that this is an AI action plan? It's not an executive order. And partially, in my understanding, it's meant to not regulate. And so maybe it's just a document. But really, I guess, what is the implication of this more broadly, do you think, in terms of its impact? Yeah, I mean, it definitely reflects this current administration's approach well beyond AI to a larger viewpoint. This one definitely is all about removing regulation, which I think is, I think that there are places where regulation could be beneficial and there are places where regulation might get in the way.

7:30And I think you have to pick carefully where you want to put that. I think, and one of the other things that I've noticed here is that, you know, there's a lot, you know, there's a lot of verbiage in it that kind of on the surface, almost anyone would say, yeah, I'm for that. But we also know if you look at the politics outside of this specific document and specific AI field, what free speech is depends, you know, in terms of interpretation depends on who you're talking to. If you put a Democrat and a Republican in the room and talk about what free speech is, you may get some different answers on what that is and what's implied into that.

8:09And so it's one of those situations where we are left to interpret what the intent of some of these words means along the way. And it's almost impossible to do that without kind of taking in the larger political environment, which sadly, I think detracts from the document. it in a lot of ways. I think it could be better. I'll also note that it doesn't, there's a whole bunch of recommendations in it and there's no funding for any of those recommendations that is noted. And so it's interesting about in the sense of where do you go from here? Whether it's a good idea or a bad idea from my perspective, it's still left with, well, great.

8:52Maybe I like that idea. How are we going to do that now? Yeah, I think as opposed to some other maybe executive orders that had some direct implications right away, this is, again, an action plan. And I guess that's what I was meaning in terms of impact. Like, what impact is this going to have other than being a document? I think if people are out there and maybe one implication that could be thought of is, universities applying for grants, companies applying for SBIRs or other types of government interactions. I think in the current climate of those things getting approved, if it's AI related, it probably needs to kind of tie into this action plan for it to have a greater chance of success.

9:44And tying into that action plan kind of has this trickle down effect to all of these grants and innovation that happen across university and across small business and et cetera, et cetera. And so there is maybe this trickle down effect, but it's not a sort of direct, we're going to fund this or that, and there's this program. But maybe that is more the impact in terms of the climate of what is funded and how that affects businesses and universities or research institutions that get government funding. Yep, I totally agree. It's kind of funny, we've talked about some of the ambiguity as we've gone forward into this, and the good and the bad of removing regulation a little bit, the ambiguity of the free speech component of it.

10:32One thing that I'll say that I like the intent of is the promotion of open source and open weight AI, which are both explicitly pointed out in several points in the document. And I know that we have long talked about that on the show. And so I think that's one of those those inspirational moments, though there's not a lot of how we're going to get there, I must say. Yeah, actually, and just to give maybe, you know, we've talked about a little bit of the impact. We've set it some stage. I love this kind of point that you're bringing up about the open access models, which, of course, I'm very passionate about.

11:12But maybe it would be helpful also for our listeners. There's kind of a bit of a structure to this document. If you're listening to this in your car and you don't kind of have it in front of you, of course, we'll link it in the show notes. But there's kind of these sections and they're split up into, quote, pillars of the AI action plan. So kind of these key pillars. The first pillar is accelerate AI innovation. The second pillar is build American AI infrastructure. The third pillar is lead in international AI diplomacy and security. So there's multiple things under this first pillar. So the first pillar being accelerate AI adoption.

11:57And you've highlighted a couple of those things that might be more on the kind of controversial side related to removing regulations. specifically maybe even rolling back some of those things related to the NIST AI risk management framework, protecting, quote, free speech in AI models. Looking at this, quote, AI systems must be built from the ground up with freedom of speech and expression in mind and be free of top-down ideological bias. So there's that discussion of kind of the free speech piece under the innovation. Then they kind of go to this discussion like you talk about, about promoting open source and open weight AI, which, of course, is one of those elements that I'm really excited about.

12:47some of the recommended policy actions they talk about is ensuring access to large-scale computing for startups and academics, partnering with leading technology companies to increase research, and fostering this sort of environment that creates a supportive environment for open models because they view this as, obviously see this as kind of a part of the future, which I would, of course, agree with. And with open AI becoming open again and releasing open models for the first time a couple of weeks ago, I think that certainly is a trend that we're seeing. But yeah, I love that piece. Of course, again, it's this mixed bag of things that as you go through each of these pillars, you kind of have to parse through.

13:38What else stands out to you, Chris, in this kind of accelerate AI innovation pillar. There's a lot of interesting things there about world-class data sets, the science of AI, interpretability, adoption in government, adoption within the Department of Defense, lots of things in this innovation pillar. What else stands out to you? I think one of the first things that I noticed, It was, you know, so like once again on the surface, addressing data set quality and stuff is great at that at that outermost layer of the onion. But as soon as you dig into it in the document and, you know, people listening to us may be on both sides of this politically, but they immediately go for a political objective directly under that, which is basically kind of removal of DEI concerns.

14:34concerns. And for those of you who may not be familiar with the DEI acronym, it's diversity, equity and inclusion. And that that is a big focus of policy across all fields within the Trump administration. And so that had we not been inserting some of the politics into it, I would have been encouraged to see that under it, you know, but that's with each of these things, you kind of have to inspect it and see how much politics is in it. And not in fairness, not every point that they're making under either this first pillar of Accelerate AI innovation or the subsequent pillars that we'll talk about include explicit politics.

15:13But I know in my first past, that was the first thing was trying to filter through some of that issues and actually get to the meat of it. And then think about, A, many of the things that are suggested are already being done. So it's not new. But for those things that are not being done, how do you get there? And what's the funding to make things happen that aren't already happening.

15:52Well, friends, when you're building and shipping AI products at scale, there's one constant complexity. Yes, your wrangling models, data pipelines, deployment infrastructure, And then someone says, let's turn this into a business. Cue the chaos. That's where Shopify steps in, whether you're spinning up a storefront for your AI-powered app or launching a brand around the tools you've built. Shopify is the commerce platform trusted by millions of businesses and 10 % of all U.S. e-commerce, from names like Mattel, Gymshark, to founders just like you. with literally hundreds of ready-to-use templates, powerful built-in marketing tools, and AI that writes product descriptions for you, headlines, even polishes your product photography.

16:37Shopify doesn't just get you selling, it makes you look good doing it. And we love it. We use it here at Changelog. Check us out, merch.changelog.com. That's our storefront. And it handles the heavy lifting too. Payments, inventory, returns, shipping, even global logistics. It's like having an ops team built into your stack to help you sell. So if you're ready to sell, you are ready for Shopify. Sign up now for your$1 per month trial and start selling today at Shopify.com slash practical AI. Again, that is Shopify.com slash practical AI.

17:19Well, Chris, just kind of tying up this first pillar of the AI action plan, Accelerate AI Innovation, I think a couple potential implications that people could have, you know, we're obviously we can't read the whole thing here on the podcast, but we've just hit a few of the highlights of this first pillar. But I think there are going to be these sorts of debates over what defines bias in an AI model, because there's certainly, there is, and again, I'm just kind of coming at it from a, I guess, a technology standpoint, there could be bias as related to maybe DEI types of things in in one way or another, for example, gender or whatever that might be.

18:10But there's also bias corresponding to very real world implications around technology and security. For example, the bias of one model to be more susceptible to prompt injections than another model. So that is a bias, but it has nothing to do with these other categories. And so I think there is going to be this kind of bias in a sense is a term of art in our world. And in another sense, it's a politicized thing in our country. So I think there's going to be a little bit of tension there. I think also with the kind of rollback of some of this regulation, it's going to put more pressure, I think, on businesses to strike the balance between innovation and the safeguards that they need to put in place because without explicitly being forced to, you're going to see companies, I think, at a very high profile level, make critical mistakes in their AI applications and suffer some pretty hard brand consequences because of this.

19:23And it's going to be, I mean, that in itself, just the commercial pressure is going to force companies to think about their kind of self-regulation. And so companies are going to have to figure out this balance for themselves, I think, as it's not kind of flowing down guidance-wise from standards bodies in the government. Agreed. Yeah. Well, that kind of gets to pillar one, accelerate AI adoption. Pillar two, build American AI infrastructure. Sounds exciting. What stood out for you here? It does. You know, one of the things and I almost I forgot to mention it in pillar one and it applies to pillar two is one of the subtleties I'm noticing here when they're talking about building out AI infrastructure is, you know, we there is already quite a bit of AI infrastructure in the US.

20:14We have a lot of major players with big investments in the commercial sector that all have government and military and intelligence links and things like that. And so this applies across the board. It applies to commercial. It applies to government, etc. One of the things that really stood out to me in some of these suggestions was that there's a subtlety of picking winners and losers here. In most of these suggestions, there is already something in effect that does these different things. It may not be optimized, and that's open for debate. It may be a different organization from how they're envisioning it in the thing.

20:54But there is a subtle picking of winners and losers through the entire document in terms of approach and sometimes explicitly who's responsible and in some cases who they're likely laying it with. And so that's one of the is an overview that particularly applies to the infrastructure section pillar two would be that I can almost envision the lobbying from certain large companies that contributed to the infrastructure section as I was reading through the documents. I'll leave those companies unnamed, but probably our listeners can come to some of those conclusions themselves. So it definitely, I think that there's some interesting influences that are at play here behind the scenes in terms of the specific choices and verbiage being used.

21:45Yeah. And just to give people an idea of some of the things that are mentioned in the Build American AI Infrastructure pillar of the plan, there's sort of creating streamlined permitting for data centers, semiconductor manufacturing, et cetera, development of the grid, restoring semiconductor manufacturing, kind of onshoring that, along with a number of security related things. So, you know, bolstering the cybersecurity of critical infrastructure, creating secure by design AI technologies and applications, a mature federal capacity for AI incident response. So very, very interesting. I find this interesting kind of at the same time, there's kind of a rollback of regulation in the public sector around things like maybe what would have flowed into regulation from either the NIST AI risk management framework or previous executive orders or that sort of thing.

22:51There is this pressure from the government side where they're talking about, you know, making sure and trying to force somehow. I don't know exactly how that's kind of vague in the document, but to ensure that AI systems that the government relies on, particularly for national security, are protected against spurious or malicious inputs. Some of the language from the document, promoting resilient and secure AI development. having the DoD lead in collaboration with NIST, this sort of refining of generative AI frameworks, roadmaps, toolkits. It sounds like a lot of kind of what has happened in the NIST AI risk management framework, but maybe led from a different direction.

23:39So there's kind of this, I'm seeing two things at once, this sort of what was kind of rolling more to the commercial side, but now is being geared more towards government implementation of AI, but also being led from a different direction, being more led from the DOD kind of national security perspective versus like a NIST. Yeah, there is a lot of kind of government-ish in all the things, military and otherwise, that are prevalent through it. And once again, as we're looking through the section, the subtlety of the outcomes that are possible here are pretty big. And I'll give you one example of that.

24:26You know, they mention even though the third pillar refers to AI diplomacy and security and the second pillar, there's a couple of points where they talk about semiconductor manufacturing in terms of streamlining the permitting of that. And there's one that's called Restore American Semiconductor Manufacturing. And that has profound implications on the commercial space and not only on our foreign policy, but on our national security, that listeners of this who may not be very familiar with foreign policy and military concerns, a big strategic reason why the United States is kind of the leader in defense of Taiwan concerns.

25:11concerns Taiwan's responsible for its own defense but we have some obligations there that often come out in the news that people would have seen and a lot of that has to do with the fact that the global semiconductor manufacturing industry is very well entrenched there and it's very complicated to replicate that elsewhere um that that is not a trivial thing to do and so um whether or not we are successful according to this document and doing what it's uh advocating uh it would have implications either way in terms of things outside of AI altogether. And to your point, kind of leading into that, it definitely has a very, you know, different in, you know, shifting.

25:52There's some stuff that's assigned to NIST, but there's also stuff that might have traditionally been with NIST or other similar organizations that are often more commercially focused and moving those into more of a DOD sphere of influence as opposed to that. So yeah, Yeah, there's definitely a policy, an overall policy shift to be felt with the document going forward. Yeah, I'm kind of fascinated a little bit in this AI infrastructure pillar, this idea that one of the last thing they mentioned is this sort of federal capacity for AI incident response. And this is, I don't know, I was just trying to parse through a little bit of this, like, what is an AI incident?

26:35And, you know, some of this terminology, I'm sure, is, you know, intentionally vague in certain places. But at least in terms of this AI incident, they say prudent planning is required to ensure that if systems fail, the impacts to critical services or infrastructure are minimized and response is imminent. And, you know, they want to promote this kind of incorporation of AI incident response actions into incident response doctrine and best practices. So actually, this was another kind of point, like, other than the fact that I was a little unclear what they mean by AI incident response. I did like that this was another element in this pillar that I kind of liked from a certain perspective in the sense that I really believe that there's a lot of people kind of developing, quote, AI applications and automations and that sort of thing without a lot of best practice cybersecurity.

27:37knowledge. And then over here, there's the cybersecurity folks who aren't sleeping at night, because they know all this is going on. But it's sort of, you know, whatever it is, like the executives or the board are pushing this AI stuff forward. They don't want to bring the cybersecurity folks in because that's going to slow things down and the naysayers or whatever. But I do really think that there's a benefit to have this kind of more cross-functional collaboration between cybersecurity and AI development, and actually bake in some of the known AI threats into cybersecurity playbooks, and some of the cybersecurity best practices into AI application development.

28:19I think if you look back, you could make the parallel that when data science was big and hyped, right? There was sort of data scientists building things that DevOps and infrastructure people couldn't support. And there wasn't a lot of cross-functional collaboration on those things. And we kind of work through a lot of that. I think there's a similar thing happening right now in terms of AI development and this sort of cybersecurity industry. So I do like that kind of crossover in a lot of ways that they talk about under this AI incident response piece. It is. And while we're talking about that particular point, you know, and as we were discussing kind of the power shift that you're seeing.

29:05I note that in their third recommendation policy note, they shift who is in charge of that by going to, they say, led by DOD, which is the Department of Defense, DHS, which is the Department of Homeland Security, and the ODNI, which is the Office of the Director of National Intelligence in coordination with OSTP, which is the Office of Science and Technology Policy, NSC, which is the National Security Council, OMB, which is the Office of Management Budget, and the Office of National Cyber Director. And so they are clearly putting defense and intelligence at the front end of that and kind of backloading it with more technology orientation on that.

29:51And that's very different from how we've seen it in the past. They have essentially flip-flopped how that has traditionally been addressed, to your point, which there's potentially good and bad to that approach. It will definitely create a different set of policies when you have a different set of lead agencies with their priorities addressing that. And so how that how that ends up rolling out in terms of the discrete future executive orders and policy statements that come out and how that funding is allocated will be interesting to see in the years ahead. I will also say it will be interesting to see how this rolls out into a next administration down the road, whether that administration be Democratic, a Democrat or Republican administration.

30:42All right, Chris, we are on to the third pillar. The third pillar is lead in international AI diplomacy and security. That is the third pillar of the America's AI action plan from the White House. So lead in international AI diplomacy and security. There are a few things under this pillar here. There's discussion of exporting American AI to allies and partners. There's an explicit call out of, you know, countering Chinese influence and strengthening AI compute export control enforcement, which I think maybe is geared towards some of that. Yeah. How does Nvidia stuff get over to China. Indeed. That's right.

31:38Yeah. Actually, there's two. So there's strengthen AI compute export control enforcement, which is what I was talking about. Then there's another point, which like it's a double click, plug loopholes in existing semiconductor manufacturing export controls and align protection measures globally, ensure that the US government is at the forefront, invest in biosecurity, which maybe we can talk about that one here in a sec, which is, you know, why might that be interesting? But yeah, any sort of first reactions to this set of things? Well, while we have certainly seen current domestic politics in the previous pillars, this is without a doubt the most explicitly political pillar with turning to foreign policy.