National Security Commission on Artificial Intelligence: Closing Conversation – American AI Initiative
Transcript
So, we’re really honored to close out the day with one of my heroes, Michael Kratsios, Chief Technology Officer of the United States. And I want to go over a few of the things that Michael has achieved in his last months. So, I use as my benchmark, in the last six months, I think I watched half a season of Breaking Bad and then got antsy and read the rest of the plot summaries on Wikipedia. Here is what Michael has achieved in the last several months. An AI executive order in February. In May, the OECD AI Principles. In June, the AI R&D Strategic Plan. In September, the Nitered Supplemental Budget. What are you planning to do in the next nine months? How are you gonna top that?
Well, first off, thank you so much. I’m extraordinarily delighted to be here. The efforts done by this commission are just absolutely incredible and they’re an incredibly valuable compliment to everything administration has been driving since essentially inauguration. And having such an incredible set of commissioners driving this agenda, I think is, we’re very lucky as a nation to have this group working for the American people. That being said, we’ve been extraordinarily excited about what we’ve been able to drive on, on AI. I think our sort of big, big initiative was to make sure we got the executive order out in February. And now we can jump into that a little bit, but I think, you know, that we have big pillars on how we’re coordinating better R&D across the agencies. There’s a lot of workforce related issues on that, international engagement that we touched on and I think the last piece that we haven’t put out publicly yet, but should be coming out in the coming weeks is our OIRA or essentially, White House Regulatory Guidance Memo to our agencies. So, we’re gonna have sort of a first of its kind and document actually anywhere in the world, where we have a document with sort of legal backing to go out there and tell the agencies how they should be thinking about regulatory or non-regular approaches to AI technologies. So, if you’re in the Department of Transportation and you’re trying to regulate autonomous vehicles, if you’re FAA trying to regulate drones, if you’re the FDA trying to regulate AI powered medical diagnostics, these are all agencies that just think carefully about the implications of artificial intelligence in the domains that they’re overseeing. And you’re gonna see a lot of, sort of the core themes that the Unites States generally thinks about when talking about technology regulation. We want to create environment that incentivizes our next generation technologists to make investments and to make breakthroughs here in the United States. So, but doing that while also maintaining all the values that we hold so dear. So, it’s something we’re gonna put out for comment in the next couple weeks and there’s lots more people in the room that I hope will take some time and tell us what we did wrong so we can fix it.
Well, a lot of the smart people in the room have previously worked in the interagency process. They know how hard it is to accomplish things quickly. You all accomplished a lot quickly. Can you tell me about some of the biggest challenges that you faced in getting the agencies and departments on the same sheet of music?
Yeah, I think what is challenging about AI and I think you’re all sort of experts in that space here so you know, it’s something that touches a wide wide range of agencies. So, if you’re attempting to put together an interagency group that’s gonna come to some sort of interagency consensus on a document that the president of the United States himself is going to sign, that’s a pretty tall order. And I think if we just think of the two biggest equities we have, those security equities and we have sort of civilian equities, and in almost every technology priority that we’re driving. So, if we take a step back and say, what is the main tech agenda of the White House and this administration? It is to ensure American leadership and emerging technologies. So, we want to and have been working towards crafting and propagating national strategies on everything from AI to quantum to 5G and so down the line. And for each of those, we sort of see the world as sort of a two-sided coin. There’s a promote and there’s a protect side of that of that particular technology and for us to come to some sort of interagency consensus, we have to find some balance between those two things. And I think, you know, we had our big event at the White House in March of last year where we brought together stakeholders and opened up the public conversation on what should be in a nationally AI strategy and when we launched that strategy, it took us almost a year. We signed it out in February. So, you can see that it took some time but at the end of the day I think we have a pretty comprehensive piece of work there and something that we’re extraordinarily proud of.
So, I know that the protect part of this is the one that most strongly intersects with the work that the commission has done. I wonder, given that security is just one part of the executive order, if you can tell us what will you think we got right and what we got wrong on the national security side of AI?
So, I think as a commission, I think you guys have done an extraordinary job of kinda wrapping up the very large themes that need to drive whatever our sort of national security posture is on artificial intelligence. So, I think a couple key ones first, that always come to mind to me and you know, I see General Shanahan here, I think being able to integrate artificial intelligence into the way that all of our armed forces do their work, it’s something that’s a really really really hard task and the DoD is a massive agency and everyone has their own equities and their own interests and their own way of doing things. And being and finding pathways to be able to integrate this technology across all the various branches, I think is something that we need to do a lot of work on and I think it’s something we have a great team who are on it. So, I think that’s something that we’re kinda very, very very much aligned on. The other thing that I think is cross-cutting between both the security side and civilian, is the talent question. This desire of how can we continue to bring in the most talented AI scientists and technologists from the private sector and from academia into our federal family here to drive the outcomes that we need. So, continue to drive that. And I think the third piece is something that our director, Tippy, talks a lot about and this is the idea of partnerships. This idea that the world of technological innovation is fundamentally changed. This idea that there’s a couple of very large companies that have all the answers for all of the technology solutions that the DoD is looking for is over and we need to find a way to create better pathways and connections with start ups and smaller companies that can make huge breakthroughs that can deeply impact the way that we operate and the way that we deliver on our missions. So, finding ways to kinda bring those folks into the fold is something that I think we agree on a lot.
And what did we get wrong?
I don’t know, it sounded so good to me across the board. You know, I think one place where I think we wanna fine-tune but I think we’re getting to a good position is, how we can when ramp up on research and develop. I think the administration has made a commitment to prioritizing artificial intelligence in a way that’s never before been done in the history of our country. So, President Trump was the first in history to include artificial intelligence explicitly in an R&D budget and a budget that was sent to Congress and that was done twice. In our priorities memos that we propagated from the White House to the agencies, we called out artificial intelligence and machine-learning for the first time in history in our three memos under this president. So, I think we’re aligned in prioritizing it. I think where I think we should work together on and find a good way forward is, what is the right number if we wanna start cranking up the amount of research and development spending done by the federal government, what is the right number there and how can we do it in a way that the system can absorb those dollars in the most cost-effective way possible? And I think we’re aligned, very much aligned on the high level goal. I think we should work together on finding those specifics and work with Congress and OMB and other stakeholders to get to a place that works for everyone.
So, following up on that point. China has a published strategy to achieve global AI leadership. There are various estimates of how much China’s spending. You all have sponsored, I think a really careful analysis by IDA, sorta casting doubt on some of the ways that those estimates have been made. What do you think though, in terms of the financial commitments that the United States should be making, what’s the rough order of magnitude that you would guess is the right level for us?
Yeah, to me I think it was, we’re very interested in this question, because I think there’s, based on sort of who’s in this room and the cameras back there and everyone else, it’s very clear that journalists care a lot about covering this horse race between us and our adversary on the way we pursue the development of artificial intelligence globally. And I think you know, frankly, I think there’s a lot of misinformation around the way that some other folks around the world are prioritizing artificial intelligence. We know that senior leadership, senior leaders in China have expressed a desire to have China lead the world in AI in a decade. But, the question is, what are they actually doing? And I think the very smart people like your team who are trying to dig through this stuff and through IDA, and I think what needs is careful evaluation. If you rewind another 20 years and look back at sort of the housing boom and the bubble that happened in China, the expectations of what was actually going on and what the press was talking about for many years did not match what the reality was at all on the ground and there was a huge huge disconnect between what was being reported and the reality of the situation. So, we need the smartest minds out there to be able to sort of dig through and do the important research on figuring out where we actually are, so we can think about how it stacks. What I think is very different about our model and the Chinese model, is that the dollars that they throw around are not ones that I think even if they were true, should be compared at all with the way that our federal government spends money. If we look at you know, for us, we have an extraordinarily vibrant private eco-system that spends a large amount of private sector dollars on AI innovation. Those are the types of dollars that don’t, that would be necessarily rolled up in the way that the Chinese think about their numbers. I think we just need to make sure that whenever we do have that conversation, it’s apples to apples. Now, to me I think if we’re thinking about how we can kind of do a progressive roll up, I think thinking in the oars of magnitude of doubling or so, what we’re doing today I think is a goal that we should all strive for and it’s something that I think could make a big difference in the eco-system and one that the eco-system can actually absorb.
So in the last panel, we heard a lot about the kinds of strategic advantages that the United States enjoys by having allies. We have friends and that’s not true with some of our competitors. They don’t have many friends. You have helped to lead the work that OEC did on the AI principles and I think that was one of the first sort of multilateral papers on issuing a set of AI norms. What do you think is the next step for the United States to lead on values internationally?
Yeah, I mean, I think our office is very proud of the work we did at the OCD. And in an environment where our administration is very thoughtful about entering new multilateral agreements and takes those decisions very very seriously, we got to a point we’re able to get that agreement signed, signed in Paris in May. So, it’s something we’re very proud of and I think it speaks to how critically important it is for us to come together with our allies who share our values. To me, I think what’s frightening for many Americans and many people in the West, is the way that the Chinese have twisted technology to really, to use it in a way which flagrantly violates the way we think about the world. So, when artificial intelligence is used to servail people, to imprison minorities, to violate human rights, it’s something we as the West cannot stand for. And I think that is why we, as sort of, Western democracies need to come together and reassert that these are the values that we deeply believe in. And what’s more important and I think what’s really fundamental about this issue, is as I’ve talked earlier, we think a lot about promote and protect and we can go around the world and talk about how important it is to have secure communications networks, how important it is to make sure our research eco-system is protected from people who are attempting to do malicious things. But, at the same time, we need to think about the other side of the coin. We need to come together with out allies and talk about how we can drive and make sure that the next generation technologies are developed in the West so they are underpinned by our western common democratic values and I think that’s what’s absolutely critical. And the OECD’s the first step and it’s something that we need to continue to do with our allies, because they’re as eager as we are to lead the world in technological discovery.
Can you say a little bit about what you saw as the biggest surprises in that process of working with the other OECD members and trying to formulate a set of principles?
In all honesty, I think we were surprised with how close we got the OECD principles to reflect what the president directed in his executive order. I think we went into that process very skeptical about the ability to move the room in a direction that was more pro-innovation and less pro-preemptive regulation. And we had a great team that went over there and I think we were able to have a really good conversation with them about stressing how whatever we do, we need to make sure that we’re not stifling innovation at the end of the day. Because we may have you know, the greatest regulations in the books, but that’s not actually helping drive next generation innovation and it’s certainly not the rules that our adversaries have when they’re making their discoveries. So, for us I think what I was most surprised is how close we got it to what the president directed in his executive order and I think it shows that if we can engage smartly, early and with clear eyes, I think we can actually move the, move our allies and directions closer to the way we think about the world.
So, OSTP is a pretty privileged place to sit because you’re able to cherry pick from some of the best technical talent across the federal government. It’s often been a challenge for parts of the federal government to bring in technical talent as well as talent that might be in adjacent areas relevant to our AI policies going forward. Things like, trade analysis and trade strategy, tech policy and tech law. So, can you describe some of the ways in which you think government is gonna have to meet this challenge?
Yeah, I mean we have struggled and I think it’s something that kinda the entire administration thinks about generally when we bring in talent in the administration. And I think the last administration thought very carefully about this, especially around the way that you can bring tech talent in to help solve sort of citizen facing technology issues. They created the United States Digital Service which we have continued to support and continue to try to give the resources they need to bring talent in. There was a Presidential Innovation Fellows program which was actually signed into law by Obama on the way out and I think that’s another great great option where we can bring in talent for tours of duty. But, each of these programs has its own limitations and I think we need to continue to think about ways to either through the NDAA or other options, think about how we can help clear pathways for talented individuals who want to make a difference for our country, to be able to come here quickly, make a difference and then go back to what they were doing before. This idea that we can get the latest and greatest technologist to make a 20 year commitment to working somewhere in the federal government, I think that’s, we shouldn’t stop trying for that, but I think we should also try to find other lines of effort or pathways to bring people in.
Building off that, there’s this observation attributed to Bill Joy, that most of the smartest people always work for somebody else. And there’s a great addition to that you all made to the National AI R&D strategic plan adding a priority on expanding public-private partnerships. What are some of the approaches that you see being most likely to succeed in building stronger partnerships between government industry, academia, making those relationships more fluid?
Yeah, that could not be more important. It’s something that Andrew and I have talked about at length and I think it goes back to, at its core it goes back to this talent question. I think we all are very familiar with the issue that sort of very high-end AI talent is facing. Do they stay at academic institutions? If they go work for private companies, is the work that they’re doing then tied up in the sort of closed ball garden of wherever they’re working and how do you move people from academia into the private sector and so on, to sort of drive innovation? To us, we think that there’s important partnership models that can happen where, if you’re able to bring entrepreneurs and innovators with academics and private institutions together along with the federal government, you can make a really big difference. And I think Congress is on our side on that. Many of you may have tracked the National Quantum Initiative Act that was signed by the president on December 21st. It’s launched kinda the national effort on quantum leadership, at least on the civilian side. So, it’s three main agencies that are driving it. And as part of that legislation there was a call to create quantum consortia. And again, you had private institutions partnering with academic institutions work with the federal government to create spaces where folks can do this research that can quickly translate into sort of commercialized stuff. And I think a way that we have tried to incorporate that in the way we’re thinking about artificial intelligence, we’ve already seen sort of the first roots of that take hold and that’s an effort by the National Science Foundation to set up artificial intelligence consortia. And those have been great. Again, I will agree that the dollar amounts assigned to them, I think you could add a couple zero’s, but I think we’re at least as a start, being able to kinda build that connective tissue on how you can create good partnerships.
So, I think General Shanahan has pointed out, that in so many of, the really, outside potential applications of AI in DoD, it’s to process automation. It’s to those back office tasks that right now take so much human effort and aren’t really taking advantage of the human skills that some of these exquisitely trained officers have been given. What are some examples of process automation within government that you’re most excited about, whether it’s logistics or contracting or whatever else?
I mean, there are all sorts. I think the contracting stuff I think is very fascinating. One effort that we’re working on is kinda under sort of AI used at looking at regulations. So, this idea that the federal register is absolutely massive and the way that we can kind of look at it and sort of optimize regs in a way or try to find regulations that don’t necessarily kinda meet the world that we live in today, whether it’s you know, regulations talk about using fax machines, whatever it may be. There’s an effort underway there. To me, I think any sort of, sort of supply chain related logistics questions, some of the stuff that the DoD is looking at can make a really big difference. You know, we hosted an event at the White House about a month ago, where we kind of launched our, sort of larger interagency effort on AI in government and I think one of the realizations out of that event was that the bar is really really low and I think it’s something that Jake and folks at DoD have seen as well as you can make a pretty big impact early on because there’s just a lot of low hanging fruit. That being said, I think we need to, as we try to find this, we also need to do a great job of sort of, educating, training up and finding ways that we can work with people who wanna implement AI, but it’s not something that they’ve done before. So, one thing that we’re thinking on the civilian side, is launching a Center of Excellence for artificial intelligence at the, at GSA. And this is a model we’ve used for other sort of IT modernization efforts where we know, for example, that multiple agencies need to move from the cloud, need to move to the cloud for email services, whatever it may be. And 80-90 percent of the work that needs to be done to make that transition is the same, whether you’re at HUD or at SBA or wherever it may be. So, there’s no reason to reinvent the wheel every time. So, creating a hub like that at GSA, they can at least provide an 80 percent solution for a lot of, you know, maybe the five most common processes where we could use AI. Could be a really great way to start baking these things into a number of agencies.
So, Michael, I know you do multiple talks like this a day. This is a rare instance though where you’ve got most of the military and civilian leadership relevant to AI in one room. What are the key messages that haven’t come up either in my questions or in the questions on various other panels during the day, that you want to make sure to get across?
This has probably come up but I just wanna reiterate it. When we as the White House, think about what is the national technology policy of the White House, the number one issue that we are attempting to tackle is to ensure American leadership in artificial intelligence. It’s something we’ve been driving hard on for two years and we have gone on out of way to make sure that our strategy was signed out by the president himself. And if we look back at other technology strategies that were sort of created by administrations in the past, very few have been signed by the president himself. It’s something that we took a lot of painstaking effort to make sure that it got there and maybe it took a little bit longer than we ultimately wanted, but result is, we have something as the backing of the president. And I think we’re extraordinarily excited to drive that agenda and it’s something that we think we can work with folks in the commission to make sure we can do a great job on. And I think what’s fantastic about this effort is we have strong tie ins with Congress. Because, this is not something that DoD can do alone or the White House can do alone, or Congress can do alone. It’s something that we all need to do together and I think I’m extraordinarily excited we have such brilliant minds in the commission to help kinda bring all these pieces together and kinda bring together Congress that can do incredible work on a legislative front with us at the administrative level to kinda bring it all together.
For the last three years, I’ve admired not only your clarity of expression but also your concision. I don’t think there are any wasted words in the executive order. So, in one word, what AI-enabled invention are you most enthusiastic about?
Well, (laughing) it sounds borderline, well, I think healthcare and I think I will give a small sort of personal antidote. My father has Parkinson’s and when I think about the ways that we can sort of use these incredible technologies to help Americans and people all around the world, who are suffering from these debilitating diseases. I think we can make an incredible and huge impact and it’s so inspiring to see our national labs working with amazing researchers and doctors and scientists all across the country to do these, to use our HBC infrastructure, to use AI to drive these next generation medical breakthroughs and anything we can do to make that possible is something that I’m deeply committed to personally and I think that it’s something that can make a huge difference for our country.
Thank you. Please join me in thanking Michael for his leadership. Thank you. (applause)
All right. Can everybody here me? As I said in this moir and in my opening remarks, I said this is gonna be a good day for those of us that are passionate about AI. The fact that 90 percent of the room is still full is a demonstration that the conversation went really good and the feedback from the panelists and from you guys that we’ve got on the report and the way ahead is really enormously helpful as we draft recommendations. The second thing I said, is that no breaks. So, really, kudos to all of you for surviving only 20-minute break during the lunch. So appreciate that. Many thanks to everybody for coming in today, especially my, thanks to our commissioners who have been fabulous in the last eight months. It’s been a joy working with all of you. Thank you to everybody who took a day off to come here and be part of this. We will continue engaging with you, getting your feedback and direct input to our way forward. And lastly, I just wanna thank the staff. I have the best staff in DC and I like to brag about that all the time. So, thank you guys. You are the best, truly. And with that, I’m not gonna hold you from the event that Jason mentioned and I cannot endorse it. Thank you.