WEBVTT - The Story: Are the US and China in a Tech War? w/ Jake Sullivan 0:00:13.119 --> 0:00:16.880 Welcome to Tech Stuff. I'm Os Voloscian. On this podcast, 0:00:16.960 --> 0:00:20.520 we cover emerging technologies and how they reflect but also 0:00:20.680 --> 0:00:26.720 shape the way we live. Artificial intelligence, robotics, energy batteries, 0:00:27.080 --> 0:00:31.640 space exploration. More often than not, though the conversation comes 0:00:31.640 --> 0:00:35.599 back to the same place, China. It's a topic and 0:00:35.640 --> 0:00:38.800 a country that everyone has an opinion about, but where 0:00:38.880 --> 0:00:43.800 few have true insight. Today's guest is different. According to 0:00:43.840 --> 0:00:46.559 a headline last year in Wired magazine, he has quote 0:00:46.880 --> 0:00:50.320 the American who waged a tech war on China. The 0:00:50.400 --> 0:00:54.160 article continues, China is racing to unseat the US as 0:00:54.200 --> 0:00:58.040 the world's technological superpower, not if Jake Sullivan can help it. 0:00:58.600 --> 0:01:00.040 As soon as I read the article, I know, I 0:01:00.120 --> 0:01:02.560 knew I wanted to have Jake Sullivan on the show. 0:01:03.280 --> 0:01:06.320 He served as National Security Advisor for the duration of 0:01:06.400 --> 0:01:10.080 Joe Biden's presidency, and according to my queries on groc 0:01:10.200 --> 0:01:13.679 Gemini and chat GPT, he spent more time with his 0:01:13.760 --> 0:01:18.240 Chinese counterparts, including Chi chimping, than any US national security 0:01:18.280 --> 0:01:22.360 advisor since Henry Kissinger. But today, a conversation about waging 0:01:22.360 --> 0:01:25.840 a tech war on China, Autonomous weapons on the battlefield 0:01:25.880 --> 0:01:29.560 of Ukraine and far beyond, and what comes next for 0:01:29.640 --> 0:01:32.960 a man dubbed a future potential president by Hillary Clinton. 0:01:33.520 --> 0:01:35.000 Jake Sullivan, Welcome to tech Stuff. 0:01:35.360 --> 0:01:36.240 Thanks for having me. 0:01:36.319 --> 0:01:38.640 I wanted to start by asking you this, which is 0:01:39.200 --> 0:01:41.800 you are known as a foreign policy person, right, I mean, 0:01:41.840 --> 0:01:45.560 working in the State Department, working on the Iran nuclear 0:01:45.600 --> 0:01:51.240 deal during the Obama administration, national Security advisor under Joe Biden, 0:01:52.040 --> 0:01:54.560 and yet more and more you seem to be a 0:01:54.600 --> 0:01:58.080 tech policy person. And I'm curious you know how that 0:01:58.120 --> 0:02:02.120 shift began and when you started to see tank as 0:02:02.200 --> 0:02:04.800 such a crucial lens for national security. 0:02:05.280 --> 0:02:09.680 It really began for me after twenty sixteen and I 0:02:09.720 --> 0:02:12.760 served as a senior policy advisor on Hillary Clinton's campaign, 0:02:12.960 --> 0:02:14.800 And it was over the course of that campaign that 0:02:14.880 --> 0:02:19.560 I came to recognize first the degree to which the 0:02:19.600 --> 0:02:23.560 intersection of domestic policy and foreign policy was where everything 0:02:23.680 --> 0:02:29.120 was happening. And second, at that intersection was technology, both 0:02:29.240 --> 0:02:31.639 what we were doing in the United States to invest 0:02:31.840 --> 0:02:35.280 at the cutting edge and what was happening in the 0:02:35.360 --> 0:02:38.440 US China tech competition. So I spent my years out 0:02:38.440 --> 0:02:41.840 of government really digging into a set of technology, economic 0:02:41.960 --> 0:02:44.800 and national security issues and when I came in as 0:02:44.880 --> 0:02:48.280 National Security Advisor for the first time ever the National 0:02:48.360 --> 0:02:52.160 Security Council, we stood up a Directorate on Technology and 0:02:52.240 --> 0:02:56.160 National Security so that we could really create an engine 0:02:56.240 --> 0:02:58.680 room for the kind of policy work I felt needed 0:02:58.720 --> 0:03:01.280 to be done to make sure that the United States 0:03:01.320 --> 0:03:04.040 was in the best possible position to make technology work 0:03:04.080 --> 0:03:05.560 for us rather than against US. 0:03:06.160 --> 0:03:08.760 Yeah, it's interesting when you when you leave off this, 0:03:09.200 --> 0:03:12.200 you know way you go for an exit interview, when 0:03:12.200 --> 0:03:15.440 you sat for the Wide magazine piece why there? And 0:03:15.480 --> 0:03:17.600 did you like the headline the American who waged a 0:03:17.639 --> 0:03:18.720 tech war against China? 0:03:19.120 --> 0:03:21.080 Great question on the headline. I can't say I love 0:03:21.120 --> 0:03:23.560 the headline because I personally you may have noticed, don't 0:03:23.639 --> 0:03:25.040 use the phrase tech war. 0:03:25.240 --> 0:03:26.919 No, it would be it would be ill advised for 0:03:27.520 --> 0:03:29.520 something in your position to use the word war, I 0:03:29.520 --> 0:03:30.840 guess when talking about China. 0:03:31.000 --> 0:03:33.880 Yeah, exactly, And I don't think of it in those terms. Actually, 0:03:33.919 --> 0:03:35.760 I think of it as on a much more common 0:03:35.800 --> 0:03:39.160 sense basis, what do we need to do to promote 0:03:39.360 --> 0:03:41.840 the advancement of American technology and what do we need 0:03:41.880 --> 0:03:46.160 to do to protect American technology from being used against 0:03:46.280 --> 0:03:50.320 us by competitors like the People's Republic of China, and 0:03:50.320 --> 0:03:52.600 that was the strategy that we pursued. It was not 0:03:52.880 --> 0:03:56.680 a war, but it was hard nosed in the sense 0:03:56.760 --> 0:03:58.760 that we felt that for too long we had been 0:03:58.800 --> 0:04:01.800 sending our most advanced tool use technologies to China for 0:04:01.880 --> 0:04:04.200 them to turn back against us, and we weren't going 0:04:04.240 --> 0:04:06.840 to allow that to happen, particularly when it came to 0:04:06.880 --> 0:04:08.920 advanced semiconductors. 0:04:08.480 --> 0:04:12.120 When it comes to China specifically, I mean, what was 0:04:12.120 --> 0:04:16.280 your worst fear about China getting an edge and what 0:04:16.320 --> 0:04:19.159 was your mission to prevent that from happening. 0:04:19.640 --> 0:04:23.279 Well, Look, I've said since the start of my time, 0:04:23.560 --> 0:04:26.160 from when I came in as National Security Advisor, that 0:04:27.360 --> 0:04:31.320 China makes no bones about the fact that it seeks 0:04:31.360 --> 0:04:36.160 to become the world's leading economic, diplomatic, and technological power. 0:04:36.920 --> 0:04:40.880 And my view is that the United States is better off, 0:04:40.920 --> 0:04:43.039 and frankly, the world is better off if the United 0:04:43.080 --> 0:04:47.640 States sustains its position as the world's leading power, including 0:04:47.640 --> 0:04:50.960 the world's leading technology power. That is the way that 0:04:51.040 --> 0:04:53.680 we are most likely to be able to ensure that 0:04:53.760 --> 0:04:57.480 this technological revolution which you have spoken to so many 0:04:57.560 --> 0:05:00.520 different brilliant people about works more for us than it 0:05:00.560 --> 0:05:03.800 works against US, and so I worried that China was 0:05:03.839 --> 0:05:08.520 going to achieve supremacy or dominance in the key technologies 0:05:08.560 --> 0:05:10.640 of the future. And wasn't just my worry. I think 0:05:10.680 --> 0:05:13.159 a lot of people back in twenty twenty one were 0:05:13.240 --> 0:05:15.920 essentially saying, you know, the US is going to lose 0:05:15.920 --> 0:05:18.120 out to China when it comes to AI and other 0:05:18.279 --> 0:05:22.600 areas too. China has too many advantages. The United States 0:05:22.680 --> 0:05:24.600 is too disorganized. It's just not going to get its 0:05:24.600 --> 0:05:27.120 act together. And I was determined to make sure that 0:05:27.160 --> 0:05:29.960 we competed as effectively and vigorously as possible so that 0:05:30.000 --> 0:05:31.760 we could sustain the edge. But I also knew it 0:05:31.800 --> 0:05:34.040 was going to be a tough competition because China has 0:05:34.120 --> 0:05:35.880 a lot of capability to bring to bear in this 0:05:36.000 --> 0:05:37.960 area as well as in other areas, some of which 0:05:37.960 --> 0:05:40.280 you mentioned at the outset of the podcast, whether it's 0:05:40.400 --> 0:05:44.360 robotics or autonomy or you name it. So I can't 0:05:44.360 --> 0:05:46.360 say I went to sleep at night with a single 0:05:46.400 --> 0:05:49.760 fear other than I felt it was my job to 0:05:49.839 --> 0:05:54.160 help position the United States as best as possible in 0:05:54.200 --> 0:05:57.919 this really sustain competition with China over the future of 0:05:58.400 --> 0:06:01.400 key technologies that are going to find so much of 0:06:01.520 --> 0:06:04.840 national security, but also economics and society in the years ahead, 0:06:05.120 --> 0:06:07.480 and AI is essential to that as any. 0:06:08.160 --> 0:06:12.280 There is this theory called the Fucidity's trap, which has 0:06:12.320 --> 0:06:15.200 been popularized by Graham Allison, who I know is a 0:06:15.240 --> 0:06:17.720 colleague at at Harvard now and kind of a mentor, 0:06:17.760 --> 0:06:20.040 which essentially says that you know, if you look at 0:06:20.040 --> 0:06:23.600 the history of a declining great power and a rising 0:06:23.680 --> 0:06:27.040 great power in should we say in our ten cases, 0:06:27.080 --> 0:06:29.360 that will necessarily lead to war? And I think that's 0:06:29.360 --> 0:06:31.880 been kind of like a bipartisan framework for thinking about 0:06:32.120 --> 0:06:35.800 US and China. Is that a framework that you believe 0:06:35.880 --> 0:06:38.679 to be true that there is a about fifty percent 0:06:38.800 --> 0:06:42.320 chance that there's inevitability of a capital w war? Like, 0:06:42.360 --> 0:06:45.120 how do you apply that framework to you're thinking about 0:06:45.200 --> 0:06:47.280 China and then in turn your policy on tech. 0:06:47.680 --> 0:06:51.160 I think Graham Alison's book is excellent. It deserves to 0:06:51.200 --> 0:06:53.760 be read by everybody. I have it here in my 0:06:53.880 --> 0:06:56.400 office at Harvard, and not just because he's my colleague 0:06:56.400 --> 0:07:00.640 down the hall. But I do not believe that it 0:07:00.720 --> 0:07:05.119 is excellent because it is describing an inevitable conflict between 0:07:05.160 --> 0:07:06.760 the US and China. That this is some kind of 0:07:06.839 --> 0:07:08.680 law of nature. In fact, he points out in the 0:07:08.720 --> 0:07:11.680 book that there are cases where a rising power and 0:07:11.800 --> 0:07:16.280 established power didn't go to war. And my view is 0:07:16.320 --> 0:07:19.360 that the US and China are going to be in 0:07:19.400 --> 0:07:24.840 a sustained competition. It will be intense, it will be challenging, 0:07:25.560 --> 0:07:28.560 but it is critical, it is vital that we manage 0:07:28.600 --> 0:07:32.000 that competition so that it does not spill over into 0:07:32.040 --> 0:07:35.400 conflict or confrontation. And my view is we can do that, 0:07:36.000 --> 0:07:38.160 and I actually think the blueprint for that was laid 0:07:38.160 --> 0:07:40.120 out over the four years of the Biden administration. We 0:07:40.120 --> 0:07:43.600 were able to enhance America's competitive position, take tough actions, 0:07:44.120 --> 0:07:47.200 and also engage in sustained diplomacy so that we didn't 0:07:47.320 --> 0:07:50.160 move in the direction of war with China, which would 0:07:50.200 --> 0:07:52.800 be a catastrophe not just for the US and China, 0:07:52.840 --> 0:07:53.840 but for the entire world. 0:07:54.400 --> 0:07:56.440 He let's talk about some of the tough measures. You 0:07:56.480 --> 0:08:00.960 mentioned two countries, Holland and Japan. That this kind of 0:08:01.000 --> 0:08:06.120 fascinating secret meeting that you convene with your adoption Japanese counterparts. 0:08:07.200 --> 0:08:08.200 What are you trying to achieve? 0:08:08.840 --> 0:08:11.640 So I'll be advised in what I say, because having 0:08:11.680 --> 0:08:15.040 some discretion in the diplomatic relationships with Japan and the 0:08:15.080 --> 0:08:18.600 Netherlands was critical to being able to achieve the kind 0:08:18.640 --> 0:08:23.320 of consensus we were able to achieve on semiconductor manufacturing 0:08:23.320 --> 0:08:27.200 equipment export controls. So, for your listeners, there's two types 0:08:27.240 --> 0:08:30.600 of export controls. There's the controls on very high end ships, 0:08:31.080 --> 0:08:35.240 I think H one hundred Nvidia chips or the Blackwells, 0:08:35.840 --> 0:08:41.160 and then there are controls on the manufacturing equipment used 0:08:41.600 --> 0:08:45.880 to make the highest end AI chips and other high 0:08:45.960 --> 0:08:51.320 end ships. And that equipment is primarily produced by three countries, 0:08:51.520 --> 0:08:55.840 the United States, the Netherlands, and Japan, and so having 0:08:55.880 --> 0:08:59.040 all three of us come together around a common understanding 0:08:59.520 --> 0:09:04.040 of what components and what machines necessary for the manufacture 0:09:04.040 --> 0:09:08.320 of semiconductors would be controlled was essential for such controls 0:09:08.360 --> 0:09:14.720 to be effective, and that required persuasion, consultation, really laying 0:09:14.760 --> 0:09:17.520 out the challenge as we saw it, and it also 0:09:17.600 --> 0:09:21.440 required a deep level of technical detail, which is why 0:09:21.800 --> 0:09:24.240 this really was not just about the diplomats or the 0:09:24.320 --> 0:09:27.079 national security advisors being in the room. It was about 0:09:27.120 --> 0:09:30.200 technical experts from each of the three countries really working 0:09:30.240 --> 0:09:32.520 through the reality of this so that it wasn't just 0:09:32.520 --> 0:09:33.640 a fly by night effort. 0:09:34.200 --> 0:09:38.640 One policy goal was to prevent China from buying advanced 0:09:38.679 --> 0:09:42.120 chips from companies like Nvidia, and the other was to 0:09:42.120 --> 0:09:46.079 prevent China from developing the technology to produce them domestically, 0:09:46.120 --> 0:09:49.840 which was more what the Japanese Dutch diplomatic effort was. 0:09:50.000 --> 0:09:56.679 Around January twentieth, literally the day you leave office, Deep Seek, 0:09:56.760 --> 0:10:01.920 the Chinese AI Company announces a new you reasoning model 0:10:02.280 --> 0:10:06.320 that performs on par with the best American models and 0:10:06.440 --> 0:10:09.280 was able to be developed in spite of the export controls. Now, 0:10:09.320 --> 0:10:11.400 some people say they claim it's because they actually were 0:10:11.400 --> 0:10:14.199 able to develop such as sophisticated technology, they didn't need 0:10:14.200 --> 0:10:17.280 as many chips. Others say they just got around the 0:10:17.280 --> 0:10:19.640 export controls. I don't know if you know which of 0:10:19.640 --> 0:10:21.680 those you have more of a hypothesis to be true. 0:10:22.360 --> 0:10:24.600 But either way, what was your reaction to this moment? 0:10:24.640 --> 0:10:26.560 I mean, someone said this was basically the moment that 0:10:26.920 --> 0:10:28.800 everything you were working on was designed to prevent. 0:10:29.840 --> 0:10:31.960 Well, keep in mind, this is a I'm glad you 0:10:32.040 --> 0:10:33.640 used the word moment because it was a moment. It 0:10:33.640 --> 0:10:35.920 was a moment in time in an ongoing competition, and 0:10:35.960 --> 0:10:39.000 of course since then, open Ai has come out with three. 0:10:39.360 --> 0:10:42.040 Deep Seek has come out with yet another version of 0:10:42.080 --> 0:10:44.959 its R one model, and we'll see more models coming 0:10:44.960 --> 0:10:49.360 out as we go. For me, deep Seek sort of 0:10:49.360 --> 0:10:52.440 showed me number one the power of a concerted pr 0:10:52.520 --> 0:10:56.880 campaign by China to effectively say resistance is futile. These 0:10:56.920 --> 0:11:00.360 export controls have failed. Give up America. I would just 0:11:00.440 --> 0:11:02.920 point out that it was the number one issue they 0:11:03.040 --> 0:11:05.200 raised with me in every meeting, which suggests to me 0:11:05.320 --> 0:11:07.920 they didn't think the export controls were totally useless. 0:11:08.040 --> 0:11:10.040 The number one issue is why you try to prevent 0:11:10.120 --> 0:11:11.200 us from buying chips? 0:11:11.480 --> 0:11:16.080 Right? Why have you imposed these semiconductor export controls on us? 0:11:16.600 --> 0:11:19.199 And you know, I spent a lot of time explaining 0:11:19.280 --> 0:11:21.760 our theory of small yard high fence, and what we 0:11:21.800 --> 0:11:26.719 saw is the difference between a national security logic to 0:11:27.040 --> 0:11:30.040 export controls in some kind of economic or technology blockade, 0:11:30.040 --> 0:11:31.840 which I do not believe that we are engaged in 0:11:31.880 --> 0:11:32.320 with China. 0:11:32.640 --> 0:11:35.400 They've responded just to your small yard high fence metaphor 0:11:35.440 --> 0:11:37.679 with big yard iron cut in right correct. 0:11:37.960 --> 0:11:41.640 That that is what they said, And I said, well, 0:11:41.679 --> 0:11:44.280 we see it very differently. I see it as a 0:11:44.320 --> 0:11:46.360 small yard and a high fence. And part of the 0:11:46.360 --> 0:11:48.439 reason I know it's small is I know in the 0:11:48.480 --> 0:11:50.600 meetings that I'm conducting how much we're keeping out of it, 0:11:50.679 --> 0:11:53.960 how much continuing technology trade there is between the US 0:11:53.960 --> 0:11:56.440 and China. So but the other thing about deep Seat 0:11:56.440 --> 0:11:59.040 that I think is really important for people to recognize 0:11:59.240 --> 0:12:04.679 is the big breakthrough was really an open AI breakthrough 0:12:04.679 --> 0:12:07.400 that then Deep Seek more or less replicated and then 0:12:07.440 --> 0:12:11.560 improved upon. And I don't actually know exactly I'm not 0:12:11.640 --> 0:12:14.040 inside Deep Seek, so I can't say, but it seems 0:12:14.040 --> 0:12:17.680 on the balance of the evidence that the deep Seek 0:12:17.760 --> 0:12:20.400 story is the story that involves them having access to 0:12:20.440 --> 0:12:24.160 a range of Western chips, some of which they got 0:12:24.600 --> 0:12:27.480 before the export controls ever went into place, and some 0:12:27.679 --> 0:12:30.960 they got between the first set in twenty two and 0:12:31.000 --> 0:12:33.840 the second set in twenty three when in video basically 0:12:33.920 --> 0:12:37.840 designed around the controls what they call the H eight hundred, 0:12:38.480 --> 0:12:41.800 So there was an H one hundred, which was at 0:12:41.840 --> 0:12:46.600 one point the most elite chip. In Vidia basically reduced 0:12:46.600 --> 0:12:49.320 the interconnect speed to get around the control created the 0:12:49.480 --> 0:12:52.400 H eight hundred for the Chinese market that got fixed 0:12:52.400 --> 0:12:55.160 in twenty three. And so I think the right way 0:12:55.200 --> 0:12:58.720 to think about the export control story is it started 0:12:58.720 --> 0:13:02.840 in twenty two. The first draft of those export controls 0:13:02.840 --> 0:13:05.400 in twenty two we learned lessons from because this was 0:13:05.440 --> 0:13:08.800 a new undertaking, and one of those lessons involved how 0:13:08.800 --> 0:13:12.280 to make sure that there couldn't be this easy design around, 0:13:12.280 --> 0:13:14.360 this easy workaround, and that's why we updated them in 0:13:14.400 --> 0:13:17.480 twenty three and updated them again in twenty four. So 0:13:18.280 --> 0:13:22.439 these are not going to be one hundred percent foolproof. 0:13:23.240 --> 0:13:26.080 What they are designed to do is, to the maximum 0:13:26.080 --> 0:13:30.760 extent possible, make sure that the highest end national security 0:13:30.840 --> 0:13:34.640 related technology, these very high end AI chips are not 0:13:35.120 --> 0:13:37.680 going to China to be used against America currents allies. 0:13:38.040 --> 0:13:41.280 And you know, I can't guarantee zero leakage, that it's 0:13:41.280 --> 0:13:44.640 not a dynamic scenario where China and others are working 0:13:44.720 --> 0:13:46.679 to get around them. We just have to keep at 0:13:46.720 --> 0:13:49.160 it and continue to learn and continue to update. 0:13:49.679 --> 0:13:52.400 So just to play a bank to you, you'll view 0:13:52.400 --> 0:13:59.119 as far from controls on exports of advanced chips being futile. 0:13:59.200 --> 0:14:02.400 And even if deep Seat were able to distill knowledge 0:14:02.520 --> 0:14:07.280 from open AI's models, there is a value in these 0:14:07.320 --> 0:14:11.200 export controls. And maybe deep Seak's success was because the 0:14:11.240 --> 0:14:15.480 export controls weren't effective until let's say twenty three, and 0:14:15.600 --> 0:14:18.640 really your legacy will be making sure the next Deep 0:14:18.679 --> 0:14:19.680 Seat moment doesn't happen. 0:14:20.160 --> 0:14:23.560 Yeah, I mean, I think realistically time will tell. What 0:14:23.600 --> 0:14:26.600 I would just argue is I believe it's already had 0:14:26.640 --> 0:14:30.920 an impact, because I think it has placed quite a 0:14:30.960 --> 0:14:35.520 limit on China's access to an incredibly important resource for 0:14:35.600 --> 0:14:39.960 frontier AI development, Compute, and that we can expect that 0:14:40.400 --> 0:14:44.880 impact to be seen over the course of the rest 0:14:44.880 --> 0:14:47.640 of twenty five, twenty six, twenty seven. You know, if 0:14:47.680 --> 0:14:51.400 you look at public statements from Deep Seek CEO, one 0:14:51.440 --> 0:14:54.120 of the things he talks about is access to Compute 0:14:54.200 --> 0:14:57.480 being a challenge for him, something the Chinese government doesn't 0:14:57.520 --> 0:15:01.720 like to acknowledge, but it something he's sort of very 0:15:01.920 --> 0:15:04.840 publicly talked about. The China's effort to have its own 0:15:04.920 --> 0:15:08.280 ship has not met with the kind of success that 0:15:08.360 --> 0:15:11.480 the Chinese propagandas want to suggest it has, either in 0:15:11.560 --> 0:15:13.960 terms of their performance or in terms of the scale 0:15:14.000 --> 0:15:17.000 of what they're capable of producing. They cannot produce in 0:15:17.040 --> 0:15:20.800 a year anything remotely resembling what, for example, in video 0:15:20.880 --> 0:15:23.840 can produce. Not to mention the Google TPUs or Trainium 0:15:24.880 --> 0:15:29.720 or other American design silicon. So I think those core 0:15:30.320 --> 0:15:34.080 observations are intact sitting here today in June of twenty 0:15:34.120 --> 0:15:36.760 twenty five, and now let's see how things play out. 0:15:36.960 --> 0:15:39.520 I want to talk about video for a moment. I mean, obviously, 0:15:39.560 --> 0:15:41.600 Jensen Whang, the SEO of and Video, has been out 0:15:41.640 --> 0:15:47.040 there loudly and publicly castigating in effect your policy and 0:15:47.080 --> 0:15:50.320 also saying, you know, necessity is the mother of invention 0:15:50.440 --> 0:15:53.320 and in fact us export controls. And we've heard this 0:15:53.440 --> 0:16:00.520 argument before, spurred this great development in Chinese domestic developments 0:16:00.520 --> 0:16:02.680 and technology. How do yours boned? 0:16:03.040 --> 0:16:06.320 Well? First, it's an interesting statement is the suggestion that 0:16:06.360 --> 0:16:09.400 if deep Seak had more Nvidia chips, it would not 0:16:09.680 --> 0:16:13.840 have invented R one. I mean, that is a to me, 0:16:13.880 --> 0:16:18.200 a kind of bizarre statement. But the argument that I 0:16:18.320 --> 0:16:24.160 frequently encounter in this area is, aha, Look, China's motivated 0:16:24.280 --> 0:16:26.560 when it comes to AI and motivated when it comes 0:16:26.560 --> 0:16:30.120 to semiconductor production. Because you are placing limits on the 0:16:30.200 --> 0:16:33.720 chips and the manufacturing equipment they can access, You've motivated them. 0:16:33.880 --> 0:16:36.760 And this is really Jensen's argument, and I think that 0:16:36.840 --> 0:16:41.400 this misses just the facts and the sequence of what's 0:16:41.480 --> 0:16:44.320 unfolded over the last decade. It was China who came 0:16:44.360 --> 0:16:46.920 out years before these export controls were ever in place 0:16:47.360 --> 0:16:49.760 and announced to the world that they were going to 0:16:49.800 --> 0:16:52.920 be the world leader in ai by twenty thirty. And 0:16:52.960 --> 0:16:55.480 that one of the areas that made in China twenty 0:16:55.520 --> 0:16:57.480 twenty five, by the way, a policy that was put 0:16:57.520 --> 0:16:59.640 in place more than a decade ago was going to 0:16:59.680 --> 0:17:05.040 focus on, was chips. And so the semiconductor export controls 0:17:05.240 --> 0:17:09.000 did not create this big push by China. The semiconductor 0:17:09.040 --> 0:17:12.760 export controls responded to this big push by China. And 0:17:12.840 --> 0:17:15.520 so I think that this narrative of necessity as the 0:17:15.520 --> 0:17:18.760 mother of invention kind of gets the sequence exactly backwards. 0:17:19.119 --> 0:17:22.399 It suggests that the big Chinese push all happened post 0:17:22.440 --> 0:17:26.880 export controls, when in fact it long predated the export controls, 0:17:26.880 --> 0:17:29.320 and the export controls were in part motivated by saying, well, 0:17:29.320 --> 0:17:31.359 wait a second, China's kind of come out to the 0:17:31.359 --> 0:17:32.760 world and said this is what we want to do. 0:17:32.880 --> 0:17:35.480 We're telling all of you as saying, well, we're not 0:17:35.520 --> 0:17:38.320 going to make it easier for you, because you know, 0:17:38.440 --> 0:17:40.679 we want to sustain our advantages. 0:17:40.240 --> 0:17:42.600 Just to play devil's advocate, though. I mean, obviously, any 0:17:42.920 --> 0:17:45.720 economy has a demand side and a supply side, right, 0:17:45.800 --> 0:17:49.399 and clearly you know you're right. The demand side from 0:17:49.440 --> 0:17:52.040 the Chinese Communist Party and the government was there get 0:17:52.080 --> 0:17:56.040 to ANI supremacy by twenty thirty on shore chip manufacturing. 0:17:56.800 --> 0:18:00.240 But there's a lot of demand site mandates from the 0:18:00.280 --> 0:18:05.159 Chinese Communist Party which don't translate into enormous successes. The 0:18:05.240 --> 0:18:09.159 supply side issue, which was US policy is saying well, 0:18:09.160 --> 0:18:11.959 you can't have this from US, may have ultimately been 0:18:11.960 --> 0:18:15.680 a much more powerful and motivating signal to Chinese technology 0:18:15.720 --> 0:18:18.040 companies than the dictates of their own government. 0:18:18.720 --> 0:18:21.000 Well, I would make a couple points in response to that. 0:18:21.080 --> 0:18:23.720 The first is that in a bunch of the other 0:18:23.880 --> 0:18:27.400 key areas that China also identified mid and China twenty 0:18:27.480 --> 0:18:31.880 twenty five, you've seen huge progress in China, huge progress, right. 0:18:32.000 --> 0:18:35.200 So somehow the demand and supply sides lined up when 0:18:35.240 --> 0:18:39.560 it came to things like robotics or electric vehicles, or 0:18:39.840 --> 0:18:42.040 you go down the list of what they identified as 0:18:42.040 --> 0:18:44.440 their sectors. It's not like AI stands out as something 0:18:44.440 --> 0:18:46.800 that all of a sudden blipped off the map because 0:18:46.800 --> 0:18:49.359 of our export control. So I find it a little 0:18:49.359 --> 0:18:55.200 hard to square with what has been a very distinctive 0:18:56.200 --> 0:19:00.879 presidentially dictated policy, of which there are are not many 0:19:01.160 --> 0:19:03.520 that look like made in China twenty twenty five, or 0:19:03.560 --> 0:19:06.359 where they have put as much energy as they have 0:19:06.480 --> 0:19:07.199 in this area. 0:19:07.760 --> 0:19:10.000 I know that your Chinese counterpart often like to ask 0:19:10.040 --> 0:19:13.200 you this kind of rhetorical question, where's the line between 0:19:13.200 --> 0:19:18.359 economic policy and national security? Physical question? And I'm curious, 0:19:18.720 --> 0:19:21.520 you know, did you and President Biden look with envy 0:19:22.160 --> 0:19:25.920 at President She's ability to onboard massive new energy onto 0:19:25.960 --> 0:19:30.760 the grid to dictate the decision making at the top 0:19:30.840 --> 0:19:32.959 Chinese tech firms in a way that the US government 0:19:33.000 --> 0:19:36.200 and the tech firms are increasingly or often intention I mean, 0:19:36.880 --> 0:19:39.720 was there something about the Chinese model in terms of 0:19:39.800 --> 0:19:44.480 preparing for this new industrial and AI revolution that you 0:19:44.480 --> 0:19:45.439 wish you could have borrowed? 0:19:45.840 --> 0:19:48.800 Well, look, anytime you're working at the White House and 0:19:48.840 --> 0:19:55.280 you're dealing with bureaucracy, red tape regulation, the thought is 0:19:55.280 --> 0:19:57.359 never far from your mind. Hey, if I was just 0:19:57.520 --> 0:20:00.800 all powerful and President Biden was all powerful, we could 0:20:00.840 --> 0:20:03.280 just not have to waste time with any of this. 0:20:03.680 --> 0:20:06.080 But then you pause and you say, hey, wait a second, Actually, 0:20:06.080 --> 0:20:09.040 democracy is a pretty good form of government one. And 0:20:09.040 --> 0:20:14.000 two the American technology model, which is messier, which is 0:20:14.040 --> 0:20:21.119 more decentralized, which is in many ways more maddening, also 0:20:21.200 --> 0:20:23.240 really works. And it's a good thing to bet on. 0:20:24.000 --> 0:20:26.560 And that's why I think we're in such a disturbing 0:20:26.640 --> 0:20:29.320 moment right now, because, look, some of the advantages of 0:20:29.320 --> 0:20:31.879 the US has. One is the ability to attract talent 0:20:32.240 --> 0:20:35.200 from all over the world. Another is this ecosystem a 0:20:35.240 --> 0:20:39.960 basic research funding supplied by the government, research universities, and 0:20:40.040 --> 0:20:42.840 the private sector. And two of those three pillars are 0:20:42.880 --> 0:20:46.320 being knocked out by this administration. It strikes me that 0:20:47.080 --> 0:20:50.680 the most recent announcement saying basically, we're going to try 0:20:50.680 --> 0:20:53.680 to really dramatically reduce the number of Chinese grad students 0:20:53.680 --> 0:20:58.200 and undergraduate students and researchers in the United States, all 0:20:58.200 --> 0:21:00.160 of these are self harming moves because all of them 0:21:00.640 --> 0:21:04.800 take away from this unique model. The US is built 0:21:04.840 --> 0:21:08.840 to build and sustain an innovation edge over time. So 0:21:08.880 --> 0:21:11.240 the Chinese have their way of doing it. We have 0:21:11.320 --> 0:21:14.280 to look at that and say, that's a formidable competitor. 0:21:14.800 --> 0:21:18.119 But it is precisely that observation that we are dealing 0:21:18.119 --> 0:21:21.399 with the formidable competitor. That motivated us to say, what 0:21:21.440 --> 0:21:22.919 do we need to do on the promote side, what 0:21:22.920 --> 0:21:25.840 do we need to do to push the boundaries, and 0:21:25.880 --> 0:21:27.400 then what do we need to do on the protect 0:21:27.440 --> 0:21:30.320 side to make sure that our most advanced technologies aren't 0:21:30.320 --> 0:21:33.560 being used against us? So that's it was that sense 0:21:33.640 --> 0:21:38.399 of real understanding that this was going to be a 0:21:38.440 --> 0:21:41.679 hard race in many of these different technology areas that 0:21:41.800 --> 0:21:44.440 motivated us to take the policy steps that we took. 0:21:46.560 --> 0:21:50.240 After the break, Jake Sullivan on how the Biden administration 0:21:50.840 --> 0:21:56.200 was preparing for the possibility of agi I, a system 0:21:56.480 --> 0:22:00.760 that is superior to humans and effectively old COCA tasks, 0:22:01.400 --> 0:22:04.400 and also Jake talks about the role of autonomous weapons 0:22:04.400 --> 0:22:12.520 in current conflicts from Ukraine to Gaza. I want to 0:22:12.560 --> 0:22:16.719 zoom out a bit, Jake. One of the most interesting 0:22:16.920 --> 0:22:21.280 conversations I've seen this year was between as reclined ere 0:22:21.320 --> 0:22:25.119 I say it and Ben Buchanan, the White House Special 0:22:25.119 --> 0:22:29.119 Advisor on a AI for the Biden administration. Essentially that 0:22:29.160 --> 0:22:32.919 they had a conversation where Ben Buchanan said it was 0:22:32.960 --> 0:22:38.200 the belief of the Biden administration that artificial general intelligence 0:22:38.400 --> 0:22:41.119 is going to come within the next four years, i e. 0:22:41.640 --> 0:22:44.840