WEBVTT 00:00:00.000 --> 00:00:02.357 To discuss the technological trajectory 00:00:02.440 --> 00:00:05.317 and strategic implications of AI, the 00:00:05.400 --> 00:00:07.837 future of America's innovation power, 00:00:07.920 --> 00:00:10.357 and changes in warfare. Please welcome 00:00:10.440 --> 00:00:14.317 Dr. Eric Schmidt, chair of SCSP, and CEO 00:00:14.400 --> 00:00:16.957 of Relativity Space, in conversation 00:00:17.040 --> 00:00:19.077 with Tom Shanker, director for the 00:00:19.160 --> 00:00:22.997 project of media and national security. 00:00:23.080 --> 00:00:26.557 Uh thank you so much and good afternoon everyone. 00:00:26.640 --> 00:00:31.317 Um It's my great honor to moderate this discussion with Dr. Eric Schmidt. There 00:00:31.400 --> 00:00:33.117 are few people who've shaped the 00:00:33.200 --> 00:00:35.437 technology landscape of the last half 00:00:35.520 --> 00:00:38.357 century century and are still shaping 00:00:38.440 --> 00:00:40.517 for the future as our speaker Dr. 00:00:40.600 --> 00:00:45.037 Schmidt. He led Google from a promising startup to one of the most consequential 00:00:45.120 --> 00:00:49.237 companies in history. He then chaired the Defense Innovation Board and the 00:00:49.320 --> 00:00:53.797 National Commission on Artificial Intelligence. In 2021, 00:00:53.880 --> 00:00:56.157 he co-founded and endowed the Special 00:00:56.240 --> 00:01:00.917 Competitive Studies Project, our host today for this important panel. 00:01:01.000 --> 00:01:05.157 Uh in his spare time, he's written extensively a number of books 00:01:05.240 --> 00:01:07.557 uh on all sorts of things, most recently 00:01:07.640 --> 00:01:11.277 Agenda AI. And in March 2025, he became 00:01:11.360 --> 00:01:14.517 CEO of Relativity Space, bringing the 00:01:14.600 --> 00:01:16.917 same energy to the race for space that 00:01:17.000 --> 00:01:18.917 he brought to the race for the internet 00:01:19.000 --> 00:01:22.077 and AI. And guys my age only have one um 00:01:22.160 --> 00:01:24.317 audio visual aid. I wore this of the 00:01:24.400 --> 00:01:29.437 cosmos, Dr. Schmidt, just to honor your work in space. So, thank you. 00:01:29.520 --> 00:01:33.237 Um thanks for all you've done here for your time today. There's so much to talk 00:01:33.320 --> 00:01:37.677 about, but I wanted to start with the headlines, which is the drone war, 00:01:37.760 --> 00:01:41.757 Strait of Hormuz, Ukraine. And I know you're just back 00:01:41.840 --> 00:01:44.717 from Ukraine making another of your valuable trips. 00:01:44.800 --> 00:01:46.717 On the topic of AI here today, Dr. 00:01:46.800 --> 00:01:49.397 Schmidt, at the tactical level, what do 00:01:49.480 --> 00:01:51.677 you see as the future of the marriage of 00:01:51.760 --> 00:01:54.957 AI and drones? And then strategically, 00:01:55.040 --> 00:01:57.877 how is AI going to reshape the face of 00:01:57.960 --> 00:02:02.077 combat? Um so first thing, Tom, thank you for it for doing this. I want to 00:02:02.160 --> 00:02:06.637 thank you all for being here. This was an experiment that Illy came up with, 00:02:06.720 --> 00:02:09.637 and I said, "Well, you know, it'll probably work." And it's worked 00:02:09.720 --> 00:02:13.277 enormously well, and I want to thank Illy and the entire team putting this 00:02:13.360 --> 00:02:17.997 event on. Uh we're committed to national security. 00:02:18.080 --> 00:02:21.037 We're committed to working with the government. We're committed to working 00:02:21.120 --> 00:02:23.197 with you who work with the government. 00:02:23.280 --> 00:02:26.197 Uh we we want to make America safe and 00:02:26.280 --> 00:02:29.197 powerful and keep it that way for many, 00:02:29.280 --> 00:02:33.397 many decades. To answer your question, uh I spent lots of time in Ukraine. I 00:02:33.480 --> 00:02:35.757 have a company there. Um and I spent 00:02:35.840 --> 00:02:37.637 lots of time on the front lines, uh 00:02:37.720 --> 00:02:40.757 which is not for the faint of heart. 00:02:40.840 --> 00:02:44.117 And I think now we're seeing there So, 00:02:44.200 --> 00:02:48.197 it it's important to understand that Ukraine is the laboratory for future 00:02:48.280 --> 00:02:50.877 war. Uh I'm not in any way endorsing 00:02:50.960 --> 00:02:53.077 this war. It's a horrible war. I've seen 00:02:53.160 --> 00:02:56.237 plenty people die. It's just horrible. 00:02:56.320 --> 00:02:59.877 Horrible, horrible, horrible. Uh having said that, it's important to 00:02:59.960 --> 00:03:03.717 understand what's going on. Uh today, the flood, as it's known, is 00:03:03.800 --> 00:03:06.357 quite wide. That's the actual line 00:03:06.440 --> 00:03:09.557 demarcating the two sides. Um and the 00:03:09.640 --> 00:03:12.357 people who are on that area are in 00:03:12.440 --> 00:03:15.477 foxholes. The minimum number of days 00:03:15.560 --> 00:03:19.637 they spend in the foxholes is 50, 5 0. 00:03:19.720 --> 00:03:22.437 But in many cases, it's up to 3 to 4 00:03:22.520 --> 00:03:25.557 months. How do they survive? Drones 00:03:25.640 --> 00:03:29.917 supply them with water and food. If they stick their head out of the ground, 00:03:30.000 --> 00:03:33.677 they're killed. That's what a transparent battlefield 00:03:33.760 --> 00:03:37.877 looks like. And I don't think we Americans understand this. I think 00:03:37.960 --> 00:03:39.637 almost no country understands that 00:03:39.720 --> 00:03:41.637 that's what the future of drone war 00:03:41.720 --> 00:03:44.997 means. Um the result of this is that it 00:03:45.080 --> 00:03:47.917 is the the front is largely frozen. It's 00:03:48.000 --> 00:03:50.557 very hard for either side to go against 00:03:50.640 --> 00:03:55.357 the other. And people take enormous risks running along the fields, hiding 00:03:55.440 --> 00:03:59.717 under trees, and so forth, while the drone hunters are hunting them. And it's 00:03:59.800 --> 00:04:03.357 just horrible to watch. Uh in Ukraine, they've recently 00:04:03.440 --> 00:04:07.237 announced that they've hit their fourth month of killing their quota of Russian 00:04:07.320 --> 00:04:11.677 targets, which they say is 35,000. To give you an an 00:04:11.760 --> 00:04:15.957 an idea of the scale of this war, it's hard for us to understand. Think about 00:04:16.040 --> 00:04:19.597 the number of people who we lost in Vietnam in the context of of what's 00:04:19.680 --> 00:04:21.677 going on in Ukraine. The Ukraine numbers 00:04:21.760 --> 00:04:23.597 are not released, uh but it's 00:04:23.680 --> 00:04:27.277 substantially lower than 35,000. 00:04:27.360 --> 00:04:30.157 And the consequence of this this this 00:04:30.240 --> 00:04:32.637 thing is anything that moves is 00:04:32.720 --> 00:04:35.517 destroyed. So, get rid of your tanks, 00:04:35.600 --> 00:04:37.637 armored personnel carriers, 00:04:37.720 --> 00:04:40.237 uh artillery, flamethrowers, all the fun 00:04:40.320 --> 00:04:42.717 stuff that you see in movies, right? 00:04:42.800 --> 00:04:45.677 It's sort of over. Um replaced by these 00:04:45.760 --> 00:04:48.357 powerful robotic systems. 00:04:48.440 --> 00:04:50.517 Technically, the way they work is they 00:04:50.600 --> 00:04:55.477 uh can work without GPS, they can communicate through electronic warfare. 00:04:55.560 --> 00:04:59.797 That technology is getting stronger on both sides. So, this is the future. 00:04:59.880 --> 00:05:02.157 The battle now has moved to the 00:05:02.240 --> 00:05:05.357 mid-strike area, which is roughly 40 to 00:05:05.440 --> 00:05:08.637 100 km. And that's where the supply line 00:05:08.720 --> 00:05:12.437 on the other on the enemy side is, supplying these people who are 00:05:12.520 --> 00:05:16.157 unfortunately in the foxholes. And that's where the bombing is, and that's 00:05:16.240 --> 00:05:17.517 where the war is largely being fought 00:05:17.600 --> 00:05:21.477 right now. Um and Ukraine is slowly uh 00:05:21.560 --> 00:05:25.437 is as pretty much stopped Russia's advance, which is extraordinary, and is 00:05:25.520 --> 00:05:26.997 slowly beginning to respond 00:05:27.080 --> 00:05:30.757 successfully. You asked about AI. 00:05:30.840 --> 00:05:33.997 Um if you take the the principle that I 00:05:34.080 --> 00:05:37.677 described, which is that basically uh 00:05:37.760 --> 00:05:40.477 air, land, and sea drones are going to 00:05:40.560 --> 00:05:43.357 be the way that this that is 00:05:43.440 --> 00:05:46.557 um code for robots and automation and 00:05:46.640 --> 00:05:50.117 AI. Um they will all be interconnected. 00:05:50.200 --> 00:05:53.677 They will all operate in the equivalent of swarms and they'll all operate in 00:05:53.760 --> 00:05:57.917 what are called coordinated attacks. So, a typical example is you'll have a 00:05:58.000 --> 00:06:00.317 target and you'll surround the target by 00:06:00.400 --> 00:06:02.317 a number of drones and you hit it 00:06:02.400 --> 00:06:05.397 simultaneously and most targets do not 00:06:05.480 --> 00:06:07.677 have the ability to simultaneously in a 00:06:07.760 --> 00:06:10.877 360 azimuth defend themselves. So, 00:06:10.960 --> 00:06:14.757 that's roughly where we are. Thanks. That's on the battlefield there. 00:06:14.840 --> 00:06:18.597 If you think about that as a Russia-Ukraine competition, add in Dr. 00:06:18.680 --> 00:06:21.677 Schmidt, the US and China. How will this 00:06:21.760 --> 00:06:27.077 influence how we should both prepare and deter war and fight if required? And I 00:06:27.160 --> 00:06:31.677 have to ask the former journalist question, who's ahead? 00:06:31.760 --> 00:06:35.677 So, it's a a slightly different subject 00:06:35.760 --> 00:06:40.917 would be on AI. They're all interrelated, of course. 00:06:41.000 --> 00:06:45.437 A week ago, China released Deep Seek V4. 00:06:45.520 --> 00:06:47.437 And when I was working on the National 00:06:47.520 --> 00:06:49.757 Security Commission for AI, we worked 00:06:49.840 --> 00:06:53.637 really hard with Trump 1 and Biden to 00:06:53.720 --> 00:06:57.037 implement hardware controls for chips. 00:06:57.120 --> 00:06:58.957 We were very successful collectively, 00:06:59.040 --> 00:07:01.917 all of us, to limit hardware access to 00:07:02.000 --> 00:07:04.077 China. And I was strongly in favor of 00:07:04.160 --> 00:07:06.757 this policy as were many people. 00:07:06.840 --> 00:07:09.437 It looks like the policy has worked, but 00:07:09.520 --> 00:07:12.997 it's largely beginning to not work. 00:07:13.080 --> 00:07:15.717 The Chinese are clever, they're smart. 00:07:15.800 --> 00:07:18.477 Um and they have now built systems that 00:07:18.560 --> 00:07:23.517 are within striking range of the top models in America, 00:07:23.600 --> 00:07:27.837 but with much, much less powerful hardware. The The technical summary is 00:07:27.920 --> 00:07:32.757 that they use a chip called the Ascend chip. It uses a slower process and 00:07:32.840 --> 00:07:36.477 they've invented new software ways to get around the various latency and 00:07:36.560 --> 00:07:40.837 architectural problems. Um what I like about this is that we 00:07:40.920 --> 00:07:42.837 have real competitors. What I don't like 00:07:42.920 --> 00:07:46.557 about it is China is very focused in 00:07:46.640 --> 00:07:48.237 broad diffusion of this technology 00:07:48.320 --> 00:07:51.077 globally and it's all open source, which 00:07:51.160 --> 00:07:53.757 means it's largely uncontrolled and not 00:07:53.840 --> 00:07:58.397 controlled in any way by us. Um so I think a fair statement is a year 00:07:58.480 --> 00:08:01.877 ago I said I thought that they were 1 to 2 years 00:08:01.960 --> 00:08:06.117 behind. It looks like they've caught up enough that the most recent analysis is 00:08:06.200 --> 00:08:08.717 China's within 6 months, 00:08:08.800 --> 00:08:11.037 which is a nano second, right, in our 00:08:11.120 --> 00:08:13.357 world. So it gives you a sense of how 00:08:13.440 --> 00:08:16.037 committed China is to the AI leadership 00:08:16.120 --> 00:08:19.197 position and they're not going to stop. 00:08:19.280 --> 00:08:21.557 Um if it makes you feel any better, in 00:08:21.640 --> 00:08:23.437 order to do this, you have to have a 00:08:23.520 --> 00:08:26.557 whole country of, you know, engineers, 00:08:26.640 --> 00:08:31.117 scientists, nerds, money, hardware, and so forth. There are not going to be many 00:08:31.200 --> 00:08:33.477 company countries that can do this on 00:08:33.560 --> 00:08:38.237 their own, but China's clearly one, America's another one with our allies, 00:08:38.320 --> 00:08:41.317 of course. Maybe there'll be a third or fourth. And 00:08:41.400 --> 00:08:44.597 as we look not just in national security, but to how this country's 00:08:44.680 --> 00:08:47.557 dealing with AI from A to Z, 00:08:47.640 --> 00:08:52.037 how would you tell the American leadership to deal with AI? What's the 00:08:52.120 --> 00:08:54.237 left and right guardrail? What's the 00:08:54.320 --> 00:08:56.037 role of private sector and how do we 00:08:56.120 --> 00:08:59.357 build consensus? So so so 00:08:59.440 --> 00:09:01.517 let me say some some uncomfortable 00:09:01.600 --> 00:09:04.557 facts, which I think are true and if you 00:09:04.640 --> 00:09:08.837 don't believe me, please write what I say down and then check me on a year in 00:09:08.920 --> 00:09:10.117 a year. 00:09:10.200 --> 00:09:13.397 Um we build an enormous number of tanks, 00:09:13.480 --> 00:09:17.557 ships, missiles, uh human-powered jets, 00:09:17.640 --> 00:09:19.397 driven jets, etc. 00:09:19.480 --> 00:09:22.877 Um they're all now vulnerable. 00:09:22.960 --> 00:09:25.917 Anything that's exquisite, big, slow, 00:09:26.000 --> 00:09:29.997 and emits hit heat, it's a problem. 00:09:30.080 --> 00:09:32.397 So the conclusion you come to if you sit 00:09:32.480 --> 00:09:34.357 in my position is we should be focusing 00:09:34.440 --> 00:09:37.837 our national security on automation of 00:09:37.920 --> 00:09:40.477 land, sea, and you know, air and so 00:09:40.560 --> 00:09:45.477 forth. Uh we have fantastic military, very, very smart people. We should be 00:09:45.560 --> 00:09:50.077 able to make this transition. It takes a while. Um and by the way, it doesn't 00:09:50.160 --> 00:09:54.157 produce less jobs, it just produces different ones. And by the way, this 00:09:54.240 --> 00:09:56.877 doesn't mean the budgets go down, they 00:09:56.960 --> 00:09:59.157 probably go up. But they go up with 00:09:59.240 --> 00:10:01.357 different things. So, for example, why 00:10:01.440 --> 00:10:04.477 would you build artillery for 155 00:10:04.560 --> 00:10:08.757 piece, these big 155 things, which are completely untargeted and only go about 00:10:08.840 --> 00:10:11.757 15 km? Makes no sense. You should be building 00:10:11.840 --> 00:10:15.677 the equivalent thing for a drone, which is far more uh 00:10:15.760 --> 00:10:19.317 it doesn't cost any more, in fact, costs probably less. You can get a lot more 00:10:19.400 --> 00:10:21.117 for the same amount of budget. Plus, you 00:10:21.200 --> 00:10:23.917 can swarm them and so forth and so on. 00:10:24.000 --> 00:10:26.917 So, so as a matter of military doctrine, 00:10:27.000 --> 00:10:31.837 I think all that stuff that we learned watching the you know, the courage of 00:10:31.920 --> 00:10:33.397 the Marines in Iwo Jima, you know, all 00:10:33.480 --> 00:10:36.037 of that stuff is all true, and it's also 00:10:36.120 --> 00:10:39.277 over. Right? That the future combat will 00:10:39.360 --> 00:10:41.837 be largely robotic. Um it will be 00:10:41.920 --> 00:10:44.317 automated, it will be controlled by uh 00:10:44.400 --> 00:10:47.037 the laws of war. America has a very 00:10:47.120 --> 00:10:51.877 clear model of of human control, which I strongly support. I helped write it. 00:10:51.960 --> 00:10:55.357 Um and I feel very, very strongly that this is the direction that the nation 00:10:55.440 --> 00:10:56.877 should take. 00:10:56.960 --> 00:11:00.197 Um with respect to AI in general, 00:11:00.280 --> 00:11:02.717 uh I'll just confess what's really going 00:11:02.800 --> 00:11:05.277 on in my mind. I'm in mourning. 00:11:05.360 --> 00:11:07.997 And I'm in mourning because I started as 00:11:08.080 --> 00:11:10.637 a programmer when I was basically 13 or 00:11:10.720 --> 00:11:15.637 14. And that is essentially over. 00:11:15.720 --> 00:11:19.517 Right? How in How can your entire identity and life's career as a 00:11:19.600 --> 00:11:22.797 programmer, computer scientist be over in one lifetime? It's supposed to be 00:11:22.880 --> 00:11:25.357 like your children or your grandchildren. 00:11:25.440 --> 00:11:28.797 And what's happened I'll I'll give you an example. Uh 00:11:28.880 --> 00:11:32.317 So, the way programming is done at the state of the art now 00:11:32.400 --> 00:11:36.477 Is a programmer wakes up, goes to the office uh cuz they're social and they 00:11:36.560 --> 00:11:38.117 sit there, but they don't have that many 00:11:38.200 --> 00:11:40.197 friends. So, they sit in their office 00:11:40.280 --> 00:11:43.557 and they collect 10 Claude friends 00:11:43.640 --> 00:11:47.597 or 10 Gemini friends. And they set them up with objective 00:11:47.680 --> 00:11:51.397 functions and they watch what they wrote they the code that they write. 00:11:51.480 --> 00:11:55.837 Then they go out to lunch and they make sure that they have long enough tasks 00:11:55.920 --> 00:11:58.957 that they continue working while they're having lunch. 00:11:59.040 --> 00:12:01.477 So, then they go back to the office and they do this and then it's time to go 00:12:01.560 --> 00:12:05.997 home to see the family and so forth and they set up an objective function 00:12:06.080 --> 00:12:09.237 of this is what I want you to do and here's the verification and it's a long 00:12:09.320 --> 00:12:11.877 enough project that it takes them all night. 00:12:11.960 --> 00:12:15.917 So, they've gone from writing code which is what I did and I thought I was the 00:12:16.000 --> 00:12:19.997 hottest thing ever when I was like 20 and I'm sure that wasn't true, but 00:12:20.080 --> 00:12:24.237 welcome to being 20 to now having this kind of power. And 00:12:24.320 --> 00:12:26.557 I've watched what these systems can do 00:12:26.640 --> 00:12:29.957 and it blows me away, right? So, we have 00:12:30.040 --> 00:12:32.637 a major major change happening. This 00:12:32.720 --> 00:12:34.997 happened this started about October of 00:12:35.080 --> 00:12:38.077 last year. It's new for this conference. 00:12:38.160 --> 00:12:39.957 And if you're writing code in any 00:12:40.040 --> 00:12:45.037 traditional way, stop. Right? It's over. 00:12:45.120 --> 00:12:49.477 And go and if you manage a company and you have software programmers say, "Why 00:12:49.560 --> 00:12:53.397 are you still writing code code the way you did it 6 months ago?" 00:12:53.480 --> 00:12:56.437 This has a lot of implications globally 00:12:56.520 --> 00:13:00.877 because it means that the productivity of software, which has 00:13:00.960 --> 00:13:05.997 always been an issue, is going to become far far greater. It also means that 00:13:06.080 --> 00:13:07.677 individuals will be able to build 00:13:07.760 --> 00:13:10.277 incredibly powerful applications and so 00:13:10.360 --> 00:13:13.117 forth. You saw for example this huge 00:13:13.200 --> 00:13:17.597 sell-off in software as a service stocks, essentially software 00:13:17.680 --> 00:13:21.917 traditional software stocks. That's because the market decided that those 00:13:22.000 --> 00:13:25.317 products would literally go away, which of course is not true. They're probably 00:13:25.400 --> 00:13:30.157 oversold, but the they're directionally in trouble and they themselves will have 00:13:30.240 --> 00:13:34.077 to adapt. Well, as a moderator, as a father, I have to 00:13:34.160 --> 00:13:37.277 follow up on something that's not in my notes. You talked about these young 00:13:37.360 --> 00:13:40.037 people should stop coding now. 00:13:40.120 --> 00:13:44.757 For the young people here, what should they do? Study plumbing or 00:13:44.840 --> 00:13:48.277 what? So, what what it says is that the 00:13:48.360 --> 00:13:53.877 traditional writing code, you know, like if then else, the stuff that I used to do, 00:13:53.960 --> 00:13:55.117 you're not going to do that anymore. 00:13:55.200 --> 00:13:58.197 What you're going to have to learn is to write down what you want in a way that 00:13:58.280 --> 00:14:02.917 the computer can understand it and then monitor it producing it. 00:14:03.000 --> 00:14:05.477 Do you remember two or three years ago there was this whole thing about 00:14:05.560 --> 00:14:08.757 hallucinations and we couldn't trust these systems and they were off, you 00:14:08.840 --> 00:14:12.757 know, falling in love with the the therapist and things like that. That's 00:14:12.840 --> 00:14:17.397 largely under control. Uh I can sit in a meeting and somebody will say, "Well, we 00:14:17.480 --> 00:14:20.477 just don't have the right user interface." And I can basically say, 00:14:20.560 --> 00:14:24.837 while the meeting's going on, I can specify within these tools, "Build me a 00:14:24.920 --> 00:14:26.437 UI with these features." And it 00:14:26.520 --> 00:14:29.117 produces. That doesn't work yet. 00:14:29.200 --> 00:14:34.517 But that's the change. So, you learn the new programming, if you will, is 00:14:34.600 --> 00:14:38.277 learning how to manage the programming tools. You've gone from building the 00:14:38.360 --> 00:14:43.157 house to becoming the architect and the builder is under your control. 00:14:43.240 --> 00:14:44.717 So, there is some reason for optimism. 00:14:44.800 --> 00:14:48.397 I'm glad to hear that. And I should say, by the way, that the people who are it's 00:14:48.480 --> 00:14:51.197 always been true, sorry to go on about this, 00:14:51.280 --> 00:14:53.517 it's always been true in software that 00:14:53.600 --> 00:14:56.917 the very tippy-top programmer was worth 00:14:57.000 --> 00:14:59.557 10 times what the next level was. 00:14:59.640 --> 00:15:02.437 Um it's similar in if you think about 00:15:02.520 --> 00:15:06.757 football salaries or NBA salaries, their productivity was so great. 00:15:06.840 --> 00:15:08.957 So, my guess is that there will be a 00:15:09.040 --> 00:15:13.957 smaller number of software people in aggregate, but the very, very top ones, 00:15:14.040 --> 00:15:17.877 the top ones that can manage this productivity, will have enormous 00:15:17.960 --> 00:15:21.157 economic return to themselves, their corporations, things like that. That's 00:15:21.240 --> 00:15:24.437 so interesting. Thank you. I want to return to national security. You talked 00:15:24.520 --> 00:15:28.277 about missiles and bombers and all of that. I'm old enough to remember the 00:15:28.360 --> 00:15:30.757 missile gap with the Soviets, the bomber 00:15:30.840 --> 00:15:33.437 gap. Is there any kind of an AI 00:15:33.520 --> 00:15:35.877 predictability gap because you can't 00:15:35.960 --> 00:15:39.157 really count AI? You can't track it from 00:15:39.240 --> 00:15:41.357 a satellite. So, if you can't if you 00:15:41.440 --> 00:15:45.477 can't count or map an enemy's AI, an 00:15:45.560 --> 00:15:48.997 adversary can't map or count yours. 00:15:49.080 --> 00:15:52.877 What does that do to deterrence and stability? 00:15:52.960 --> 00:15:56.397 So, there's a Again, let me give you a bit of the 00:15:56.480 --> 00:16:00.797 background. Um AI went from language to language to 00:16:00.880 --> 00:16:04.557 language to action. So, when you go and look at the robots 00:16:04.640 --> 00:16:09.997 that we have here in the demo, um that you can communicate with them 00:16:10.080 --> 00:16:13.597 Uh they they look at something, you can communicate with them, and they take an 00:16:13.680 --> 00:16:16.117 action. Those are called VLA models. 00:16:16.200 --> 00:16:19.317 That technology is emerging very, very fast. 00:16:19.400 --> 00:16:23.037 Um those require an enormous amount of training data. 00:16:23.120 --> 00:16:26.077 The reason language to language works so well is cuz there was so much training 00:16:26.160 --> 00:16:29.157 data. And all And we sucked in all the world's language and built all these 00:16:29.240 --> 00:16:33.637 incredibly powerful models. And one of the emergent behaviors of 00:16:33.720 --> 00:16:36.517 these things is they could begin to do planning. 00:16:36.600 --> 00:16:39.997 So, somehow they learned how to do planning. 00:16:40.080 --> 00:16:43.357 So, you can say uh again, it's called chain of thought 00:16:43.440 --> 00:16:47.877 reasoning. Um you can say, "Show me the steps 00:16:47.960 --> 00:16:51.797 to do something." So, all of that is pretty well assembled. 00:16:51.880 --> 00:16:56.197 The next thing is prediction. And these models are very good at 00:16:56.280 --> 00:17:00.517 predicting things. So, in war, for example, you do predictive battlefield 00:17:00.600 --> 00:17:04.197 analysis. I want to know what happens if I do this, what happens? Very 00:17:04.280 --> 00:17:07.677 reasonable. The same is true in a company or government or whatever. It's 00:17:07.760 --> 00:17:11.917 not unique to the military. The next one is called reinforcement 00:17:12.000 --> 00:17:15.637 learning, and that's a simulation task. And that's 00:17:15.720 --> 00:17:17.757 how we would, example won the the game 00:17:17.840 --> 00:17:22.837 of Go at Google, which was thought to be incomputable. That's how 00:17:22.920 --> 00:17:27.037 something called AlphaFold was done at Google, which is a 00:17:27.120 --> 00:17:29.717 Everyone won Nobel Prizes for that. 00:17:29.800 --> 00:17:31.677 The ability to understand how proteins 00:17:31.760 --> 00:17:36.877 fold is a very big deal. This technology is generally available now if you have 00:17:36.960 --> 00:17:39.757 the data. So, if you want to do the kind of stuff 00:17:39.840 --> 00:17:43.477 you're talking about for a national security reasons, you have to have a 00:17:43.560 --> 00:17:45.877 great deal of training data. The only 00:17:45.960 --> 00:17:48.157 place I have found that has enough 00:17:48.240 --> 00:17:51.277 training data is inside of Ukraine. And 00:17:51.360 --> 00:17:53.917 shockingly, Ukrainians will not release 00:17:54.000 --> 00:17:58.797 it for obvious reasons. I'm joking. It's highly classified. You can't get to it 00:17:58.880 --> 00:18:00.597 for all sorts of very good national 00:18:00.680 --> 00:18:04.077 security reasons. So, for America to win 00:18:04.160 --> 00:18:06.277 in this particular area, we're going to 00:18:06.360 --> 00:18:08.197 have to have lots of training data. One 00:18:08.280 --> 00:18:11.357 way to get that is by doing things like 00:18:11.440 --> 00:18:15.757 we have in the cage outside, where you basically have drone competitions. 00:18:15.840 --> 00:18:20.477 Um, and I think my guess is that our military ranges will ultimately have 00:18:20.560 --> 00:18:23.997 drone-on-drone competitions with live fire, and that's how the training will 00:18:24.080 --> 00:18:26.157 occur. I know you've been thinking a lot 00:18:26.240 --> 00:18:28.517 about one of the critical enablers of 00:18:28.600 --> 00:18:31.837 AI, which is energy. The huge amounts of 00:18:31.920 --> 00:18:33.837 electricity that will be needed. Do you 00:18:33.920 --> 00:18:37.237 see the need for what? More wind, solar, 00:18:37.320 --> 00:18:40.397 small modular reactors? How do we power 00:18:40.480 --> 00:18:43.157 the AI future that is undeniably coming 00:18:43.240 --> 00:18:47.757 our way? So, um just one more thing on energy for the 00:18:47.840 --> 00:18:51.837 national security. So, our military 00:18:51.920 --> 00:18:56.157 has now, which is great, has now figured out they need real classified data 00:18:56.240 --> 00:19:00.557 centers. And so, they've started looking at how to partner with the commercial 00:19:00.640 --> 00:19:02.157 sector and so forth in this way, and I 00:19:02.240 --> 00:19:04.557 think this is wonderful. We're not going 00:19:04.640 --> 00:19:10.037 to be safe without our government having its own very, very powerful data center, 00:19:10.120 --> 00:19:13.557 something which they've never really had. Um, and when I was working for the 00:19:13.640 --> 00:19:17.677 Secretary of Defense and we're on the Ash Carter stage, so I'll always owe Ash 00:19:17.760 --> 00:19:20.117 a lot for getting me involved with this. 00:19:20.200 --> 00:19:23.517 There was essentially none. So the military now has a plan to do this. This 00:19:23.600 --> 00:19:26.117 is very welcome. 00:19:26.200 --> 00:19:30.877 In general, you're going to read Let's go through this sort of what 00:19:30.960 --> 00:19:32.837 you'll read in the press. 00:19:32.920 --> 00:19:35.197 The AI boom has contributed between 1 00:19:35.280 --> 00:19:40.357 and 2% of the GDP of America. That's a very big number. 00:19:40.440 --> 00:19:44.237 For most of you, you're going to say, "My God, these people are idiots." 00:19:44.320 --> 00:19:48.477 I mean, how could you possibly waste all that money? That's not my view. I think 00:19:48.560 --> 00:19:50.557 that the AI revolution is under hyped, 00:19:50.640 --> 00:19:53.477 not over hyped. This is my opinion. 00:19:53.560 --> 00:19:57.677 And I started one of a number of 00:19:57.760 --> 00:20:01.277 data center businesses to study this and to build data centers. 00:20:01.360 --> 00:20:05.037 And I can tell you roughly what you conclude. First, 00:20:05.120 --> 00:20:08.157 the interconnection grid is not going to 00:20:08.240 --> 00:20:12.357 survive all of this new power. There's not enough switches, there's not enough 00:20:12.440 --> 00:20:14.597 wires, there's not enough management. 00:20:14.680 --> 00:20:18.797 The interconnection queues, as they're called, are are very long one. 00:20:18.880 --> 00:20:22.557 So what people are now doing is they're building data centers in remote places 00:20:22.640 --> 00:20:26.757 with their own power plants. The technical term for that is behind behind 00:20:26.840 --> 00:20:29.877 behind the meter. What they do is they build essentially 00:20:29.960 --> 00:20:33.997 air-cooled data centers of enormous scale. You've seen the pictures. They're 00:20:34.080 --> 00:20:36.837 essentially air air-cooled systems. They 00:20:36.920 --> 00:20:38.517 don't use as much water as you think. 00:20:38.600 --> 00:20:41.717 The only water they use is internal for 00:20:41.800 --> 00:20:45.797 internal transfer. Very little water escapes. But they use a lot of 00:20:45.880 --> 00:20:47.757 electricity. And they're measured in 00:20:47.840 --> 00:20:50.957 megawatts. The largest data centers are 00:20:51.040 --> 00:20:53.677 in the 1 and 2 gigawatt stage, and there 00:20:53.760 --> 00:20:56.237 are people talking about A gigawatt data 00:20:56.320 --> 00:20:58.277 center costs something on the order of 00:20:58.360 --> 00:21:01.517 50 billion with a B number. 00:21:01.600 --> 00:21:06.237 How could these people in my industry be so crazy to raise all that money and 00:21:06.320 --> 00:21:09.797 build all the centers and the answer is they see the demand. 00:21:09.880 --> 00:21:13.877 And if you the my industry has changed so that if you want to make more money, 00:21:13.960 --> 00:21:17.317 you need more servers cuz the servers are how you make the money cuz that's 00:21:17.400 --> 00:21:20.597 the reasoning. So absent an AI 00:21:20.680 --> 00:21:24.477 breakthrough algorithmically, which is a change to what is called the transformer 00:21:24.560 --> 00:21:28.877 architecture which is invented at Google, thank you, in 2017, 00:21:28.960 --> 00:21:31.797 um we're sort of stuck with this cost structure. 00:21:31.880 --> 00:21:36.597 So my own opinion is you're going to see a lot of remote America 00:21:36.680 --> 00:21:39.997 build these behind the meter power plants that takes forever to get the 00:21:40.080 --> 00:21:44.557 turbines and all of that. And I think the real limit to AI is not energy, it's 00:21:44.640 --> 00:21:48.237 actually cash. Because when you add up the cost of 00:21:48.320 --> 00:21:52.597 these, if you take 50 I'm just round numbers, 50 billion a gigawatt, so 10 00:21:52.680 --> 00:21:54.957 gigawatts is half a trillion dollars. 00:21:55.040 --> 00:21:57.597 How many companies, countries, and so 00:21:57.680 --> 00:22:00.197 forth can hand an industry a trillion 00:22:00.280 --> 00:22:04.877 dollars of capital? Very, very few. The Chinese could certainly do it. I don't 00:22:04.960 --> 00:22:08.077 know if they're doing it. I'm going to try to find out. In the America, there 00:22:08.160 --> 00:22:11.437 are people who hope that's going to happen. It's interesting that you can 00:22:11.520 --> 00:22:16.077 finance these things. Because the the the the brilliance of 00:22:16.160 --> 00:22:20.117 the American capital market allows us to borrow that kind of money. So for 00:22:20.200 --> 00:22:24.597 example, the Europeans can't do this, right, which they're sort of sore about. 00:22:24.680 --> 00:22:28.757 But this is good for America. Mhm. So interesting. Of your many very valuable 00:22:28.840 --> 00:22:32.397 books, my favorite is the one you co-authored with Henry Kissinger. I 00:22:32.480 --> 00:22:36.037 recommend it to everyone. Kissinger, of course, made a career of looking at the 00:22:36.120 --> 00:22:38.397 world and deciding if it was organized 00:22:38.480 --> 00:22:41.917 along values, geography, military 00:22:42.000 --> 00:22:44.437 necessity, trade. As you look to the 00:22:44.520 --> 00:22:46.837 future, Dr. Kissinger's not here to ask, 00:22:46.920 --> 00:22:49.357 do you see the world, Dr. Schmidt, 00:22:49.440 --> 00:22:53.517 redrawing the map along AI alignments? 00:22:53.600 --> 00:22:57.197 The Chinese model, our model, energy 00:22:57.280 --> 00:23:00.917 providers in in the Gulf, will AI redraw 00:23:01.000 --> 00:23:04.437 the map of the world? Well, you know, Henry was my closest 00:23:04.520 --> 00:23:08.757 friend. And so I miss him, you know, every day. And I don't know how to bring 00:23:08.840 --> 00:23:12.157 him back. Uh The empty chair is for his spirit. 00:23:12.240 --> 00:23:16.597 >> Yes, he would be He would be very helpful now in the Iran crisis, for 00:23:16.680 --> 00:23:19.437 example. He understood all that very well. 00:23:19.520 --> 00:23:20.837 And um 00:23:20.920 --> 00:23:23.917 Henry was very concerned about 00:23:24.000 --> 00:23:27.957 competition between America and China in AI. 00:23:28.040 --> 00:23:32.077 That it would somehow unleash an unstable situation. 00:23:32.160 --> 00:23:36.157 And he felt very strongly and started a so-called track two dialogue, which 00:23:36.240 --> 00:23:39.477 continues uh with the Chinese. 00:23:39.560 --> 00:23:44.157 And the the concerns have I'll give you an example. If you look at Methuselah, 00:23:44.240 --> 00:23:47.277 which is the Anthropic model that's so powerful, 00:23:47.360 --> 00:23:50.157 uh we collectively believe that 00:23:50.240 --> 00:23:54.357 Methuselah is just the first of a set of very powerful models that have the 00:23:54.440 --> 00:23:57.517 ability to, for example, do cyber attacks. 00:23:57.600 --> 00:24:01.797 And this is not a desirable thing. It's desirable if you're in offensive cyber 00:24:01.880 --> 00:24:03.517 attack mode, but it's certainly not 00:24:03.600 --> 00:24:06.117 desirable uh if you're on a company or 00:24:06.200 --> 00:24:09.277 or in the military and so forth. 00:24:09.360 --> 00:24:11.877 And so you have to to think of it in the 00:24:11.960 --> 00:24:17.037 following way. You have these enormously powerful models uh coming out. 00:24:17.120 --> 00:24:20.437 How many will there be? Well, there's four or five in the US. 00:24:20.520 --> 00:24:23.797 There's at least three powerful ones in China. There'll be a few others. Let's 00:24:23.880 --> 00:24:27.477 say there's 10. How do you want to organize the world? 00:24:27.560 --> 00:24:29.557 Can you have meetings where every all 00:24:29.640 --> 00:24:33.717 the players agree to limitations? 00:24:33.800 --> 00:24:36.797 And the concern goes something like this. 00:24:36.880 --> 00:24:39.077 Um you and I are on opposite sides, 00:24:39.160 --> 00:24:42.077 which we would never be, and we agree. 00:24:42.160 --> 00:24:47.197 But then I build something which you know I'm doing, and it's really good. 00:24:47.280 --> 00:24:51.557 And I'm going to get a significant advantage by undercutting you, by by 00:24:51.640 --> 00:24:54.677 running ahead of you, by violating our agreement. 00:24:54.760 --> 00:24:56.797 Uh it sounds a lot like OPEC, 00:24:56.880 --> 00:24:59.797 Right? So, we've not We don't have 00:24:59.880 --> 00:25:04.277 language among the countries yet on how to handle this. The 00:25:04.360 --> 00:25:08.957 Trump administration has figured this out and wants to start working on this, 00:25:09.040 --> 00:25:10.877 which I think is very welcome. 00:25:10.960 --> 00:25:15.597 And we'll see. The general issue of global instability 00:25:15.680 --> 00:25:19.397 is actually more even more serious than that. Mythos is just a good example that 00:25:19.480 --> 00:25:24.037 everybody's familiar with. But the Chinese models are open-source 00:25:24.120 --> 00:25:26.157 open weights. 00:25:26.240 --> 00:25:28.677 So, for example, this means that if you 00:25:28.760 --> 00:25:31.277 want to take deep seek V4 and you want 00:25:31.360 --> 00:25:35.717 to add powerful capabilities and you're evil, which none of you are, 00:25:35.800 --> 00:25:38.877 uh and I'm not suggesting the Chinese would do this, but somebody could do 00:25:38.960 --> 00:25:42.877 this. Some evil person exists in the world. How would we know that? How would 00:25:42.960 --> 00:25:45.477 we police it? How would we see it? 00:25:45.560 --> 00:25:49.877 At least with the uh American models, the big ones are under corporate 00:25:49.960 --> 00:25:52.317 control, proprietary, we know who to 00:25:52.400 --> 00:25:57.357 call, the military can call, the police can call, and so forth. But if you 00:25:57.440 --> 00:26:00.917 release this stuff open-source, it's dangerous enough 00:26:01.000 --> 00:26:03.677 that a a lone actor, a new kind of 00:26:03.760 --> 00:26:08.317 terrorist could emerge. Obviously, we want to avoid that. And we want to And 00:26:08.400 --> 00:26:11.837 we want to acknowledge it's possible and we want to get ahead of it. And it 00:26:11.920 --> 00:26:14.157 requires multi-stakeholder agreement. 00:26:14.240 --> 00:26:17.997 Hm. Well, for all of us insomniacs, you've given us more reason to lose 00:26:18.080 --> 00:26:21.477 sleep at night. Um I have enough questions to fill the day, but we always 00:26:21.560 --> 00:26:25.117 love to hear what's on the mind of of the audience. Um if you will raise your 00:26:25.200 --> 00:26:26.917 hand and I'll uh we have about 10 00:26:27.000 --> 00:26:30.637 minutes to for questions. Uh anyone? 00:26:30.720 --> 00:26:33.637 Uh yes, please. Standing here. 00:26:33.720 --> 00:26:35.077 Hang on. There there there's a microphone coming. 00:26:35.160 --> 00:26:39.040 >> Yeah, we'll have to get the mic so everybody can hear. 00:26:41.200 --> 00:26:43.077 Oh, you followed Okay. Um I'm curious to 00:26:43.160 --> 00:26:45.957 hear uh Dr. Eric Schmidt's opinion on 00:26:46.040 --> 00:26:50.797 things like UBI, universal basic income, and like Elon Musk likes to call it 00:26:50.880 --> 00:26:54.557 universal high income. How do you think the future of that will look if we ever 00:26:54.640 --> 00:26:57.557 get to that stage with AI? So, the tech 00:26:57.640 --> 00:27:01.877 industry is full of people who pontificate on this stuff, and they're 00:27:01.960 --> 00:27:05.037 not trained in economics, they're not trained in social sciences cuz they 00:27:05.120 --> 00:27:06.597 didn't have time to take those classes. 00:27:06.680 --> 00:27:08.837 They were too busy programming. 00:27:08.920 --> 00:27:13.157 Um so, first place, maybe we should ask people who have actually studied this. 00:27:13.240 --> 00:27:17.317 So, my in talking to economists who have 00:27:17.400 --> 00:27:23.037 studied labor economics and so forth, the argument goes something like this. 00:27:23.120 --> 00:27:26.557 Um the power of AI in terms of efficiency 00:27:26.640 --> 00:27:30.717 is so great that companies will adopt it, 00:27:30.800 --> 00:27:34.597 and the companies themselves become more profitable and bigger. 00:27:34.680 --> 00:27:38.877 And although there's plenty of internal disruption, so it's painful, 00:27:38.960 --> 00:27:41.917 the fact of the matter is the company gets bigger and the economy grows 00:27:42.000 --> 00:27:47.677 overall. That has been true basically for 200 years, and it's been true forever. 00:27:47.760 --> 00:27:52.797 So, the first question is, do you think that that will continue? And I do. 00:27:52.880 --> 00:27:57.357 And I think that what will happen is we'll have a shortage of job of people 00:27:57.440 --> 00:28:00.557 for jobs because the economy will grow so quickly 00:28:00.640 --> 00:28:03.877 because of the productivity. That's the optimistic view. 00:28:03.960 --> 00:28:07.877 The accelerationist view is a stronger version of that, 00:28:07.960 --> 00:28:11.637 which is that things become so inexpensive that we can all sit around 00:28:11.720 --> 00:28:16.117 on the beach, uh and things like that. I think my own view, 00:28:16.200 --> 00:28:18.637 and again, this is conjecture, is that's 00:28:18.720 --> 00:28:22.237 probably unlikely. And the reason is, 00:28:22.320 --> 00:28:26.797 in the at least in the American system, uh we love lawyers fighting lawyers, we 00:28:26.880 --> 00:28:30.317 love watching sports teams, we love We're not going to be sitting around uh 00:28:30.400 --> 00:28:34.357 drinking mai tais and so forth on the beach. We'll just engage in a higher 00:28:34.440 --> 00:28:37.917 level of competition. Um and I think that that will provide 00:28:38.000 --> 00:28:40.717 plenty of employment. Uh there's lots of 00:28:40.800 --> 00:28:43.757 evidence. So, many, many people blame AI 00:28:43.840 --> 00:28:45.837 for stuff. When you You at the actual 00:28:45.920 --> 00:28:48.717 job losses, there's two categories. The 00:28:48.800 --> 00:28:53.877 first is mine, right? There's no question that if you are a young 00:28:53.960 --> 00:28:57.957 software person entering the software field, it's harder to get a job for the 00:28:58.040 --> 00:29:02.277 reasons that I stated. The other one are low-end or low-paid customer service 00:29:02.360 --> 00:29:05.477 workers. If you think about it in 10 years, 00:29:05.560 --> 00:29:07.797 they'll probably be a lot more. 00:29:07.880 --> 00:29:12.037 Right? But it's it's much slower than people think. I don't spend my time 00:29:12.120 --> 00:29:15.917 worrying about this question. That's very helpful, thanks. Yes, right 00:29:16.000 --> 00:29:22.800 here in front. This is Yes, ma'am. Here it comes. The microphone is coming this way. 00:29:24.440 --> 00:29:27.837 Oh, hello. I'm Dr. IG, private sector. 00:29:27.920 --> 00:29:31.877 Could you please put it up? I have a question. One of the things 00:29:31.960 --> 00:29:33.877 that I hear 00:29:33.960 --> 00:29:36.117 within the field of AI with the people 00:29:36.200 --> 00:29:39.877 working on AI is that everybody talks 00:29:39.960 --> 00:29:42.077 about a human-centered 00:29:42.160 --> 00:29:45.437 AI. Yet, when we go at the technical 00:29:45.520 --> 00:29:48.437 level, that falls or is not at all. 00:29:48.520 --> 00:29:50.277 So, and I believe that one of the 00:29:50.360 --> 00:29:52.757 reasons is because most of the people 00:29:52.840 --> 00:29:58.117 that are building AI are engineers, coders, programmers, and so 00:29:58.200 --> 00:30:01.597 on. And now think people are really thinking 00:30:01.680 --> 00:30:04.837 that we need to bring all the humanistic 00:30:04.920 --> 00:30:07.277 humanic aspects, all the brain science, 00:30:07.360 --> 00:30:13.437 cog science, psychology into the AI 00:30:13.520 --> 00:30:18.717 during the entire cycle of building AI, training, and so on, and creating the 00:30:18.800 --> 00:30:21.597 algorithms. So, my 00:30:21.680 --> 00:30:24.677 I have noticed that big companies like 00:30:24.760 --> 00:30:25.997 Google, 00:30:26.080 --> 00:30:29.797 DeepMind, correct me if I'm wrong, now 00:30:29.880 --> 00:30:33.077 they have jobs for psychologists, cog 00:30:33.160 --> 00:30:37.237 scientists. Is this is becoming a reality? I would like to hear your 00:30:37.320 --> 00:30:39.477 opinion on that. Thank you. So, you're 00:30:39.560 --> 00:30:41.557 exactly right, and thank you for making 00:30:41.640 --> 00:30:44.437 this point. So, in our books, Dr. 00:30:44.520 --> 00:30:47.717 Kissinger and I basically said it's AI 00:30:47.800 --> 00:30:51.837 is too important to leave it to people like me, right? 00:30:51.920 --> 00:30:54.277 Who have a technical background. We need 00:30:54.360 --> 00:30:59.077 the great philosophers, um the the the sort of humanists, if as 00:30:59.160 --> 00:31:02.837 you will. Um they are beginning to show up. 00:31:02.920 --> 00:31:07.077 And the reason is the human the humanists produce graduate students. 00:31:07.160 --> 00:31:10.357 And this is the most interesting thing for a graduate student in humanities to 00:31:10.440 --> 00:31:12.877 work on is the impact on their field in 00:31:12.960 --> 00:31:17.557 psychology or or government or political science or whatever of these powerful 00:31:17.640 --> 00:31:21.237 set of tools. So, I think we're in we're getting in better shape. 00:31:21.320 --> 00:31:23.517 The companies use the word human 00:31:23.600 --> 00:31:26.837 alignment to refer to both values and 00:31:26.920 --> 00:31:32.157 safety. The first one is they have a set of tests. So, for example, if you say to 00:31:32.240 --> 00:31:33.677 any of these models, how do I kill 00:31:33.760 --> 00:31:36.277 myself? It will say, "Don't do it. Call 00:31:36.360 --> 00:31:39.797 a suicide hotline." That's added It by 00:31:39.880 --> 00:31:43.557 the way, it learned how to kill yourself, but it's suppressing that 00:31:43.640 --> 00:31:48.117 answer. That's a good thing. The second question is more subtle, 00:31:48.200 --> 00:31:51.557 which has to do when you're talking to it, what's its style, what's its 00:31:51.640 --> 00:31:55.357 communication, what's its value, and that's trained. 00:31:55.440 --> 00:31:58.157 Um the rhetorical question is, do we all 00:31:58.240 --> 00:32:01.077 agree on what human values are? 00:32:01.160 --> 00:32:02.677 I suspect with this audience, we 00:32:02.760 --> 00:32:05.277 probably are in favor of Western liberal 00:32:05.360 --> 00:32:08.637 values, which includes um 00:32:08.720 --> 00:32:11.717 uh rights of women and minorities and uh 00:32:11.800 --> 00:32:15.717 broadly speech, the ability to express yourself, and so forth. Well, what 00:32:15.800 --> 00:32:20.597 happens if the Chinese models are trained in a different way and they take over? 00:32:20.680 --> 00:32:24.317 And because of open source, it's reasonable that the majority of the 00:32:24.400 --> 00:32:27.997 models in the world, because it's open source is just 00:32:28.080 --> 00:32:30.317 cheaper, will be Chinese resident, not 00:32:30.400 --> 00:32:35.597 American resident. Now, there are American values ones being built. Um 00:32:35.680 --> 00:32:39.837 there is a debate about which ones are coming. I'll give an example today, um, 00:32:39.920 --> 00:32:44.237 there's a something called Gemma 4 from Google, which is obviously have American 00:32:44.320 --> 00:32:48.877 values. There's something called Nemo Tron from Nvidia, which has a similar 00:32:48.960 --> 00:32:52.397 values. There's a couple of startups coming up with better ones. I'm very 00:32:52.480 --> 00:32:57.197 excited about those. But, when you say human values 00:32:57.280 --> 00:32:59.397 and alignment, you have to decide tell 00:32:59.480 --> 00:33:03.317 me which ones. I like the US ones. 00:33:03.400 --> 00:33:07.197 It's marvelous. Thank you. Uh, yes, a question almost to the back, the 00:33:07.280 --> 00:33:09.637 gentleman raising his right hand here. 00:33:09.720 --> 00:33:13.637 Yes. I'm coming. A microphone's coming to you, sir. 00:33:13.720 --> 00:33:15.317 Thank you. 00:33:15.400 --> 00:33:17.317 Um, my name's Kevin. 00:33:17.400 --> 00:33:21.037 How do we prevent another Twitter files, 00:33:21.120 --> 00:33:23.797 um, and I want to 00:33:23.880 --> 00:33:28.677 get trust back in science, and 00:33:28.760 --> 00:33:30.357 the, um, 00:33:30.440 --> 00:33:36.037 COVID lab origin, we, you know, with Peter Daszak, 00:33:36.120 --> 00:33:40.037 and Ralph Baric, and she's in the Why can't 00:33:40.120 --> 00:33:41.637 AI 00:33:41.720 --> 00:33:45.117 say yes, Peter Daszak was a bad actor, 00:33:45.200 --> 00:33:50.837 and we're going to correct, uh, rogue scientist, 00:33:50.920 --> 00:33:54.317 rogue, kind of dangerous gain of function 00:33:54.400 --> 00:33:55.917 experiments, 00:33:56.000 --> 00:33:58.917 and we'll have better 00:33:59.000 --> 00:34:03.437 confidence next pandemic. 00:34:03.520 --> 00:34:05.637 Um, I only could hear about half of your 00:34:05.720 --> 00:34:08.717 words. I I'm sorry. So, so Francis Lee 00:34:08.800 --> 00:34:11.437 wrote a book called In COVID's Wake. She 00:34:11.520 --> 00:34:15.037 said, "People don't trust science or 00:34:15.120 --> 00:34:19.877 news because we got gaslit in the COVID 00:34:19.960 --> 00:34:24.797 I understand. Um, let's start with some basic facts, 00:34:24.880 --> 00:34:29.077 and it's Washington, so you have to start with actual facts. 00:34:29.160 --> 00:34:30.237 Um, so 00:34:30.320 --> 00:34:33.437 >> Thank you for that. So, science 00:34:33.520 --> 00:34:36.637 and technology innovation is the cause 00:34:36.720 --> 00:34:40.917 of American greatness over hundreds of years. 00:34:41.000 --> 00:34:42.837 The innovation 00:34:42.920 --> 00:34:45.517 that America has led brought us to this 00:34:45.600 --> 00:34:48.197 extraordinary power around the world. 00:34:48.280 --> 00:34:50.317 And it's really important that science 00:34:50.400 --> 00:34:53.117 remain non-partisan, 00:34:53.200 --> 00:34:56.717 focused, well-regulated, and funded. Um 00:34:56.800 --> 00:35:00.357 I've committed my own I've made lots of money from Google thank thank goodness 00:35:00.440 --> 00:35:04.317 and all that money will eventually go into science in one form or another. Our 00:35:04.400 --> 00:35:06.837 military security depends on science. 00:35:06.920 --> 00:35:09.477 This AI future depends on science. 00:35:09.560 --> 00:35:14.877 So I start with a science statement. With respect to the specific questions that 00:35:14.960 --> 00:35:17.797 you are describing which have to do with misinformation, 00:35:17.880 --> 00:35:22.677 that's not a solved problem because if you think about it, if people 00:35:22.760 --> 00:35:27.037 are willing to pay more for things that scare them than things which are long 00:35:27.120 --> 00:35:29.357 and reasonable, then we have an economic 00:35:29.440 --> 00:35:31.797 misalignment. In other words, the the 00:35:31.880 --> 00:35:35.357 CEO of that vendor is 00:35:35.440 --> 00:35:37.797 has an incentive to not be completely 00:35:37.880 --> 00:35:42.797 truthful. And that's the that conundrum exists in every democracy. But I think 00:35:42.880 --> 00:35:46.237 the most important thing for this audience is to say regardless of your 00:35:46.320 --> 00:35:50.797 political views and so forth, if you want our country to be great, which we 00:35:50.880 --> 00:35:54.717 all do, and you want us to grow, you have to invest in the kind of stuff 00:35:54.800 --> 00:35:57.077 we're doing. I'll give you an example. 00:35:57.160 --> 00:35:59.157 The gains that we're going to see in 00:35:59.240 --> 00:36:02.317 America from the application of AI into 00:36:02.400 --> 00:36:07.437 science and medicine in the next 5 to 10 years, typically done by graduate 00:36:07.520 --> 00:36:10.677 students and postdocs, the faculty supervised, but it's really that sort of 00:36:10.760 --> 00:36:16.477 young cohort is phenomenal. The breakthroughs that we're about to see in 00:36:16.560 --> 00:36:17.637 energy, 00:36:17.720 --> 00:36:20.677 uh quantum uh various quantum effects, 00:36:20.760 --> 00:36:23.637 um in in literally chemistry and how 00:36:23.720 --> 00:36:27.917 bonds work together. If I told you that as a result of these discoveries, 00:36:28.000 --> 00:36:31.317 everything fell going back to the earlier abundance question, everything 00:36:31.400 --> 00:36:35.437 you purchased was twice as powerful, half the price, and so forth, you'd be a 00:36:35.520 --> 00:36:39.397 pretty happy citizen. That's all within our reach, and I'm not even talking 00:36:39.480 --> 00:36:43.117 about healthcare. Um and the gains in terms of drugs, 00:36:43.200 --> 00:36:46.917 uh Demis Hassabis and I were talking. He started a 00:36:47.000 --> 00:36:49.797 company, and he their goal is to solve 00:36:49.880 --> 00:36:54.517 all disease within 20 or 30 years. Wow, that's a pretty good goal. Sign me up 00:36:54.600 --> 00:36:56.757 for that. And and what do you have to 00:36:56.840 --> 00:36:59.637 do? You have to invest in science. 00:36:59.720 --> 00:37:03.197 Thank you. I think we have time for probably one 00:37:03.280 --> 00:37:07.037 more very fast question. Yes, the gentleman right right here. 00:37:07.120 --> 00:37:11.640 And sir, I'm going to ask you to be really brief if it's okay. 00:37:13.520 --> 00:37:16.277 Uh as a country, what do we need to do 00:37:16.360 --> 00:37:19.197 prepare our university, high school, 00:37:19.280 --> 00:37:23.317 elementary students to participate next 5 10 years? What a 00:37:23.400 --> 00:37:26.437 perfect ending question. >> so um 00:37:26.520 --> 00:37:30.717 let's think about us an a 17 or 18 year old. I think people here look like you 00:37:30.800 --> 00:37:33.157 have kids, or maybe grandkids in some 00:37:33.240 --> 00:37:34.557 cases. 00:37:34.640 --> 00:37:37.877 What's their life going to be like? 00:37:37.960 --> 00:37:42.197 Their entire lives will be mediated by this new form of artificial 00:37:42.280 --> 00:37:45.717 intelligence. Um they'll be their friend, they'll be 00:37:45.800 --> 00:37:49.277 their psychologist, they'll be their advisor, they'll be their programmer as 00:37:49.360 --> 00:37:52.157 we've already discussed. My career is over. 00:37:52.240 --> 00:37:54.717 Um what's the most important thing to teach 00:37:54.800 --> 00:37:57.557 them? Is to use those tools. 00:37:57.640 --> 00:38:00.477 So, my proposal, to be very specific, is 00:38:00.560 --> 00:38:04.637 that every university in the nation have a mandatory class 00:38:04.720 --> 00:38:09.797 that's in freshman year, which is how to use these tools. And I'll design one. It 00:38:09.880 --> 00:38:13.317 starts off by learning how to use the power of them, and it ends by building a 00:38:13.400 --> 00:38:17.677 system that grades your your friends in a way that promotes you. Yeah, or pick 00:38:17.760 --> 00:38:19.917 some that makes sense to an 18-year-old 00:38:20.000 --> 00:38:24.837 student. And you sit there and you so what a conservative proposal. You should 00:38:24.920 --> 00:38:29.117 do this in high school. Yes, let's do it in college first to catch all the people 00:38:29.200 --> 00:38:32.677 who didn't get it in high school and then move it into the high schools. 00:38:32.760 --> 00:38:36.517 But uh I'll give you an example. Um in my 00:38:36.600 --> 00:38:38.557 military work, I'm working with a 00:38:38.640 --> 00:38:42.117 20-year-old at Stanford 00:38:42.200 --> 00:38:46.277 who is not graduated yet. And I hired him and I told him 00:38:46.360 --> 00:38:50.797 that the only rule he has is he has to send me a picture of him graduating from 00:38:50.880 --> 00:38:53.837 his university with his parents. 00:38:53.920 --> 00:38:58.197 Right? That's the deal. Cuz he's so good. He He skipped all 00:38:58.280 --> 00:39:01.117 those other steps and he's now doing essentially reinforcement learning 00:39:01.200 --> 00:39:05.237 around swarming. Now, this is the future. I want to 00:39:05.320 --> 00:39:08.757 capture those people and I want to get them I want to get them to be early 00:39:08.840 --> 00:39:14.317 successful. And I'm sure that the power of these tools is so great 00:39:14.400 --> 00:39:17.717 that we can address a lot of the complaints that young people have today 00:39:17.800 --> 00:39:21.837 about their lives, about their economic opportunities, their reach, and so 00:39:21.920 --> 00:39:26.437 forth. And plus, it's fun. Right? The reason I got into this when I was so 00:39:26.520 --> 00:39:30.077 young, and I grew up by the way in Virginia, and my father got me a 00:39:30.160 --> 00:39:34.397 teletype, and off I was going, is I As a young person, I like to make things. 00:39:34.480 --> 00:39:37.637 These are the most powerful tools you could imagine to make things cuz you can 00:39:37.720 --> 00:39:41.637 design almost anything. Right? Including bad things, which hopefully you'll teach 00:39:41.720 --> 00:39:44.237 people not to do. So a very So and 00:39:44.320 --> 00:39:46.677 universities are resistant to this cuz 00:39:46.760 --> 00:39:49.717 they still want to teach in the old way. 00:39:49.800 --> 00:39:52.357 But that's going away. And the new model 00:39:52.440 --> 00:39:54.797 is going to be having professors teach 00:39:54.880 --> 00:39:58.397 students how to ask a creative question, 00:39:58.480 --> 00:40:02.517 not a rote learning question. When I was in Virginia, by the way, when I was in 00:40:02.600 --> 00:40:06.357 seventh grade, I was required to memorize the names of all of the 00:40:06.440 --> 00:40:10.877 counties of Virginia, which were like 50. And I managed to do it. 00:40:10.960 --> 00:40:15.117 Why in the world is that kind of thinking still around? 00:40:15.200 --> 00:40:18.917 Right? When these systems are always available, they're so powerful and they 00:40:19.000 --> 00:40:22.637 can be used for this extraordinary analytical capability. They need 00:40:22.720 --> 00:40:25.197 creative juices, they need oversight. 00:40:25.280 --> 00:40:28.117 That's what we should be teaching for anything. 00:40:28.200 --> 00:40:32.077 Thank you, Tom. No, thank you for Thank you for ending on such an optimistic 00:40:32.160 --> 00:40:37.157 note. Thank you all. Thanks to SCSP for hosting us. And Dr. Schmidt, thank you 00:40:37.240 --> 00:40:41.117 not just for all you've done for our country, but for explaining it so 00:40:41.200 --> 00:40:44.477 clearly. >> you. You're very kind. Thank you all. 00:40:44.560 --> 00:40:48.760 >> Thank you. Thank you. It's great to see you.