Kissinger and the Future of AI ft. Eric Schmidt (YouTube Video Transcript)

(00:00:00) Your YouTube transcript will appear here (00:12:00) Good evening and welcome to the John F. (00:12:03) Kennedy Jr. Forum at the Institute of (00:12:06) Politics. My name is Amo Ganjiga and I'm (00:12:09) a freshman at the college studying (00:12:11) computer science and economics and I'm a (00:12:14) member of the JFK Junior Forum (00:12:16) Committee. Before we begin, please note (00:12:19) the exit doors which are located both at (00:12:22) the park side and the JFK street side of (00:12:25) the forum. In the event of an emergency, (00:12:28) walk to the exit closest to you and (00:12:31) congregate in the JFK park. Please also (00:12:34) take a moment now to silence your cell (00:12:36) phones (00:12:38) and join me in welcoming Harvard College (00:12:40) undergraduate Matteo Kagliero. (00:12:44) [applause] (00:12:47) Good evening everyone. It is my pleasure (00:12:50) tonight to welcome you all to the John (00:12:51) F. Kennedy Junior Forum at the Harvard (00:12:53) Institute of Politics. My name is Mateo (00:12:55) Kagglero. I'm a junior at the college (00:12:56) studying applied math to computer (00:12:58) science with economics. I'm a member of (00:12:59) the John F. Kennedy Junior Forum Student (00:13:01) Committee. We gather tonight at what is (00:13:03) arguably one of the most pivotal moments (00:13:05) in human history. Artificial (00:13:07) intelligence has evolved from being a (00:13:09) futuristic concept from to becoming a (00:13:11) fundamental part of our everyday lives. (00:13:14) quietly but profoundly reshaping (00:13:16) economies, warfare, and the very fabric (00:13:18) of human society. For the first time (00:13:21) since the dawn of the nuclear era, (00:13:23) global stability may hinge not just on (00:13:25) military strength or economic weight, (00:13:28) but also on competing to secure a (00:13:30) competitive advantage on this new (00:13:32) technological intelligence. All in all, (00:13:34) AI embodies something we have never seen (00:13:36) before. a shift in how knowledge is (00:13:39) generated, a shift in how decisions are (00:13:41) made, and ultimately how humanity sees (00:13:44) its own place in the world. What will (00:13:47) our society look like 20 years from now? (00:13:49) And more importantly, how will (00:13:50) artificial intelligence have shaped it? (00:13:53) Now, to comment on these ideas, it is my (00:13:55) pleasure to introduce you all to our (00:13:56) speakers for the evening. Eric Schmidt (00:13:59) served as chairman and CEO of Google (00:14:00) between 2001 and 2011. Today, he serves (00:14:04) as chair and CEO of Relativity Space. He (00:14:06) is a founding partner at innovation (00:14:08) endeavors and is a leading voice in AI (00:14:10) and national security in the US and (00:14:12) around the globe. Graham Allison is the (00:14:14) Douglas Dylan Professor of Government at (00:14:16) the Harvard Kennedy School where he (00:14:18) served as founding dean and director of (00:14:19) its Belelfer Center for Science and (00:14:21) International Affairs. He also served as (00:14:23) assistant secretary of defense and the (00:14:25) first Clinton Administration receiving (00:14:27) the Defense Medal for Distinguished (00:14:28) Public Service. Now, without further (00:14:30) ado, please join me in opening up (00:14:32) tonight's discussion and welcoming our (00:14:33) esteemed guests to the stage. (00:14:36) >> [applause] (00:14:40) [applause] (00:14:47) >> So, thank you for a generous (00:14:49) introduction. It's a extraordinary (00:14:52) pleasure to welcome back to the JFK (00:14:55) Junior Forum, Eric Schmidt, our (00:14:57) colleague and friend uh for a discussion (00:14:59) of Henry Kissinger uh and AI and the (00:15:04) future. Uh as you can see here uh this (00:15:08) picture of Eric and Henry uh uh they (00:15:12) were best friends uh late developing (00:15:15) friendship but became very deep (00:15:18) friendship in which Eric was generously (00:15:21) the tutor for a 95year-old Henry (00:15:25) Kissinger who discovered AI hanging (00:15:28) around the end of a lecture that he had (00:15:31) given uh when Deis Hibu was uh offering (00:15:36) to or beginning to talk about AI and (00:15:40) Henry decided he needed to learn about (00:15:42) this. Uh he called me up. I told him, (00:15:45) "Henry, forget about it. You know that (00:15:48) you don't have any background in science (00:15:50) and technology." I told him, "You don't (00:15:53) know the difference between a chip and a (00:15:55) potato chip." (00:15:57) He said, "That's true, but Eric has (00:15:59) promised to to to teach me." So, we're (00:16:02) very pleased to have him here. We had (00:16:04) him here last year. This is actually uh (00:16:08) uh maybe it will become an annual (00:16:09) tradition. Uh Henry passed two years ago (00:16:14) last week. Uh so he was a hundred years (00:16:17) old. Uh to think of such an amazing life (00:16:20) over that century. Uh he was a person (00:16:23) who made a huge difference in America's (00:16:27) national security and the world and made (00:16:30) a huge difference in the lives of many (00:16:32) many people. uh some of whom began as (00:16:34) his students, some who became as his (00:16:36) tutors uh and many others. So uh Eric uh (00:16:41) has been introduced uh I would just (00:16:45) remind you a couple of things. I would (00:16:46) say f first that Eric was the chief (00:16:50) executive who took Google from a cause (00:16:53) to one of the great companies of the (00:16:55) world (00:16:57) and that's a pretty amazing thing. Uh (00:17:00) secondly, early on he had identified AI (00:17:06) as an arena of the future and that (00:17:09) Google essentially bought up all the (00:17:12) super talent in the world that he could (00:17:13) find including uh uh Deep Mind which was (00:17:18) actually the company that then brought (00:17:21) to Google uh Deus Kasib who got the (00:17:26) Nobel Prize last year for his work at (00:17:29) Google on uh protein. (00:17:32) Mustafa Sullean who's now running uh uh (00:17:36) Microsoft's consumer AI was another part (00:17:39) of the acquisition and many many many (00:17:42) many others. So the other thing to say (00:17:45) about Derek about Eric and why it's such (00:17:47) a pleasure to have him here is that if (00:17:49) one tries to make any sense of all the (00:17:52) claims that are being made about AI, (00:17:54) most of the people who talk loudest are (00:17:58) talking their own book. So when I listen (00:18:00) to Sam Alman or to folks at Enthropic or (00:18:04) folks even Mustafa now at Microsoft, (00:18:08) they have to talk their companies. (00:18:09) >> You're holding holding my book. (00:18:11) >> Well, wait a second. They have to talk (00:18:14) their company's, you know, business, (00:18:16) their future, and they are chasing what (00:18:19) they believe is the biggest pot of gold (00:18:22) at the end of the rainbow that there's (00:18:24) ever been. So, pretty hard to tell what (00:18:26) they actually think and what they're (00:18:28) saying. Eric has kind of graduated to be (00:18:31) the statesman in this domain, having had (00:18:34) a big stake in it, but now having (00:18:38) standing back and I particularly after (00:18:40) his work with Henry, basically trying to (00:18:44) say things as as clearly as he can and (00:18:47) as clearly as he can see them. So I take (00:18:50) him to be a much more valuable source of (00:18:53) clues about what's happening than (00:18:55) listening to most of the people who are (00:18:57) talking their own books. So let's start (00:19:00) with Henry and then we're going to go to (00:19:01) AI and then we'll go to questions in the (00:19:04) audience. (00:19:06) So, uh, Eric, uh, you, uh, spoke (00:19:10) beautifully at at Henry's memorial about (00:19:14) how much impact Henry had on your own (00:19:18) life and even on the, you know, the (00:19:21) questions that you were asking, the (00:19:23) things that you think that matter. So (00:19:27) tell us a little bit about more what (00:19:29) Henry meant in this relationship and (00:19:33) also for people that wouldn't have the (00:19:35) opportunity to to know him, what what (00:19:38) how might they capture a little bit of (00:19:40) this magic? (00:19:42) >> You know, when I when I first met and (00:19:44) thank you, Graham, as always, it's great (00:19:45) to be here. Um, I I met Henry when he (00:19:49) was in his early 80s and normally men in (00:19:54) their early 80s become a little off. And (00:19:58) I was struck by how brilliant he was. (00:20:00) And so I thought to myself, you know, he (00:20:02) he served in World War II. He got the (00:20:04) Bronze Star, you know, immigrated or (00:20:06) escaped from Germany because they were a (00:20:08) Jewish family. and um came to fought in (00:20:12) the war, came on the GI bill to Harvard (00:20:14) uh undergraduate and graduate and (00:20:15) ultimately was a professor here. And I (00:20:18) tried to figure out he must have been (00:20:20) really smart when he was in your (00:20:22) terrace, not that the building existed (00:20:24) at the time. So here's a quote when he (00:20:26) was an undergraduate at Harvard. quote, (00:20:30) "In the life of every person, there (00:20:33) comes a point when he realizes that out (00:20:35) of all of the seemingly limitless (00:20:37) possibilities of his youth, he has in (00:20:40) fact become one actuality. No longer is (00:20:43) life a broad plane with forests and (00:20:46) mountains and beckoning all around. But (00:20:48) it becomes apparent that one's journey (00:20:51) across the meadows has indeed followed a (00:20:54) regular path. that one can no longer go (00:20:57) this way or that but that the direction (00:20:59) is set and the limits are defined. (00:21:02) That's what he wrote when he was your (00:21:03) age. (00:21:06) Um when he was an undergraduate here he (00:21:08) holds the record for the longest (00:21:10) undergraduate thesis in the uh in the (00:21:13) college. After he submitted his (00:21:15) undergraduate thesis on Kant and the (00:21:17) meaning of the world, they instituted a (00:21:19) new rule which still applies to you that (00:21:21) your thesis can be no more than 350 (00:21:23) pages. (00:21:25) FYI, it's (00:21:26) >> true. (00:21:27) >> So, he was clearly an unusually (00:21:29) brilliant polymath. (00:21:32) And uh my own explanation for Henry (00:21:34) aside from you know I really cared (00:21:36) deeply about him (00:21:39) was that watching his family and in (00:21:41) particular his father and he we talked a (00:21:43) lot about this um see the destruction of (00:21:46) the world around him in Germany as the (00:21:48) Nazis took over (00:21:50) and when they fled and he saw the damage (00:21:53) to his father and his father's identity (00:21:54) and so forth. Henry after the war (00:21:57) decided that he would do everything he (00:21:58) can could to avoid a future war. Now, (00:22:02) you can quibble with or argue very much (00:22:05) with the things he did, but you cannot (00:22:07) argue that his that his goal was not (00:22:10) what I just said. The the data supports (00:22:12) that the historians all agree. You can (00:22:15) disagree with the specific tactics, but (00:22:17) at the end of the day, what he sought is (00:22:19) he sought a a world that did not have a (00:22:22) World War II. and think about the issues (00:22:25) that he faced in the 50s and 60s. He (00:22:27) used to tell me about this small group (00:22:30) and he always thought that the most (00:22:31) interesting policy things happen with (00:22:33) small groups and this was a (00:22:35) collaboration between MIT and Harvard (00:22:37) and Rand uh which in the early 1950s de (00:22:41) um um invented mutually assured (00:22:44) destruction and I said Henry who was in (00:22:46) this group is oh anime like [laughter] (00:22:49) you know give me a break right all the (00:22:51) most famous people. So uh he was lucky (00:22:56) uh to be born at the right time. He was (00:22:58) clearly one of the smartest people at (00:23:00) alive at the time. He was lucky to get (00:23:02) out of Germany before his family and he (00:23:04) were killed. He was lucky to be part of (00:23:06) the GI Bill. Um and he had all sorts of (00:23:10) interesting stories. My favorite was one (00:23:12) day I told him you have to go to the (00:23:13) doctor. You know he's old and he goes (00:23:15) okay. And I said I want you to go to the (00:23:17) Mayo Clinic. And I said okay. He's a New (00:23:20) Yorker and the people in Mayo are like, (00:23:22) you know, Midwestern, soft-spoken, (00:23:24) pleasant, nice, not like the intense New (00:23:26) Yorker types. He goes to the Mayo (00:23:28) Clinic, comes back and he said, "I loved (00:23:30) it." I said, "Why would you like the (00:23:32) Mayo Clinic in the middle of Rochester, (00:23:33) Minnesota?" He said, "It reminded me of (00:23:35) the war." I said, "What?" And he goes, (00:23:37) "Well, he was in a he didn't speak (00:23:40) English when he came to the country." (00:23:41) And he worked as a um (00:23:44) uh in a um shaving brush factory where (00:23:47) they made shaving brushes. and he was (00:23:49) drafted as everyone expected and he went (00:23:51) to war and he went with a Wisconsinbased (00:23:53) American group. So he indexed his (00:23:57) identity couldn't hardly speak English (00:23:59) and nobody could understand his English (00:24:01) anyway. He indexed to that group. So he (00:24:03) had all these sort of touching little (00:24:05) stories. I have lots of stories like (00:24:06) that about how he became an American and (00:24:09) of course became an American citizen and (00:24:12) you know the rest of the history. (00:24:14) >> Well, I would say a fantastic person. I (00:24:16) had the good fortune to enroll in a (00:24:20) course here (00:24:21) >> as a student. (00:24:22) >> What was that like? (00:24:22) >> 1965, (00:24:25) God help us, uh, taught by Henry. Uh, (00:24:28) and ever thereafter was part of his (00:24:32) entourage. And I think the thing that (00:24:35) struck me most about him was this (00:24:40) strategic acumen. That is the ability to (00:24:43) take a problem and lift it up to its (00:24:46) whole (00:24:48) 360 degrees of strategic challenge and (00:24:51) then find the points at which the policy (00:24:57) process could be impacted. And uh I (00:25:02) wrote about uh for his 100th birthday (00:25:06) which reached to write a little you know (00:25:08) something that Eric wrote something I (00:25:10) wrote something and I said really (00:25:12) thinking about Henry uh he was not about (00:25:17) if you think about he was often (00:25:18) criticized for his realism or ruthless (00:25:22) realism or real politic but if you look (00:25:26) and look and see what he was doing uh it (00:25:29) wasn't n't raw re real politic (00:25:32) uh just for the advancement of the (00:25:35) interests of one country. It was always (00:25:38) about the construction of a viable order (00:25:42) to prevent catastrophic war. He had (00:25:45) lived through the experience of the (00:25:46) holocaust and the most deadly war in (00:25:49) history. He foresaw the prospect of a (00:25:54) nuclear war that could actually (00:25:57) extinguish life on Earth. (00:25:59) He was caught up in the Cold War, the (00:26:02) fiercest rivalry that we'd seen to that (00:26:05) point between the US and the Soviet (00:26:08) Union. And all the time he was trying to (00:26:11) reach beyond that to some (00:26:14) way to survive for both the US and for (00:26:18) for uh for fellow human beings. So (00:26:23) that's I think that's how he kind of (00:26:26) backed into (00:26:28) the AI issue which for him was (00:26:32) another generation of a somewhat (00:26:35) analogous problem. Well, he had written (00:26:37) his undergraduate thesis on Kant and the (00:26:39) meaning of meaning essentially and um he (00:26:43) was very when he went to hear Demis (00:26:45) speak um it sort of he immediately got (00:26:49) it and he immediately said to himself (00:26:52) what does this mean to being human (00:26:55) and we are today grappling with the (00:26:57) question that he foresaw 20 years ago (00:26:59) when we when we first started working on (00:27:01) this. What does it mean to be human in (00:27:03) the age of AI? what does it mean to be a (00:27:05) child to an adult to be a leader? What (00:27:08) does it mean for economics? What does it (00:27:10) mean for jobs? You know, all of that. (00:27:12) But his core argument was that (00:27:16) this is an ethical change in the sense (00:27:18) that it's like the um uh the various (00:27:23) major changes that we've had in sort of (00:27:26) reasoning, scientific revolution, and so (00:27:28) forth. Because we as humans have never (00:27:30) had a competitor that is not human but (00:27:34) of [clears throat] similar or greater (00:27:35) level of intelligence and it is (00:27:37) unpredictable what we human will do. He (00:27:40) used to say that what would happen is in (00:27:43) magic when people don't understand (00:27:45) things they either decide that it's a (00:27:47) new religion or they take up arms. And (00:27:50) so he would say well are we going to (00:27:51) take up arms to AI or we make it a new (00:27:53) religion? And I said I hope it's a (00:27:55) religion. [laughter] Um because of (00:27:57) course I benefit from the religion I (00:27:58) guess. (00:27:59) >> So in the book that was published this (00:28:02) time last year called Genesis which was (00:28:05) the work that Eric and Henry and Greg (00:28:09) Bundy was doing up to the point at which (00:28:12) he uh passed. He here's a a passage that (00:28:16) clearly Henry wrote. It says quote (00:28:20) talking about the US and China. If each (00:28:22) wishes to maximize its unilateral (00:28:25) position in the AI space, (00:28:29) then the conditions would be such for a (00:28:32) contest between rival military forces (00:28:34) and intelligence agencies the likes of (00:28:37) which humanity has never faced before. (00:28:40) Today, in the weeks and months and days (00:28:42) leading up to the arrival of the first (00:28:45) super intelligence, a security dilemma (00:28:48) of existential nature awaits us. (00:28:52) So you've thought about this super (00:28:54) intelligence as an existential threat (00:28:57) beyond what what we can imagine. (00:29:02) Say more. Yeah. Well, um [clears throat] (00:29:06) there is something which I call the San (00:29:08) Francisco consensus. And the reason I (00:29:09) call it the San Francisco consensus is (00:29:11) everyone in San Francisco believes this. (00:29:14) It may or may not be true, but if you go (00:29:15) to San Francisco, trust me, that's what (00:29:17) they're doing. besides the usual San (00:29:19) Francisco recreational behaviors and um (00:29:24) the the basic build goes that we are (00:29:27) we're we we've gone through a language (00:29:29) revolution you understand that you know (00:29:31) language language chat GPT everybody (00:29:33) here knows what that is agents which are (00:29:36) now on the horizon and these agents (00:29:38) allow you to essentially automate tasks (00:29:41) and the key thing to understand about (00:29:42) agents is they can be concatenated so (00:29:45) you can you know do do this and this and (00:29:47) this and this and this and then they all (00:29:48) get applied together. Why am I talking (00:29:51) about agents? Because I've just (00:29:53) described workflow and that's what (00:29:56) businesses do, what's universities do, (00:29:57) it's what governments do, so forth and (00:29:59) so on. And then the next one is (00:30:01) reasoning. Now reasoning is higher order (00:30:04) function in humans and the re the (00:30:07) reasoning revolution is just beginning. (00:30:09) Um, as of today, it is technically (00:30:12) correct to say that the scaling laws (00:30:15) that operate in AI have not slowed down (00:30:17) yet. The scaling law basically says that (00:30:20) if you put more data and more (00:30:22) electricity and more chips, you get this (00:30:24) emergent behavior one after the other. (00:30:26) And you can see that for example with (00:30:28) Gemini 3, which just came out, which (00:30:30) beat uh OpenAI 5, which just came out, (00:30:33) which beat um Claude 4.5, which just (00:30:37) came out, which beat Deep Seek because (00:30:39) they copied it anyway, and blah blah (00:30:40) blah, you know. So, so it's (00:30:42) >> in case you didn't notice, Gemini 3 is (00:30:44) from Google. (00:30:45) >> Yes. I'm proud to say that we're back in (00:30:49) charge um until the next one. It's it's (00:30:52) very very competitive. And what's (00:30:54) happening is that these massive data (00:30:55) center buildouts uh which are by the way (00:30:58) one of the key drivers of the American (00:31:00) economy are helping not only lift our (00:31:02) economy but they're also building (00:31:03) building this kind of stuff and it's a (00:31:05) whole new world compared to my whole (00:31:07) experience. So the question is what (00:31:09) happens over time. So you have language (00:31:12) agents and reasoning. Well isn't that (00:31:14) what we do (00:31:16) right? We do stuff we communicate and we (00:31:19) do actions. So the San Francisco (00:31:22) consensus is that at some point that (00:31:25) stuff comes together and you get what is (00:31:27) called technically recursive (00:31:29) self-improvement. And recursive (00:31:30) self-improvement is when it's learning (00:31:32) on its own. This is not true today. (00:31:36) Today when you set up one of these huge (00:31:38) data centers, you know what they look (00:31:39) like, you have to tell it what to learn. (00:31:42) But the belief is that this is coming (00:31:43) and there's lots of evidence that this (00:31:45) is coming. The ability for computers to (00:31:47) write programs to generate mathematical (00:31:50) conjectures to discover new facts looks (00:31:52) like it's very very close. Many people (00:31:55) believe that there will be new math (00:31:57) design, new mathematicians, AI (00:31:59) mathematicians deliver in the next year. (00:32:01) So we collectively as an industry (00:32:03) believe that this is going to happen (00:32:04) soon. If you ask the San Francisco (00:32:07) people, they'll say two years, which is (00:32:10) really soon. If you ask me, I double (00:32:12) that to four years, which is really (00:32:15) soon, right? So, it's happening. It's (00:32:17) happening very quickly. I want, and (00:32:19) Henry certainly wanted it, we want it to (00:32:21) be built with American values and human (00:32:23) values. There is a point in my view, and (00:32:26) we talked a lot about this, where (00:32:28) somebody's going to have to raise their (00:32:29) hand and say, "We just went too far. (00:32:32) There's too much danger here. We don't (00:32:34) want to give that agency to the (00:32:36) computer. We want humans to be in charge (00:32:37) of it." It is not agreed to where that (00:32:40) point is, but our book spends a lot of (00:32:42) time talking about where that might be. (00:32:43) Another example would be that you (00:32:45) discovered that the computer has decided (00:32:47) on its own to get access to weapons. (00:32:50) That that's clearly a like everyone (00:32:52) would agree that's not a good idea, (00:32:53) right? It's bad enough that humans have (00:32:55) access to weapons. Imagine if the (00:32:56) computer has it and what are the (00:32:58) criteria by which and you can think of (00:33:00) many other such examples. Fund (00:33:02) fundamentally about human agency. We (00:33:04) also spend a lot of time talking about (00:33:06) the effect on children. Um we're running (00:33:09) a mass experiment on human development (00:33:12) by deploying these systems that are (00:33:14) incredibly addictive. Um whether it's on (00:33:17) iPad or phone or you know whatever for (00:33:20) young people who may not be certainly (00:33:22) not adults may not have their own (00:33:24) identity and they can really be (00:33:25) manipulated. So what does it mean to be (00:33:27) a child whose best friend is non-human? (00:33:29) aside from maybe becoming like a super (00:33:32) nerd. But I don't know what does it (00:33:34) mean. We we we have no data. We don't (00:33:35) know what it means to young young boys (00:33:37) and girls to their development to their (00:33:39) ability to associate. Will they (00:33:41) ultimately rebel and say, "I only love (00:33:43) people. I hate computers." Like, you (00:33:44) know, kids rebel. We we just don't know. (00:33:48) >> So, can you take a next slide? I'm (00:33:50) sorry. I had the clicker, but I left it. (00:33:58) Can you can you guys make the next slide (00:34:00) go up (00:34:04) here? Thank you. So here the so-called (00:34:07) Kissinger challenge. This is Henry 1969. (00:34:11) So Nixon was elected in 1968, (00:34:15) became president January 1969, (00:34:19) appointed Henry, who was a professor (00:34:21) here at Harvard as his national security (00:34:24) adviser. And Henry writes, as you'll see (00:34:27) here, that anyone coming to office in (00:34:30) the late60s could not fail to be aed by (00:34:34) the unprecedented dimensions of the (00:34:37) challenge of peace. And then down to the (00:34:40) bottom line, there could be no higher (00:34:42) duty than to prevent the catast (00:34:46) catastrophe of nuclear war. (00:34:50) So that became for him a defining (00:34:54) challenge in the cold war with nuclear (00:34:57) weapons arsenals developing but quote no (00:35:01) higher duty than to prevent. So how (00:35:06) might this apply currently as we try to (00:35:08) think about the US China and AI? Well, (00:35:11) let me say that I think there's no (00:35:12) higher duty than to preserve human (00:35:15) agency and human freedom, right? The (00:35:18) things that we value collectively the (00:35:20) most. And I think that's going to be a (00:35:23) central challenge for all of you when (00:35:24) you graduate. Um, all of you will be (00:35:27) facing these issues and they're (00:35:28) complicated and they're subtle. Um, (00:35:31) imagine if the internet had been (00:35:32) invented by China and it didn't have the (00:35:35) kind of openness that uh the internet (00:35:37) has today. was under the Chinese (00:35:39) internet. Um, it looks like China is (00:35:42) pursuing a different strategy than what (00:35:45) I'm talking about. In my most recent (00:35:47) visit to China, the way I operate is I (00:35:50) ask engineers technical questions (00:35:51) because they don't lie to you, whereas (00:35:53) kind of everyone else, I'm not so sure. (00:35:55) And I eventually figured out that what (00:35:57) the Chinese are doing is they're really, (00:35:59) really focused on applying AI in their (00:36:01) businesses. And they're going to out (00:36:03) compete with us. We, you know, we will (00:36:06) lose to China for their incredible (00:36:08) adoption of AI in every product because (00:36:11) they're just relentless and they work so (00:36:13) hard. It's called 996, 9 in the morning, (00:36:16) 9 in the evening, six days a week. (00:36:18) Illegal, by the way, in China, illegal (00:36:20) in America, right? But nevertheless (00:36:22) practiced. Um, they're coming. Um, there (00:36:26) do not appear to be as focused on super (00:36:28) intelligence and the path that I was (00:36:30) describing as the San Francisco (00:36:31) consensus. Of course, that could change. (00:36:34) So it appears that the two are pursuing (00:36:37) different paths. One of the questions (00:36:38) for you all and at the graduate level (00:36:41) even undergraduates given it's Harvard (00:36:43) is start figuring out what happens when (00:36:45) these divergent paths hit roadblocks (00:36:48) right because they're both have example (00:36:50) would be in America we have essentially (00:36:53) produced no new electricity because it's (00:36:55) very hard to provision electricity. (00:36:57) China has infinite electricity because (00:37:01) of their incredible investment in (00:37:03) renewables and so forth. They built (00:37:05) something like 120 gawatts of new (00:37:07) renewable energy in the last 5 years at (00:37:09) some number like that. (00:37:10) >> No, no, no. Way way way. Every day in (00:37:14) China, Thursday, Saturday, Monday, every (00:37:18) day another gigawatt of electricity has (00:37:22) added to the grid all year. That's (00:37:24) pretty extraordinary. (00:37:24) >> All year. Every day. So my point is and (00:37:27) here and here by the way just to give (00:37:29) you an example how big is a nuclear (00:37:31) power plant about 1.5 gawatt. (00:37:34) So that's the scale of the electricity (00:37:37) revolution that's going on in China (00:37:39) again using those numbers. How many of (00:37:42) those kinds of plants have we built in (00:37:43) America? Zero. Uh and we're certainly (00:37:46) losing the renewables race to China for (00:37:48) the reasons that everybody here knows. (00:37:50) So they have a lot of power, right? We (00:37:53) don't. We have a lot of chips. they (00:37:56) don't um it sets up the competition (00:37:59) right and each will pursue different (00:38:00) ways. Um one of the technical questions (00:38:04) there's something called diffusion where (00:38:06) what you do is you take a very powerful (00:38:08) model we'll use Gemini 3 a top one at (00:38:10) the moment um and you ask it 10,000 (00:38:13) questions and you take the answers and (00:38:15) then the system can learn from the (00:38:17) questions and answers enough to mimic (00:38:20) without the expense of doing the big (00:38:22) training the big model. So again, (00:38:25) thinking through the strategies that (00:38:26) China will pursue and the ch the (00:38:29) strategies that are available for (00:38:30) America um are is probably pretty (00:38:32) important. What's interesting is both (00:38:34) countries are relying on the private (00:38:36) sector to do this. You would have (00:38:38) thought in Henry's period that you would (00:38:40) have used the government, (00:38:41) >> but in fact our government cannot move (00:38:43) this quickly, the compensation systems (00:38:45) and so forth. It turns out it's probably (00:38:47) true for the Chinese as well. I don't (00:38:49) know at the at the security level if (00:38:51) this is true, but I've not found any (00:38:54) large weird Manhattan type projects (00:38:56) within China, although there's plenty of (00:38:59) people in the private companies that are (00:39:00) working hard on national security. (00:39:02) >> So, on the AI topic in general, tell us (00:39:06) uh just give us a minute or two of what (00:39:08) is the upside, the things that excite (00:39:11) you the most that are one year, two (00:39:13) year, three years that a line of sight (00:39:16) too. Well, the first question is why is (00:39:18) this madness occurring? It must be a (00:39:20) bubble and it's going to crash. (00:39:23) No, it's not a bubble. If if anything, (00:39:25) it's underhyped because you're (00:39:27) fundamentally automating businesses. And (00:39:30) the reason people are spending this (00:39:31) enormous amount of money is to automate (00:39:34) the boring parts or what they view as (00:39:35) the boring parts of their business. So (00:39:37) whether it's billing or accounting or (00:39:39) product design or delivery or inventory (00:39:41) or management or whatever, people are (00:39:43) automating those. Um, and there's an (00:39:46) awful lot there. Think about medicine. (00:39:48) Think about climate change in (00:39:49) engineering, new science. It's (00:39:51) extraordinary. (00:39:57) So, what excites the most of the ones (00:40:01) that you see that you have that you (00:40:03) [clears throat] have a line of sight to (00:40:05) that the rest of us probably are not (00:40:07) seeing yet? (00:40:09) >> Sorry, [clears throat] I'm sorry. (00:40:10) >> I have a cough. (00:40:13) I apologize for that. (00:40:17) >> We can all see in our own imagination (00:40:20) what we're thinking of and then we'll (00:40:21) see what Eric says. Yes. (00:40:23) >> Well, when I started uh when I was in (00:40:26) high school, I was an early programmer (00:40:29) and I delighted in writing code. When I (00:40:31) went to college and graduate school, (00:40:32) that's all I ever wanted to do. I didn't (00:40:34) I ignored all these history things and (00:40:35) things like this. I was the definition (00:40:37) of a nerd at the time. And everything (00:40:40) that I did in my 20s which got me to (00:40:42) where I am has now been completely (00:40:44) automated. (00:40:46) Every aspect of the programming that I (00:40:48) did, every aspect of the design is now (00:40:50) done by computers. [snorts] I recently (00:40:53) had it write a whole program for me and (00:40:56) I'm sitting there watching it define the (00:40:58) classes and the detail of the (00:40:59) interactions and so forth as it's (00:41:01) generating go like holy crap, you know, (00:41:04) the end of me. And I think watching I've (00:41:07) been doing programming for (00:41:10) 55 years. So to see something start and (00:41:13) then end in front of your own life and (00:41:15) you're still alive is really profound I (00:41:17) might say. Now computer science is not (00:41:20) going away. The computer scientist will (00:41:22) be at least until the computer scientist (00:41:24) gets replaced uh will be supervising (00:41:26) this. But the ability to generate code (00:41:30) at the power that these systems can do (00:41:32) is revolutionary. It means that each and (00:41:35) every one of you has a supercomputer and (00:41:37) a super programmer in your pocket. Now, (00:41:40) nobody here is a terrorist, right? But (00:41:43) using it's always easier to use negative (00:41:44) examples. There's plenty of I'll use a (00:41:47) stereotype young men living in the (00:41:48) basement. Their mothers give them food (00:41:50) and they sit there on the equivalent of (00:41:52) crypto for 4chan, you know, paranoia, (00:41:54) whatever. Pick your pick your poison. (00:41:57) They all have the ability to use these (00:41:59) tools to build incredibly powerful (00:42:01) systems, (00:42:03) right? Cyber attacks, other things, no, (00:42:05) whatever they care about. Um, (00:42:09) there was u there's some evidence that I (00:42:11) think it's Manion, the fellow who killed (00:42:13) the insurance executive um was into some (00:42:17) of this and there's some people were (00:42:19) looking at some of his writings. Of (00:42:20) course, he's in jail now, but that he (00:42:22) was somehow influenced with that. Now, (00:42:23) I'm not proving causality here, but it's (00:42:25) an example of some of the darkest (00:42:27) recesses of humanity. (00:42:30) You give those people these kinds of (00:42:31) tools. We have to be ready. Now, the (00:42:33) industry is well aware of this and we're (00:42:35) working on it. It's very important that (00:42:36) a defensive systems be capable. The (00:42:39) eventual solution to AI, by the way, bad (00:42:42) AI is AI fighting good AI fighting bad (00:42:45) AI and that's how it will resolve (00:42:46) itself. (00:42:48) >> Okay. Can we do the next slide, please? (00:42:50) I want to ask about the US China rivalry (00:42:54) in AI as you see it. So this apologies (00:42:58) if the slide's not clear enough but what (00:43:00) it suggests is that if we take the (00:43:03) series of indices (00:43:05) uh if you look at the performance gap in (00:43:09) January of 24 it was significantly (00:43:12) larger than it is today. And then uh (00:43:20) what do we make of this and what do we (00:43:22) make of the likely future? So the chart (00:43:25) is correct, but the people who are (00:43:29) influenced by this claim that it's not (00:43:31) going to be true for long because the (00:43:33) reasoning revolution requires so many (00:43:36) chips and so much of the magic that the (00:43:38) San Francisco people have invented using (00:43:41) that as a sort of moniker that the gap (00:43:43) will widen. (00:43:45) My own view is that the gap will widen (00:43:48) but for different reasons. I think that (00:43:50) the Chinese focus is largely, as I (00:43:53) mentioned, on embedding AI in (00:43:56) everything. Smart toasters, cars, so (00:43:58) forth and so on. They're moving much (00:44:00) more quickly than are the robots. I (00:44:03) think that the vast majority of humanoid (00:44:04) robots will be Chinese AI powered and (00:44:06) Chinese AI manufactured simply because (00:44:09) they know how to drive the cost of (00:44:10) things down. Their supply chains are (00:44:12) incredible. Their cost management, they (00:44:14) work so hard, blah blah blah, all that (00:44:15) kind of stuff. So my guess is that that (00:44:19) it's true that the gap will probably get (00:44:22) larger, but the real question is will (00:44:24) you as a consumer ultimately have a a (00:44:27) better experience as a consumer with a (00:44:28) Chinese product than a US product and (00:44:30) the answer is from a fit and finish (00:44:33) probably the Chinese product and that's (00:44:34) of concern. (00:44:36) >> So let me drill a little further on this (00:44:38) one. So one there's a half dozen (00:44:41) questions about which people are making (00:44:43) bets and you've thought about them quite (00:44:45) deeply. So one is are we going to bet on (00:44:48) uh computer chips or stacks or brains? (00:44:53) Another one is are we going to bet (00:44:55) closed or open? (00:44:57) >> Another one is are we going to bet uh (00:45:00) working hard on AGI or diffusion and (00:45:03) applications. (00:45:05) So if you go across the spectrum there, (00:45:09) uh if I look at the Chinese piece, (00:45:13) uh certainly the guys at Deep Seek think (00:45:16) 200 people, uh who just have brains can (00:45:21) have a reasoning machine that costs (00:45:24) 1,000th (00:45:25) >> of the cost of open AI. And now there's (00:45:29) six other little dragons coming along (00:45:32) from the same from the same space. So (00:45:34) that one makes me worry on the closed (00:45:37) versus open. If I remember our last (00:45:39) conversation, you had pretty much (00:45:41) concluded that open is going to be (00:45:43) closed, but all of our companies are (00:45:46) mostly closed. So what about that? And (00:45:50) then thirdly, uh maybe if there's this (00:45:54) AGI breakthrough, (00:45:56) all the rest of this other stuff won't (00:45:58) matter. But if the if the diffusion and (00:46:02) application is already working in these (00:46:06) other arenas. So just take us another (00:46:09) step on the on the rivalry. (00:46:11) >> So diffusion refers I'll work backward. (00:46:13) Diffusion refers to essentially learning (00:46:16) from one model by the by many pairs and (00:46:19) learning from as we discussed. My own (00:46:22) opinion is that which is not I'm not (00:46:24) sure but I think so that the large (00:46:27) companies will ultimately not release (00:46:28) their largest models. It'll be too (00:46:30) dangerous to do so that they'll subset (00:46:32) them. That's what I would do. No, I (00:46:34) assume that they'll come to that (00:46:35) decision. I think the most interesting (00:46:37) question is open versus closed for the (00:46:40) those of you who don't have a background (00:46:41) in this open source open weights. Uh (00:46:44) open source is what I did for decades. (00:46:47) uh when you use any form of computer (00:46:50) much of the software that you're using (00:46:52) was developed by open source which meant (00:46:54) that the source was published and people (00:46:56) would collectively advance it and there (00:46:58) was a whole movement around this which I (00:46:59) was part of. So I'm relentlessly open (00:47:02) source in my view. Um the major (00:47:05) companies are closed source largely for (00:47:07) economic reasons. Um basically if you (00:47:10) borrow $50 billion from the financial (00:47:12) market they do want a return and saying (00:47:15) to them oh by the way we're going to (00:47:16) give all the models away and you won't (00:47:18) get your return probably isn't a very (00:47:20) good um legal or financial strategy. So (00:47:23) it so the American model seems to have (00:47:25) emerged as as closed. Strangely the (00:47:29) Chinese model is completely open open (00:47:32) weights open um source. (00:47:35) Why? I don't know. Um, one possible (00:47:39) explanation is that the Chinese (00:47:40) government has figured out that they (00:47:42) were losing in closed source because (00:47:43) they couldn't get the hardware and open (00:47:45) source because it's free gives them (00:47:47) proliferation. So, one of the (00:47:49) consequences of open source open ways is (00:47:52) the vast majority of humans on the (00:47:53) planet will use Chinese models. Why? (00:47:56) Because they're free, right? And most (00:47:59) countries cannot afford the computing (00:48:01) and the data centers and so forth. (00:48:02) They're just going to take the Chinese (00:48:03) models for free and embed them. Now, is (00:48:06) that an issue? Absolutely. Because it (00:48:08) comes with Chinese values, Chinese (00:48:10) training, Chinese biases and so forth. (00:48:12) We would prefer to have it be American. (00:48:15) Um, we will see. There is a there's a (00:48:18) couple of open source projects in (00:48:19) America that I am a supporter of, but (00:48:22) they can't raise the $10 billion that (00:48:26) are needed from the public markets to (00:48:28) get to where they're going. So, they are (00:48:29) gems, but they're not at the scale. I've (00:48:33) argued that the US government should (00:48:34) help fund them. I've argued that (00:48:36) philanthropists should help fund them, (00:48:38) but I don't really know. (00:48:41) >> Okay, that's very helpful. So, one last (00:48:44) question and then we're going to take (00:48:46) questions from the audience. If you were (00:48:48) picking two or three questions that a (00:48:51) graduate student or an undergraduate (00:48:54) interested in this space, AI and (00:48:56) geopolitics should be thinking about. (00:48:59) give us two or three questions that we (00:49:02) might have somebody that has an answer (00:49:04) to or part of an answer to the next time (00:49:07) you come. (00:49:07) >> So, so a couple Well, again, we're (00:49:09) dealing with some of the smartest people (00:49:10) in the world here. So, the first (00:49:12) question is what does it mean to be (00:49:13) human in the age of AI? That is a (00:49:16) question that you can write many PhD (00:49:18) thesis on. So, basically, study history, (00:49:20) study philosophy, study how people work, (00:49:22) study economics, and then figure out (00:49:24) what this new technology is going to do. (00:49:26) We always in my industry because we (00:49:28) didn't take those classes we always (00:49:30) ignore those things right you guys are (00:49:32) capable of answering these questions in (00:49:34) ways and if we get out of line maybe you (00:49:36) can use that to call to call us out u (00:49:39) the second question has to do with this (00:49:41) China US rivalry why China versus US the (00:49:45) only two countries that are going to (00:49:46) matter and the reason is you need an (00:49:49) enormous amount of money and an enormous (00:49:51) number of people and as much as I like (00:49:53) Europe Europe is not organized doesn't (00:49:55) have enough people not have enough money (00:49:56) [clears throat] to do it. Um, India is (00:49:58) not organized enough yet to do it, (00:50:00) although they're trying and most of the (00:50:02) other countries don't have enough money, (00:50:03) don't have enough the talent, don't have (00:50:04) the right universities and so forth. And (00:50:06) then the third question has to do much (00:50:08) more with conflict. In a world where you (00:50:12) have terrorists who have access to AI (00:50:15) and you have governments that have (00:50:16) access to AI, what does conflict look (00:50:19) like? What does a terrorist attack look (00:50:22) like against a major power? Obviously, (00:50:24) I'm not endorsing that. It's a terrible (00:50:25) thing. How do we defend against it? Same (00:50:28) thing for China versus US. But what (00:50:30) about Russia versus Ukraine? What about (00:50:33) Europe versus whatever? Trying to (00:50:35) understand how conflict plays out in an (00:50:37) algorithmic warfare where the AI is (00:50:40) driving everything is a very very (00:50:42) fertile area for research and new ideas (00:50:44) and it's just starting. (00:50:48) [clears throat] (00:50:48) >> So, good topics for somebody to work on (00:50:51) in the interim. Let's start here. Please (00:50:54) introduce yourself and a short question. (00:50:57) >> Hi, my name is Teresa. I'm a second year (00:51:00) student here at the Kenny School and I'm (00:51:01) in Professor Allison's national security (00:51:03) class. Uh we had a class talk about n um (00:51:06) cyber security and most of the tech that (00:51:08) underpins us homeland securities are in (00:51:12) private hands. So what government's mod (00:51:14) governance model do you think will (00:51:16) actually help work to coordinate (00:51:18) government and private companies um to (00:51:21) during uh especially during like a AI (00:51:23) type of emergency um and given we talk a (00:51:26) lot about like USChina rivalry um and (00:51:29) despite the technology in investment (00:51:32) rivalry there's also a big difference in (00:51:35) government's model so how do you think (00:51:36) the two governance model will lead uh in (00:51:39) terms of there's a homeland security (00:51:41) cyber crisis (00:51:43) Well, first place, I think in under the (00:51:44) Trump administration, you're not going (00:51:46) to see much regulation of AI. That's (00:51:48) pretty clear. Um, in China, it appears (00:51:50) that they're allowing the the companies (00:51:52) to do whatever they want. Um, although (00:51:55) they have the laws about uh various (00:51:57) things, they don't seem to be (00:51:58) implementing them. So, it looks like (00:51:59) it's an allout business conflict. The (00:52:02) biggest concern I have is cyber attacks. (00:52:04) If you can write code the way that I've (00:52:08) seen these things write and every (00:52:10) company in my world is now using (00:52:12) programmers and AI programmers together, (00:52:15) it's extraordinary. It's happened very (00:52:16) quickly. Look at Claude Code for (00:52:18) example, the most recent one which is by (00:52:20) far at the top from Anthropic and others (00:52:23) coming. Gemini, of course, claims it has (00:52:25) a competitor, but at the moment Claude (00:52:26) Code is a bit better. It looks like um (00:52:29) if you can write code, you can also (00:52:31) write cyber attacks because the (00:52:33) objective function is easy. Just keep (00:52:35) writing code until you break something. (00:52:37) And if you have enough hardware and (00:52:38) enough energy, you can just keep doing (00:52:40) that. So I think there's going to be (00:52:42) many many more cyber attacks. Um and (00:52:44) that doesn't mean from governments. It (00:52:46) could also be from terrorists and and (00:52:48) bad organizations. And I think getting (00:52:50) organized for that would be my primary (00:52:52) concerns. (00:52:53) >> Peace. (00:52:55) Hi, thank you for coming. I'm David (00:52:57) Weidman. I'm MPPP2 here. Um, you you (00:53:01) mentioned uh the strength of current (00:53:04) closed source models. Uh, you can tell (00:53:05) me if I'm wrong, but I think open source (00:53:07) models are about a half generation (00:53:08) behind the current closed source models. (00:53:10) Uh, perhaps it's a little dangerous to (00:53:13) characterize open source verse closed (00:53:15) source as an America versus China issue. (00:53:18) Um (00:53:20) uh and my my fear is that the Silicon (00:53:22) Valley people that you're talking about (00:53:23) are seeking regulatory capture and (00:53:25) that's why they want to fearmonger (00:53:27) around open source models. Uh what would (00:53:30) it take if we wanted to establish a (00:53:32) moratorum on closed source models if (00:53:35) that's something that we seek? What (00:53:36) would it take in order to establish this (00:53:38) moratorium worldwide? I (00:53:40) >> I think I just don't agree with your (00:53:41) question. So I'm sorry. Um I I don't see (00:53:45) the open source leadership in America (00:53:48) and I see nothing but open source (00:53:50) leadership in China. I think those are (00:53:51) the facts. Um so I don't think there's (00:53:53) regulatory capture on the open source (00:53:55) side. Um and I think again under the (00:53:58) Trump administration it's highly (00:54:00) unlikely that you're going to see (00:54:01) significant regulation of the closed (00:54:04) source companies. But I think the closed (00:54:06) source decision is largely driven by (00:54:08) economics not policy. if you see my (00:54:10) point. Literally, you just can't. I (00:54:12) mean, think about the cost of these (00:54:13) things. We're talking about 10 billion (00:54:15) dollars,$20 billion dollars. How would (00:54:16) you raise that without it? (00:54:19) >> And I think Eric, this goes back to if (00:54:22) for making our list of questions, I (00:54:24) would add add to it, how does the (00:54:27) financing considerations impact the (00:54:30) strategic choices? because if it if it (00:54:33) just happens to be an accident of the (00:54:35) financial uh uh structure that wouldn't (00:54:39) necessarily reflect the national (00:54:42) interest that would just reflect the (00:54:45) capital markets as they are. (00:54:46) >> I I think that it's important to (00:54:48) acknowledge that America has the most (00:54:50) extraordinary capital market financial (00:54:53) system in the world by far. I think the (00:54:55) numbers are 60% of the volume and 90% of (00:54:59) of the value dollar and so forth is in (00:55:02) US dollars and so countries that are not (00:55:05) dollar denominated do fear um the power (00:55:08) of that financial market and we see it (00:55:10) in the ability to raise money when I was (00:55:12) in China it was quite clear in talking (00:55:14) to my friends they don't have access to (00:55:17) the depth of the financial market they (00:55:19) literally cannot get the money the (00:55:21) venture financing that was in China is (00:55:23) cut by a factor of five since three (00:55:25) years ago. Now there are many reasons (00:55:27) for that, not just our world and not (00:55:29) just the US. But without that access to (00:55:31) capital, it's very hard to to do these (00:55:34) large models with these complicated (00:55:35) training. Now you can imagine everything (00:55:38) I'm saying changing if the underlying um (00:55:42) algorithms change. There are people who (00:55:44) are working on new non-transformer (00:55:46) models which are less uh economically uh (00:55:49) expensive. There are many people who've (00:55:51) commented on the energy efficiency of (00:55:53) human brains versus the cost of these (00:55:54) data centers. And trust me, our brains (00:55:56) are complicated, but they're not energy (00:55:59) they're not energy expensive. So, so (00:56:01) again, there there there might be a (00:56:03) breakthrough that would change the (00:56:04) calculus that we're discussing. Go (00:56:06) ahead, (00:56:06) >> please. Thank you, Dr. Schmidt. Uh, my (00:56:09) name is Faton Seammanyaku. I'm an MCMPA (00:56:12) student here at Harvard Kennedy School, (00:56:14) originally from Albania. In the age of (00:56:18) AI, you and Dr. Kissinger describe a (00:56:21) world where AI systems begin to (00:56:24) interpret reality for us. If human (00:56:28) strategic judgment becomes shaped even (00:56:30) subtly by machine generated frames, what (00:56:34) becomes the anchor of responsibility in (00:56:36) global affairs? In other words, when an (00:56:40) AI influenced decision produces real (00:56:43) world consequences, who holds the moral (00:56:45) burden? the human who acted, the (00:56:48) institution that de deployed the system (00:56:51) or the algorithm that shaped the (00:56:54) perception. And how should democracies (00:56:57) redesign their institutions before this (00:57:00) ambiguity becomes a geopolitical vulner (00:57:03) vulnerability? (00:57:04) >> A very wellphrased point. So I worry (00:57:08) that the future of democracy is (00:57:11) uncertain simply because I'll use an (00:57:14) American example. We believe in free (00:57:16) speech and I'm certainly in favor of (00:57:18) absolute free speech in America. I'm not (00:57:20) in favor of boosted speech or (00:57:22) algorithmic speech. So where is the (00:57:24) line, right? If if I say something (00:57:27) that's wrong and then the algorithm (00:57:29) because my claim is outrageous decides (00:57:31) to spread it everywhere, is that (00:57:33) appropriate in a democracy? And you can (00:57:35) imagine, and again, I'm not making a (00:57:37) partisan thing. I think any any any part (00:57:39) of the political system can use this for (00:57:42) whatever effect. the ability to generate (00:57:44) misinformation that people really (00:57:46) believe is so simple now that I hope the (00:57:50) answer is better education and critical (00:57:52) thinking among human beings. But you (00:57:55) could imagine if I were evil, which I (00:57:56) hope I'm not, and I I sat down and I (00:57:59) started to flood everyone with my (00:58:01) particular unique messages that I could (00:58:03) overwhelm your belief in truth by simply (00:58:06) relentless copying and repetition. We (00:58:08) know that there's something called (00:58:10) essentially anchor bias. If you hear it (00:58:12) first, then you judge a So if I manage (00:58:14) to get the message to you first that the (00:58:16) building is on fire, which it's not, so (00:58:18) I'm not committing a crime. Um, if I (00:58:20) manage to get that first, you then (00:58:22) anchor from that point. You see the (00:58:24) danger of this. It's very real. I don't (00:58:26) think and my my answer to your question (00:58:28) is that every every democracy will face (00:58:31) this and I think democracies will make (00:58:34) different decisions based on cultural (00:58:36) values and their understanding of the (00:58:38) threat. you won't see a single answer (00:58:40) from democracies. And by the way, the (00:58:42) way authoritarian countries solve this (00:58:43) problem is they just ban it. Right? So, (00:58:46) a lot of this stuff is illegal in (00:58:47) authoritarian countries because they (00:58:48) have the ability to suppress speech. And (00:58:51) obviously, we we're all in favor of free (00:58:53) speech, (00:58:54) >> please. (00:58:55) >> Go ahead. Yes, ma'am. (00:58:57) >> Yes. Thank you very much for for coming (00:58:58) here this evening. So, uh I'm called (00:59:00) Ellen Crane and I'm a fellow at the (00:59:02) Belfare Center. So I'm wondering firstly (00:59:04) could you comment on the role of humans (00:59:07) such as you and me in the very long term (00:59:09) with regards to AI what's our what does (00:59:11) our role become and also um we often (00:59:15) frame you know because of the value (00:59:17) systems the US China debate as a (00:59:19) competition but is there is there value (00:59:22) to thinking about that in terms of a (00:59:24) collaboration and maybe more (00:59:25) interestingly you know you mentioned (00:59:27) Europe being maybe somewhat disorganized (00:59:28) but they also have enormous strength and (00:59:30) enormous talent like Mistral in France (00:59:32) and so And is there something to do (00:59:34) there with um collaboration? (00:59:36) >> Um so so a couple comments on Europe. Um (00:59:40) I was the first investor in Mistral. So (00:59:42) I'm a big MRO fan. Mal cannot raise the (00:59:46) money that their US competitors. They (00:59:48) have the same problem and they're (00:59:50) working on solutions to that. Um with (00:59:53) respect to US and China, the um (00:59:59) I started because of Henry, I I spent (01:00:02) about five years looking at Chinese peer (01:00:04) competition. (01:00:06) And I thought that it would be possible (01:00:08) to get closer to China. And I eventually (01:00:11) through his work and others discovered (01:00:13) that the Chinese were more afraid of (01:00:14) competing partnering with us than we (01:00:16) were with them. So it takes two to (01:00:19) tango, right? And I think it's highly (01:00:22) unlikely. I think you're going to see in (01:00:24) your lifetime (01:00:25) um (01:00:27) hopefully understanding that we have to (01:00:29) coexist, but it's highly unlikely that (01:00:32) the two systems will become best buddies (01:00:34) for the obvious reasons. I think on this (01:00:36) this other question about what does what (01:00:40) are humans useful for uh in the very (01:00:43) long term um it's very very clear that (01:00:46) humans are social animals and want to be (01:00:48) around other humans. It's also very (01:00:51) clear that there are some things that we (01:00:53) need medical care and so forth which (01:00:54) will be delivered by humans. I think (01:00:57) that the it's a rough way to say this is (01:01:00) most other functions will be doable by (01:01:04) computers. (01:01:06) Will we allow them? That's a question (01:01:08) for your research, right? Is where is (01:01:11) that line? My my old example was the (01:01:14) following. This was back when we were (01:01:17) building Whimo at Google. So the thought (01:01:19) experiment is New York City is nothing (01:01:21) but Whimo and competitor automobile cars (01:01:23) and the engineers at Google and the (01:01:25) other companies have figured out how to (01:01:26) completely optimize the traffic so that (01:01:29) there the streets which don't change. (01:01:31) You have the absolutely mathematically (01:01:33) true highest loading per street ever and (01:01:35) it's seamless. And then you have (01:01:37) somebody with an emergency, a pregnant (01:01:39) woman, you know, whatever who has to get (01:01:40) to the hospital and needs an exception. (01:01:43) Is there an exception button in the car (01:01:46) which says you have to violate all the (01:01:48) rules because there's something now if (01:01:50) there is not an exception button in the (01:01:52) car then that's dominance of humans by a (01:01:56) computer and humans will revolt if I go (01:01:58) back to the pickaxe versus religion (01:02:00) humans will fight that that will be seen (01:02:03) as oppression by the government in this (01:02:06) case in the governor New York City (01:02:09) whatever metaphor you want and the (01:02:11) computer companies themselves If the (01:02:13) system on the other hand is adaptable to (01:02:16) human needs and says, "Ah, we have an (01:02:18) actual, you know, medical emergency and (01:02:20) not somebody who's high on drugs or, you (01:02:22) know, whatever who's just goofing around (01:02:24) or some kid who's, you know, playing (01:02:25) with the buttons." Um, and it reasons (01:02:28) and it says, "Oh my god, I've got to do (01:02:30) everything I can to get the the person (01:02:32) to the hospital." Then it'll be more (01:02:33) accepted. So, a lot of this depends on (01:02:36) whether it integrates with our human (01:02:37) experience and needs. Does it limit our (01:02:40) freedom or does it increase freedom? (01:02:43) I've come to the view, I'm old enough (01:02:45) now to believe that preserving our (01:02:47) freedom, our freedom of thought, of (01:02:48) motion, of assembly, of gathering, you (01:02:51) know, if all of the things right u are (01:02:54) really really important. If they impinge (01:02:56) on our freedom, then they will be fought (01:02:58) and I will be leading that fight. (01:03:01) >> Please. (01:03:04) >> Namaste sir. My name is Ilma Rose. I'm (01:03:06) from India. You mentioned India in your (01:03:09) conversation. (01:03:10) >> I'm lucky to be Professor Ellison's a (01:03:12) student and to be learning from him. My (01:03:14) question is how we can bring together US (01:03:18) and India. India has got some excellent (01:03:20) talented people. How we can bring both (01:03:22) our countries together in order uh to (01:03:26) create a world where democracies (01:03:28) flourish and it's a win-win partnership (01:03:31) for both our countries. (01:03:33) >> Um I strongly agree. I've spent lots of (01:03:35) time in India and the literally because (01:03:38) of the IITs and the quality of the (01:03:41) people the um the depth of talent in (01:03:44) India is extraordinary. However, the (01:03:46) depth of computing is not um last year (01:03:49) we did a calculation there were only (01:03:51) about a thousand GPUs in the entire (01:03:53) country for 1 billion people. So I and (01:03:55) others have organized to try to to fix (01:03:58) that. Um so I was alarmed with the most (01:04:02) recent trade war uh which I think has (01:04:04) set uh India and the United States back (01:04:07) and that needs to get fixed. I don't (01:04:09) understand the trade war and I just my (01:04:11) position is that's hurting us. India is (01:04:14) our natural partner right it's a (01:04:16) democracy it's a messy democracy but we (01:04:19) are too um in in Silicon Valley most of (01:04:23) the people I work with are of um (01:04:26) essentially South Asian origin with (01:04:28) India or some of the other countries so (01:04:31) at least in Silicon Valley the Indians (01:04:33) are flourishing and I want the tightest (01:04:35) possible integration with them (01:04:37) >> thank you (01:04:38) >> please (01:04:40) my name is Josh I'm an MPP2 here the (01:04:43) Kennedy school. Eric, you outlined two (01:04:45) national strategies earlier. There was (01:04:47) pushing the frontier or pursuing (01:04:49) adoption, but ideally you have both. So, (01:04:53) thinking about adoption in the US, what (01:04:55) are some of the major barriers to (01:04:57) enterprise adoption right now? Does that (01:05:00) vary by industry and what should (01:05:01) government do to help address those (01:05:03) barriers if anything? (01:05:04) >> Um, the government usually doesn't do a (01:05:06) very good job on that. Um the industry (01:05:10) believes that there is something called (01:05:12) a technology overhang that we your (01:05:15) friendly industry have produced more (01:05:17) tools than you are using. Now that that (01:05:20) is our belief whether it's true or not (01:05:22) you can discuss um and a lot of it seems (01:05:25) to be readiness to adopt the technology (01:05:29) the fact that a lot of the stuff is (01:05:30) software most companies don't have very (01:05:32) good software people I'm sorry to say (01:05:33) that um the the changes internally that (01:05:36) are required um I personally believe (01:05:39) that this technology adoption thing is (01:05:41) just a temporary problem and that as new (01:05:44) CEOs come in and winners emerge the (01:05:47) competitive pressure which are very high (01:05:49) in America will cause that adoption. Um, (01:05:52) you don't see very much adoption in (01:05:54) regulated industries because regulation (01:05:56) is used as an excuse to not innovate. (01:05:59) But at least in the hard innovative (01:06:01) industries, I think this will get (01:06:02) solved. But the general answer is I (01:06:05) would not have the government do very (01:06:06) much because I don't think it would (01:06:07) help. Um, what would help is just having (01:06:10) every business understand that if they (01:06:12) want to make money, which is the (01:06:13) capitalist vision, they need to use AI (01:06:15) much more profoundly. Remember, you can (01:06:18) target your customers, you can serve (01:06:19) them, you can understand them, you can (01:06:21) talk to them, and so forth, all with AI. (01:06:23) Um, and a lot of this, there's a lot of (01:06:26) negative issues with this. For example, (01:06:27) it may lead to layoffs in companies. (01:06:29) There's lots of examples where low-end (01:06:31) jobs are being replaced by computers, (01:06:33) and that's obviously a job loss. That's (01:06:35) a societal problem. But from an adoption (01:06:37) perspective, the answer is just time. (01:06:40) >> Please. (01:06:41) >> Hi, my name is Aush. I'm a third-year (01:06:43) PhD student in computer science doing AI (01:06:45) research. Um my question is regarding um (01:06:48) you were mentioning China's sort of (01:06:50) intense focus on automating AI and (01:06:52) business. I guess in the valley it seems (01:06:54) like you know every day there's a new (01:06:56) startup that comes around like it's the (01:06:57) big seed targeting some automation for (01:07:00) workflow and we we also sort of have the (01:07:02) the nine to 9 to6 fad. Um what would you (01:07:05) identify as like a key distinguisher in (01:07:07) China's AI um automating business that (01:07:11) you think the US needs to catch up on or (01:07:13) at least in the the Silicon Valley (01:07:14) startups nowadays? (01:07:16) >> What is your PhD topic? Uh (01:07:17) >> I work on reasoning for large language (01:07:19) models. (01:07:20) >> Perfect. (01:07:21) Right. So what is your time frame for (01:07:23) AGI? (01:07:24) >> Uh I'd give it like six to seven years. (01:07:27) >> Yeah. You see, not San Francisco (01:07:28) consensus, East Coast consensus. (01:07:30) >> Good. (01:07:30) >> We will see. You may [clears throat] be (01:07:32) right. Um, (01:07:35) I think that the (01:07:38) it has to do, I think, with the grandeur (01:07:41) of the dreams. When I'm in China, I (01:07:44) don't hear the same rhetoric as I do in (01:07:47) California. In California, it's two (01:07:51) years, the world will change, no one is (01:07:54) ready, this is humming so fast, and so (01:07:56) forth and so on. It's this sort of (01:07:57) rhetoric that drives and it's (01:07:58) self-reroducing. It's a belief system. (01:08:00) It's like a religion. Now, it always (01:08:02) takes longer than than the dreamers say. (01:08:06) I don't hear that in China. (01:08:10) It's it's it's noticeably different. So, (01:08:12) for example, with reasoning, you know (01:08:13) what Deepseek did with R3? They did a (01:08:16) fantastic job. They invented a new way (01:08:17) of doing supervised fine-tuning. I mean, (01:08:19) really, really clever stuff. You just (01:08:21) don't hear it at a national scale. And (01:08:23) by the way, Deepseek is a national (01:08:25) champion for China, right? They (01:08:26) literally now are, you know, they're (01:08:28) they're on the list of famous companies. (01:08:30) They're getting a enormous amount of (01:08:31) money. Uh last time I met with them, (01:08:33) they said, "We've solved our hardware (01:08:35) problem," which was code for the (01:08:36) government is just going to give them a (01:08:37) lot of chips, right? Welcome to a (01:08:40) communist country. (01:08:44) >> Okay, so this gent lady here, please. (01:08:47) >> Hi, thank you so much for being here. (01:08:48) I'm Sonia. I'm a joint degree student (01:08:50) with the MBA and uh and the policy (01:08:53) school. Um, this might be a slightly (01:08:55) more out there question and and (01:08:57) open-ended, but you talk a lot about (01:08:59) what it means to be human. And I think a (01:09:02) lot of that is consciousness. Um, and (01:09:05) I'm curious when I think about theories (01:09:07) of consciousness, which obviously is (01:09:08) very illdefined, AI is hitting more and (01:09:11) more of those. And so it it like if (01:09:13) anything, we need a revolution in how we (01:09:14) think about consciousness. But I'm (01:09:16) curious how you think about (01:09:18) consciousness in AI. And if you do think (01:09:22) that there's a possibility that AI is or (01:09:24) will become conscience, what that looks (01:09:25) like, what model welfare looks like in (01:09:27) that case. (01:09:28) >> Well, let me ask let me ask you a simple (01:09:29) question. So we have I'm going to give (01:09:32) Graham credit for being conscious (01:09:35) >> and in return he appears to be conscious (01:09:37) and in return Graham is going to give it (01:09:39) to me. So for purposes of argument, the (01:09:41) computer is on the table. Yes. (01:09:43) >> And we ask the computer, are you (01:09:44) conscious? And it says yes. And so you (01:09:48) being the smart graduate student here at (01:09:50) Harvard um come up with a series of (01:09:53) questions. It answers every question (01:09:56) correctly. How do we know? How could you (01:09:59) know that the computer is conscious? How (01:10:01) would you understand its internal (01:10:03) reasoning state? Now you could (01:10:06) instrument the way its reasoning works, (01:10:08) which people are now doing. And what (01:10:09) they're doing is they're watching what (01:10:10) are called super nodes within the u (01:10:13) within the weight structure to watch how (01:10:15) it's actually making its decision. So (01:10:17) maybe you could discover consciousness (01:10:19) by inspection, but that's speculation. (01:10:22) So I sat down with a bunch of (01:10:24) neuroscientists because I didn't know (01:10:25) the answer to this question. I said, (01:10:26) "How does it how does consciousness (01:10:28) evolve?" And their theory, which is just (01:10:31) a theory, was that consciousness evolves (01:10:35) when you have dissimilar systems that (01:10:37) are working together and growing and (01:10:39) they develop an awareness of the other, (01:10:42) right? And that human consciousness (01:10:45) evolved because we needed to understand (01:10:47) that we were something, right? That the (01:10:50) the development of the id, the identity (01:10:52) and so forth and so on was necessary in (01:10:54) order to command the system. Now, (01:10:56) there's no way to prove that. So the (01:10:59) answer is I don't know but this is a (01:11:01) good qu another it's a fourth question (01:11:02) to ask this audience to work on and and (01:11:05) there's two questions. The first how (01:11:07) does it work and the second is how would (01:11:09) you verify it? (01:11:10) >> Thank you. (01:11:11) >> Well I have the unfortunate uh uh (01:11:14) responsibility to say that we've come to (01:11:17) the witching hour. We dare (01:11:18) >> Graham. This is so much fun. I love (01:11:20) this. (01:11:20) >> Well you want to stay for five more (01:11:22) minutes. (01:11:22) >> Five more minutes. This is I love these. (01:11:24) We'll stay forever. I'm sorry. I just I (01:11:26) was pro I I promised your I promised (01:11:29) your schedulers that I would get you out (01:11:30) of ahead. (01:11:31) >> So please (01:11:32) >> short questions and short answers. (01:11:34) >> Sure. (01:11:34) >> Uh thanks Eric DH here from Harvard (01:11:36) Business School, a second year student (01:11:38) there. Also have a background in (01:11:39) international relations before my MBA. (01:11:41) Um you argue convincingly about the need (01:11:43) for an IE IA or international atomic (01:11:46) energy agency equivalent for AI. So my (01:11:48) question is how does the state (01:11:50) department how does the Pentagon and (01:11:51) similar agencies need to configure (01:11:53) themselves to prepare for this the (01:11:55) coming era of super intelligence and my (01:11:58) second slightly risky question is um if (01:12:02) you are looking for someone to help you (01:12:03) this summer or this winter to work on (01:12:04) this topic I'd be very happy to (01:12:06) [laughter] (01:12:06) >> I love people with courage um so the IEA (01:12:11) question so there's a group of people (01:12:12) who are very close to me who concluded (01:12:15) that the only solution to the problems (01:12:17) that I'm answering (01:12:18) is to create the equivalent of CERN. (01:12:22) And the idea was that we would get (01:12:24) together, all of us, including China, (01:12:26) you know, everybody, and we put it all (01:12:28) in here, take all the best minds, and (01:12:30) we'd all work on this. And um as a (01:12:34) result, we would (01:12:37) build this huge future because the (01:12:38) benefits to humanity of this are so (01:12:40) great. Start thinking about every (01:12:43) disease eliminated, solutions to real (01:12:45) hard problems. the we have Megan here. (01:12:47) Energy issues, you know, she's the (01:12:49) expert. All of this kind of stuff. (01:12:52) I think it's pretty you probably say (01:12:55) it's probably pretty unlikely that's (01:12:56) going to happen. So then the next group (01:12:59) says, well then we need an an IEA which (01:13:02) is essentially a mandatory inspection (01:13:05) group for nuclear. And what into that (01:13:08) scenario what happens is you'd have to (01:13:11) go and this group would visit the data (01:13:14) centers and the algorithms of every (01:13:16) company in every country. Now that (01:13:19) occurred remember the Vienna group (01:13:22) occurred after 15 years of negotiation (01:13:25) including people like Henry and after (01:13:28) two nuclear bombs were dropped on (01:13:29) Hiroshima and Nagasaki which we can all (01:13:31) agree was hor horrific. We've not had (01:13:34) such an example. There are people that I (01:13:37) know who say the following and they (01:13:38) don't say this with evil intent. They (01:13:41) say that we won't we will have an event (01:13:45) that will force this IEA thing and we (01:13:49) hope that it is a Chernobyl level event. (01:13:52) And I said what does that mean? And they (01:13:54) said not that many people died. Okay. As (01:13:58) opposed to a nuclear attack. So there's (01:14:01) a a strain of thinking in my world that (01:14:03) is consistent with that. But I think uh (01:14:06) first I think nobody knows and it's (01:14:08) unlikely that the countries will um put (01:14:13) up with that uh until there's a real (01:14:16) crisis. I used to say that in (01:14:21) I don't know 20 years after the climate (01:14:25) is being destroyed and you know the (01:14:28) earth is melting and sea level rise (01:14:30) there's going to be a m there's going to (01:14:32) there's going to be a meeting of all the (01:14:33) powers where people will say well we (01:14:35) really screwed this up we now have to (01:14:37) fix it. So I think there is precedent (01:14:39) for a global challenge and a resolution (01:14:43) but it's very messy. It's not done (01:14:46) rationally. It's not like you and I (01:14:47) agree. There has to be some crisis and (01:14:49) some political dynamic and it can be (01:14:51) quite severe. This (01:14:52) >> is an opportunity for the next (01:14:54) generation. Just do this right quick. Do (01:14:56) I take the three questions? Just do your (01:14:58) short questions. We'll let Eric wrap up, (01:15:00) please. Yes. (01:15:02) >> Hi. Uh, thank you for your extra time. (01:15:03) Uh, my name is Kevin. I'm a second year (01:15:05) at the law school. Uh, my question is (01:15:07) about automation and jobs. Um, a couple (01:15:09) months ago, Sam Alman went on an (01:15:11) interview saying that insinuating that (01:15:13) if something could be automated, maybe (01:15:15) it wasn't a real job to begin with. Uh, (01:15:17) I want to get your thoughts on that (01:15:18) statement and if you agree with the (01:15:21) premise that if something can be (01:15:22) automated, should it? Um, and relatedly, (01:15:25) if you disagree, what are ways that we (01:15:27) as a society can decide what we should (01:15:28) and shouldn't automate? (01:15:29) >> Sure. Please fit. (01:15:31) >> Um, good night. My name is Kissia. I'm a (01:15:33) middle mid-career in PA and a former (01:15:35) Googler. So, I'm very happy to see you. (01:15:38) Um my question is regarding algorithm (01:15:40) diplomacy like Dr. Kissinger used to say (01:15:43) that constructive um ambiguity and the (01:15:46) human pause is really necessary but when (01:15:48) AI is binary are we getting to a towards (01:15:51) a world where the escalation is (01:15:53) computationally impossible and does this (01:15:55) push for a new algorithm diplomacy that (01:15:57) we could push harder. (01:16:00) >> Uh thank you. My name is Kanesk. I'm (01:16:02) from Harvard Business School. Uh my (01:16:04) question is about ethical super (01:16:05) intelligence. Uh first, do why do you (01:16:07) not believe that if America develops (01:16:09) super intelligence, other nations will (01:16:11) follow and then be able to copy it like (01:16:12) China did with DeepSeek? And then second (01:16:14) of all, if we're trying to embody it (01:16:16) with ethics, then if someone else comes (01:16:18) up with an unethical super intelligence, (01:16:19) isn't that going to be strictly more (01:16:21) free and more capable than a constrained (01:16:24) super intelligence model that America (01:16:25) tries to develop? (01:16:27) >> Sorry, one word about your question (01:16:28) again. (01:16:29) >> Uh yeah, I wanted to ask what your (01:16:31) thoughts were on Sam Alman's quote. (01:16:33) >> Thank you. (01:16:35) So um one of the things that people at (01:16:39) this level forget is that human dignity (01:16:42) involves purpose and an awful lot of (01:16:45) jobs provide purpose to an awful lot of (01:16:47) people and the loss of those jobs is a (01:16:49) major crisis not financially but (01:16:52) emotionally in terms of their meaning. (01:16:54) So in order for us to get through this (01:16:56) we're going to have to address that (01:16:58) right? we're going to have to actually (01:17:00) do the right thing and the right thing (01:17:03) will be some combination of better tools (01:17:06) and so forth. I'm not as worried about (01:17:08) the jobs issue because we're producing (01:17:10) fewer humans which I view as a major (01:17:12) crisis. Um, collectively you all are (01:17:14) having fewer children than my generation (01:17:17) which is having fewer children than my (01:17:19) parents generation and so forth. And we (01:17:21) need more humans. um partly I'm a (01:17:24) businessman so you need more customers (01:17:26) but the important point is we need more (01:17:27) humans and if we have less humans then (01:17:29) there's going to be open jobs and no (01:17:32) people to fill them AI can help take (01:17:34) people who are not ready for a job and (01:17:36) get them one word from yours aside from (01:17:38) being a future Googler (01:17:40) >> no the algorithm diplomacy (01:17:43) >> and computational so in your premise of (01:17:46) your question you describe the (01:17:48) algorithms as binary and so if you think (01:17:52) about Kissinger in 1971 when he came up (01:17:54) with strategic ambiguity. (01:17:57) Do you think that a computer in one or (01:18:00) two years will be able to invent (01:18:02) strategic diplomacy, ambiguity given (01:18:05) that it's already been done before? I (01:18:07) think the answer is yes. So I think that (01:18:09) that um the algorithms are getting so (01:18:12) intelligent that as long as it's a (01:18:14) concept that's been around in the past, (01:18:16) it can probably figure out a way to (01:18:18) apply it. So I don't agree with you that (01:18:20) it's as binary as your question being (01:18:23) and your question was (01:18:24) >> uh super intelligence (01:18:25) >> super intelligence um (01:18:31) I I think it's the same answers right (01:18:35) super intelligence in one form or (01:18:37) another is going to apply my own view (01:18:40) which is maybe different from the (01:18:41) consensus is that we're going to develop (01:18:44) brilliant AI physicists brilliant AI (01:18:47) biologists brilliant AI I chemists, (01:18:49) brilliant AI writers, brilliant AI (01:18:51) historians. Um, but that that the (01:18:55) concept and and there'll be systems that (01:18:57) can drive them. It's not at all obvious (01:19:01) that if you were Einstein in 1902 with (01:19:05) that amount of math available to you (01:19:07) that you would have had the brilliance (01:19:08) to invent um special relativity (01:19:13) using the algorithms of today. In the (01:19:16) industry there is a view that that's the (01:19:19) next really hard problem. There are (01:19:21) various theories about it. Uh one answer (01:19:23) is you could just do repetition. You (01:19:25) could just keep asking questions you (01:19:27) know the monkey of the keyboard kind of (01:19:28) thing and eventually you do it. Another (01:19:31) way is that you could give the (01:19:32) optimization function to be curiosity (01:19:34) and if you just wait long enough it'll (01:19:36) discover special relativity. But that's (01:19:38) not how Einstein did it. Einstein sat (01:19:40) with his little pen and paper at the age (01:19:42) of 18 or 17 with a lamp and figured it (01:19:45) out. We're not there yet. My own view is (01:19:48) that's going to be a very hard boundary. (01:19:52) So in other words, we're going to get to (01:19:54) superhuman behavior, which is not the (01:19:56) same thing as super intelligence, which (01:19:59) I would say Einstein and Kissinger and (01:20:00) so forth really were. (01:20:02) >> Graham, it's such a privilege to be here (01:20:04) with (01:20:04) >> such a pleasure to have you here and we (01:20:07) look forward to having you back. Let's (01:20:08) say thank you [applause] very thank you (01:20:18) >> [applause] (01:20:24) >> was