1 00:00:00,400 --> 00:00:05,597 What does super intelligence mean and when is it likely to be here? Eric, 2 00:00:05,680 --> 00:00:07,437 we've talked about this. What are your thoughts? 3 00:00:07,520 --> 00:00:11,197 >> Super intelligence is defined as 4 00:00:11,280 --> 00:00:14,557 >> AI's ability to know from chemistry to 5 00:00:14,640 --> 00:00:18,797 biology to to sports. It's already super 6 00:00:18,880 --> 00:00:22,157 to human in many ways. But the 7 00:00:22,240 --> 00:00:25,357 collaboration between humans and AI will 8 00:00:25,440 --> 00:00:30,157 be the most productive and fruitful way of of doing things. 9 00:00:30,240 --> 00:00:34,637 >> All of the evidence and fee said this very well is going to be human and 10 00:00:34,720 --> 00:00:39,197 computer interaction. I think that to get to real super intelligence, we 11 00:00:39,280 --> 00:00:40,717 probably need another algorithmic 12 00:00:40,800 --> 00:00:43,677 breakthrough. It's so important as we 13 00:00:43,760 --> 00:00:47,277 talk about AGI at ASI that the most 14 00:00:47,360 --> 00:00:52,077 important thing that we keep in mind is 15 00:00:52,160 --> 00:00:56,760 >> now that's a moonshot ladies and gentlemen. 16 00:00:57,200 --> 00:00:58,957 >> Everybody welcome to Eric Schmidt week. 17 00:00:59,040 --> 00:01:01,837 Uh two episodes this week both with Eric 18 00:01:01,920 --> 00:01:03,357 talking about some of the most important 19 00:01:03,440 --> 00:01:06,557 developments. The first is with me and 20 00:01:06,640 --> 00:01:10,317 Eric and Fay Lee uh who's building world 21 00:01:10,400 --> 00:01:12,077 models. This is a conversation we had 22 00:01:12,160 --> 00:01:15,597 live at FI in Saudi Arabia talking about 23 00:01:15,680 --> 00:01:20,477 digital super intelligence. Uh this is rare because FFE doesn't travel. It was 24 00:01:20,560 --> 00:01:22,397 her first trip to Saudi and it was an 25 00:01:22,480 --> 00:01:25,037 extraordinary conversation. The second 26 00:01:25,120 --> 00:01:27,837 episode is between my moonshot mate Dave 27 00:01:27,920 --> 00:01:29,677 Blondon and Eric Schmidt talking about 28 00:01:29,760 --> 00:01:33,117 US versus China and how to avoid, you 29 00:01:33,200 --> 00:01:37,277 know, critical breakdowns and what might be catastrophes in the future due to 30 00:01:37,360 --> 00:01:39,997 accelerating AI. All right, get ready 31 00:01:40,080 --> 00:01:45,197 for two incredible episodes. The first one, myself, Eric, and Faith Ae on 32 00:01:45,280 --> 00:01:50,520 digital super intelligence. All right, let's jump in. 33 00:01:52,880 --> 00:01:54,477 We're going to be discussing super 34 00:01:54,560 --> 00:01:57,037 intelligence. What does that mean and 35 00:01:57,120 --> 00:02:00,237 what happens when it arrives? Uh we've 36 00:02:00,320 --> 00:02:05,197 been talking about AI, AGI, now perhaps 37 00:02:05,280 --> 00:02:07,597 digital super intelligence or ASI. I 38 00:02:07,680 --> 00:02:10,237 want to start with the obvious question 39 00:02:10,320 --> 00:02:12,717 and it's one that I don't think anybody 40 00:02:12,800 --> 00:02:15,757 has a perfect answer for, but what does 41 00:02:15,840 --> 00:02:20,317 super intelligence mean and when is it likely to be here? 42 00:02:20,400 --> 00:02:22,957 Eric, we've talked about this. What are your thoughts? 43 00:02:23,040 --> 00:02:26,717 >> So, a simple Thank you, Peter, and thanks for everybody for being here and 44 00:02:26,800 --> 00:02:29,277 obviously thanks to FFE, our very close 45 00:02:29,360 --> 00:02:32,717 colleague. Um, the general accepted 46 00:02:32,800 --> 00:02:37,837 definitions of general intelligence is its human level of intelligence, AGI. 47 00:02:37,920 --> 00:02:40,637 And human intelligence you can understand because we're all human. You 48 00:02:40,720 --> 00:02:44,557 have ideas, you have friends, you have, you know, you think about things, you're 49 00:02:44,640 --> 00:02:49,197 creative. Super intelligence is defined as the 50 00:02:49,280 --> 00:02:51,517 intelligence equal to the sum of 51 00:02:51,600 --> 00:02:55,357 everyone, right? Or even better than all 52 00:02:55,440 --> 00:02:58,957 humans. And there is a belief in our 53 00:02:59,040 --> 00:03:00,957 industry that we will get to super 54 00:03:01,040 --> 00:03:03,917 intelligence. We don't know exactly how 55 00:03:04,000 --> 00:03:08,717 long. Uh there's a group of people who I call the San Francisco consensus because 56 00:03:08,800 --> 00:03:11,757 it's all they're all living in San Francisco. Maybe it's the weather or the 57 00:03:11,840 --> 00:03:15,437 drugs or something, but they all basically think that it's within 3 to 58 00:03:15,520 --> 00:03:19,997 four years. Uh I personally think it'll be longer than that. But fundamentally, 59 00:03:20,080 --> 00:03:23,037 their argument is that there are compounding 60 00:03:23,120 --> 00:03:27,997 uh effects that we're seeing now which will race us to this much faster than 61 00:03:28,080 --> 00:03:31,837 people think. >> And Fay, I don't think anybody's 62 00:03:31,920 --> 00:03:37,277 expected the performance that AI has given us so far. The scaling laws have 63 00:03:37,360 --> 00:03:40,637 given us capabilities that are extraordinary. you know, you're the CEO 64 00:03:40,720 --> 00:03:43,437 of a a new company, the founder of World 65 00:03:43,520 --> 00:03:48,157 Labs. You've been at Stanford working on this. Uh how do you think about super 66 00:03:48,240 --> 00:03:51,837 intelligence? Do you discuss super intelligence at all in your work? 67 00:03:51,920 --> 00:03:53,757 >> Yeah, that's a great question, Peter. 68 00:03:53,840 --> 00:03:57,357 And uh um you know, when Alan Turing 69 00:03:57,440 --> 00:04:00,317 dared humanity with the question of can 70 00:04:00,400 --> 00:04:03,837 we create thinking machines, um he was 71 00:04:03,920 --> 00:04:06,797 thinking about the fundamental question 72 00:04:06,880 --> 00:04:10,237 of intelligence. So the birth of AI is 73 00:04:10,320 --> 00:04:13,677 about intelligence is about the the the 74 00:04:13,760 --> 00:04:16,717 profound general ability of what 75 00:04:16,800 --> 00:04:19,037 intelligence means. So from that point 76 00:04:19,120 --> 00:04:22,877 of view AI is already born as a field 77 00:04:22,960 --> 00:04:25,597 that tries to push the boundary of what 78 00:04:25,680 --> 00:04:29,197 intelligence mean. Now in fast forward 79 00:04:29,280 --> 00:04:32,957 to 75 years after Allan Touring this 80 00:04:33,040 --> 00:04:36,077 phrase uh uh super intelligence is 81 00:04:36,160 --> 00:04:38,557 pretty hot in Silicon Valley and I do 82 00:04:38,640 --> 00:04:40,877 agree with Eric that the colloquial 83 00:04:40,960 --> 00:04:45,117 definition is um what is uh is the 84 00:04:45,200 --> 00:04:48,397 capability of AI and computers that's 85 00:04:48,480 --> 00:04:51,917 better than any human. But I do think we 86 00:04:52,000 --> 00:04:53,997 need to be a little careful. First of 87 00:04:54,080 --> 00:04:57,437 all, some part of today's AI is already 88 00:04:57,520 --> 00:05:01,437 better than any human. For example, um 89 00:05:01,520 --> 00:05:04,717 AI's ability of speaking many different 90 00:05:04,800 --> 00:05:07,837 languages, translating between, you 91 00:05:07,920 --> 00:05:11,997 know, dozens and dozens of language. 92 00:05:12,080 --> 00:05:14,957 Pretty much no human can do that. Or 93 00:05:15,040 --> 00:05:17,997 AI's ability to calculate things really 94 00:05:18,080 --> 00:05:21,277 fast. AI's ability to know from 95 00:05:21,360 --> 00:05:24,397 chemistry to biology to to sports, you 96 00:05:24,480 --> 00:05:27,357 know, the vast amount of uh knowledge. 97 00:05:27,440 --> 00:05:31,837 So, it's already super to human in many 98 00:05:31,920 --> 00:05:34,637 ways. But 99 00:05:34,720 --> 00:05:39,517 it remains a question that um can AI 100 00:05:39,600 --> 00:05:41,917 ever be Newton? 101 00:05:42,000 --> 00:05:47,597 Can AI ever be Einstein? Can AI ever be Picasso? 102 00:05:47,680 --> 00:05:48,077 M 103 00:05:48,160 --> 00:05:51,997 >> I actually don't know. For example, 104 00:05:52,080 --> 00:05:55,197 we have all the celestial data of the 105 00:05:55,280 --> 00:05:58,797 movement of the stars that we observe 106 00:05:58,880 --> 00:06:02,797 today. Give that data to any AI 107 00:06:02,880 --> 00:06:06,477 algorithm. It will not be able to deduce 108 00:06:06,560 --> 00:06:10,797 Newtonian law of motion. 109 00:06:10,880 --> 00:06:13,677 that ability that humans have. It's the 110 00:06:13,760 --> 00:06:18,157 combination of creativity, abstraction. 111 00:06:18,240 --> 00:06:22,317 I do not see today's AI or tomorrow's AI 112 00:06:22,400 --> 00:06:24,157 being able to do that yet. 113 00:06:24,240 --> 00:06:27,197 >> So, so one of the common examples that 114 00:06:27,280 --> 00:06:30,637 and Fei of course got it right is to 115 00:06:30,720 --> 00:06:34,797 think about if you had all of the knowledge in a computer that existed in 116 00:06:34,880 --> 00:06:36,317 1902. Yes. 117 00:06:36,400 --> 00:06:39,597 >> Could you invent relativity? uh 118 00:06:39,680 --> 00:06:41,837 basically the physics of today and the 119 00:06:41,920 --> 00:06:45,117 answer today is no. Um so for example if 120 00:06:45,200 --> 00:06:46,797 you look at what is called test time 121 00:06:46,880 --> 00:06:49,197 compute where the systems are doing 122 00:06:49,280 --> 00:06:52,397 reasoning they can't take the reasoning 123 00:06:52,480 --> 00:06:57,037 that they learned and feed it back into themselves very quickly. Whereas if 124 00:06:57,120 --> 00:07:00,957 you're a mathematician you prove something you can base your next proof 125 00:07:01,040 --> 00:07:03,517 on that. It's hard for the systems today 126 00:07:03,600 --> 00:07:06,557 although there are appro approximations. 127 00:07:06,640 --> 00:07:08,957 So we're we're it's We don't know where 128 00:07:09,040 --> 00:07:10,717 the boundaries are. The example that I'd 129 00:07:10,800 --> 00:07:13,837 like to use is let's imagine that we can 130 00:07:13,920 --> 00:07:16,077 get computers that can solve everything 131 00:07:16,160 --> 00:07:19,037 that we normally can do as humans except 132 00:07:19,120 --> 00:07:22,397 for these amazing set of creativities. 133 00:07:22,480 --> 00:07:25,117 How do really creative people do it? The 134 00:07:25,200 --> 00:07:27,277 best examples are that they are experts 135 00:07:27,360 --> 00:07:30,077 in one area, they see another area and 136 00:07:30,160 --> 00:07:32,237 they have an intuition that the same 137 00:07:32,320 --> 00:07:37,437 mechanism will solve a problem of a completely different area there. That's 138 00:07:37,520 --> 00:07:39,117 an example of something we have to learn 139 00:07:39,200 --> 00:07:42,237 how to do with AI. An alternative would 140 00:07:42,320 --> 00:07:45,197 be to simply do it uh in brute force 141 00:07:45,280 --> 00:07:49,357 using reinforcement learning. The problem is that combinatorily the cost 142 00:07:49,440 --> 00:07:53,837 of that is insane and we're already running out of electricity and so forth. 143 00:07:53,920 --> 00:07:58,157 So I think that to get to real super intelligence, we probably need another 144 00:07:58,240 --> 00:08:00,317 algorithmic breakthrough. >> We need another what? 145 00:08:00,400 --> 00:08:05,437 >> Algorithmic breakthrough. Another way of dealing with this. So the technical the 146 00:08:05,520 --> 00:08:08,957 technical term is called non-stationerity of objectives. What's 147 00:08:09,040 --> 00:08:13,117 happening is the systems are trained against objectives. But to do this kind 148 00:08:13,200 --> 00:08:16,877 of creativity that FE is talking about, you need to be able to change the 149 00:08:16,960 --> 00:08:18,637 objectives as you're doing them. 150 00:08:18,720 --> 00:08:22,397 >> We've seen this past year, I think GPT 151 00:08:22,480 --> 00:08:27,117 5 Pro reach an IQ of like 148, which is 152 00:08:27,200 --> 00:08:28,477 extraordinary. And of course, there is 153 00:08:28,560 --> 00:08:31,757 no ceiling on on this. I mean, it it 154 00:08:31,840 --> 00:08:34,557 loses meaning at some point, but the 155 00:08:34,640 --> 00:08:37,037 ability for every human on the planet to 156 00:08:37,120 --> 00:08:39,437 have an Einstein level, not in the 157 00:08:39,520 --> 00:08:41,757 creativity side, but intelligence side 158 00:08:41,840 --> 00:08:45,997 in their pocket changes the game for 8 159 00:08:46,080 --> 00:08:48,717 billion humans. And now with Starlink 160 00:08:48,800 --> 00:08:53,677 and with, you know, $50 smartphones, uh, it's possible that every single person 161 00:08:53,760 --> 00:08:58,397 on the planet has this kind of capability. Add to that humanoid robots. 162 00:08:58,480 --> 00:09:00,237 Add to that, you know, a whole slew of 163 00:09:00,320 --> 00:09:02,797 other exponential technologies. And the 164 00:09:02,880 --> 00:09:05,437 commentary is we're heading towards a a 165 00:09:05,520 --> 00:09:08,157 post scarcity society, 166 00:09:08,240 --> 00:09:10,877 right? Do you believe in that vision? 167 00:09:10,960 --> 00:09:13,037 Feet, 168 00:09:13,120 --> 00:09:14,957 >> I do think we have to be a little 169 00:09:15,040 --> 00:09:18,077 careful. I I know that we are combining 170 00:09:18,160 --> 00:09:20,877 some of the hottest words from Silicon 171 00:09:20,960 --> 00:09:23,837 Valley from uh AI, super intelligence, 172 00:09:23,920 --> 00:09:26,317 humanoid robots and all that. To be 173 00:09:26,400 --> 00:09:28,797 honest, I think robotics has a long way 174 00:09:28,880 --> 00:09:32,077 to go. I I think uh we have to be a 175 00:09:32,160 --> 00:09:34,237 little bit careful with the projection 176 00:09:34,320 --> 00:09:37,917 of robotics. I I think the the ability 177 00:09:38,000 --> 00:09:41,837 the dexterity of human level uh 178 00:09:41,920 --> 00:09:46,237 manipulation is um is um 179 00:09:46,320 --> 00:09:50,957 you know we we have to wait a lot longer 180 00:09:51,040 --> 00:09:54,157 to to get it. So, are we entering post 181 00:09:54,240 --> 00:09:58,840 uh scarcity? Um, 182 00:10:00,160 --> 00:10:04,077 I don't know. I I actually I'm not as uh 183 00:10:04,160 --> 00:10:07,917 bullish as a typical Silicon Valley uh 184 00:10:08,000 --> 00:10:11,437 person because I think we're entering I 185 00:10:11,520 --> 00:10:14,557 absolutely believe AI will be augmenting 186 00:10:14,640 --> 00:10:18,157 human uh capabilities in incredibly 187 00:10:18,240 --> 00:10:22,077 profound ways. But I think we we will 188 00:10:22,160 --> 00:10:25,277 continue to see that the collaboration 189 00:10:25,360 --> 00:10:28,877 between humans and AI will be the most 190 00:10:28,960 --> 00:10:32,717 productive and fruitful way of of doing things. 191 00:10:32,800 --> 00:10:35,837 >> So the projection is that AI is going to 192 00:10:35,920 --> 00:10:38,637 generate as much as 15 trillion in 193 00:10:38,720 --> 00:10:42,157 economic value by 2030. Uh the idea that 194 00:10:42,240 --> 00:10:46,557 shifting the foundation of national wealth from capital to labor to 195 00:10:46,640 --> 00:10:49,037 computational intelligence. So what's 196 00:10:49,120 --> 00:10:51,917 that implication, Eric, for the global 197 00:10:52,000 --> 00:10:56,077 economy? How are we going to see redistribution, 198 00:10:56,160 --> 00:10:59,757 if you would, of wealth or of capabilities? Are we going to see a 199 00:10:59,840 --> 00:11:03,757 leveling of the field between nation states, or are we going to see runaway 200 00:11:03,840 --> 00:11:07,837 winners? >> So in your abundance hypothesis, which 201 00:11:07,920 --> 00:11:10,637 we've talked a lot about, there may be a 202 00:11:10,720 --> 00:11:13,837 flaw in the argument because part of the 203 00:11:13,920 --> 00:11:18,077 abundance argument is that it's abundance for everyone. 204 00:11:18,160 --> 00:11:21,677 But there's plenty of evidence that these technologies have network effects 205 00:11:21,760 --> 00:11:25,757 which concentrates to a small number of winners. So you could for example 206 00:11:25,840 --> 00:11:30,637 imagine a small number of countries getting all those benefits in those 207 00:11:30,720 --> 00:11:34,237 countries. You could imagine a small number of firms and people getting those 208 00:11:34,320 --> 00:11:38,397 benefits. Those are public policy question. There's no question the wealth 209 00:11:38,480 --> 00:11:40,477 will be created because the wealth comes 210 00:11:40,560 --> 00:11:43,677 from efficiency. And every company that 211 00:11:43,760 --> 00:11:46,957 has implemented AI has seen huge gains. 212 00:11:47,040 --> 00:11:49,117 Think about here we are in Saudi Arabia. 213 00:11:49,200 --> 00:11:51,037 You have all of this oil distribution, 214 00:11:51,120 --> 00:11:53,357 all the oil networks, all the losses. AI 215 00:11:53,440 --> 00:11:56,317 can easily improve that by 10% 20%. 216 00:11:56,400 --> 00:11:58,557 Those are huge numbers for this country. 217 00:11:58,640 --> 00:12:00,877 If you look in biology and medicine and 218 00:12:00,960 --> 00:12:05,917 drug discovery, much faster drug approval cycles, much lower cost trials, 219 00:12:06,000 --> 00:12:09,997 look at materials, much more efficient and easier to build materials. the 220 00:12:10,080 --> 00:12:12,797 companies that adopt AI quickly get a 221 00:12:12,880 --> 00:12:15,197 disproportionate return. The question 222 00:12:15,280 --> 00:12:18,637 is, is are those gains uniform, which 223 00:12:18,720 --> 00:12:21,117 would be our hope, or in my view more 224 00:12:21,200 --> 00:12:25,917 likely largely centered around early adopters, network effects, well-run 225 00:12:26,000 --> 00:12:27,677 countries, and perhaps capital. 226 00:12:27,760 --> 00:12:31,197 >> But you could imagine still that we're going to see autonomous cars in which 227 00:12:31,280 --> 00:12:35,517 the ownership of a car is four times, let me put it the other way, uh being in 228 00:12:35,600 --> 00:12:39,997 an autonomous vehicle is four times cheaper than owning a car. We can see AI 229 00:12:40,080 --> 00:12:44,157 giving us the best physicians, the best health care for free in the same way 230 00:12:44,240 --> 00:12:48,077 that Google gave us access to information for free. We will see a 231 00:12:48,160 --> 00:12:50,797 massive demonetization in so much of our 232 00:12:50,880 --> 00:12:53,437 world. I think that will be available to 233 00:12:53,520 --> 00:12:56,877 anyone with a smartphone and a decent 234 00:12:56,960 --> 00:13:02,077 bandwidth connectivity. Is that still not what you think will 235 00:13:02,160 --> 00:13:05,597 happen? Do you think there's a reason something that would stop that level of 236 00:13:05,680 --> 00:13:08,477 distribution of of those services which 237 00:13:08,560 --> 00:13:10,477 we spend a lot of our money on today? 238 00:13:10,560 --> 00:13:13,197 >> I do think AI democratizes that. I 239 00:13:13,280 --> 00:13:15,357 totally agree with you. I think whether 240 00:13:15,440 --> 00:13:18,157 it's healthcare or transportation or 241 00:13:18,240 --> 00:13:21,997 knowledge AI will will uh democratize 242 00:13:22,080 --> 00:13:25,517 massively. But I agree with Eric that uh 243 00:13:25,600 --> 00:13:28,477 this increased global uh productivity 244 00:13:28,560 --> 00:13:31,197 does not necessarily translate to shared 245 00:13:31,280 --> 00:13:34,317 prosperity. Shared prosperity is a 246 00:13:34,400 --> 00:13:37,357 deeper social problem. It involves 247 00:13:37,440 --> 00:13:41,517 policy. It involves you know geopolit uh 248 00:13:41,600 --> 00:13:46,477 politics. It involves distribution and that's a different problem from the 249 00:13:46,560 --> 00:13:48,477 capability of the technology. So what's 250 00:13:48,560 --> 00:13:50,317 your advice to the country leaders that 251 00:13:50,400 --> 00:13:55,197 are here uh that are seeing ASI as a 252 00:13:55,280 --> 00:13:58,877 future for someone else and not for themselves? What should they be doing? I 253 00:13:58,960 --> 00:14:03,117 mean this is the crit the speed at which is deploying. They don't have a lot of 254 00:14:03,200 --> 00:14:06,797 time to make critical decisions. Well, it's it's worth describing where we are 255 00:14:06,880 --> 00:14:11,357 now in the United States because of the depth of our capital markets and because 256 00:14:11,440 --> 00:14:14,317 of the extraordinary chips that are available in the Taiwanese 257 00:14:14,400 --> 00:14:16,717 manufacturers, TSMC in particular, 258 00:14:16,800 --> 00:14:21,757 America has this huge lead in building these what are called hyperscalers. 259 00:14:21,840 --> 00:14:25,037 If there's going to be super intelligence, it's going to come from 260 00:14:25,120 --> 00:14:28,077 those efforts. That's a big deal. If 261 00:14:28,160 --> 00:14:30,637 there is super intelligence, imagine a 262 00:14:30,720 --> 00:14:32,637 company like Google inventing this, for 263 00:14:32,720 --> 00:14:35,757 example. I am obviously biased. Um, and 264 00:14:35,840 --> 00:14:40,797 what's the value of being able to solve every problem that humans can't solve? 265 00:14:40,880 --> 00:14:45,357 It's infinite. >> Sure. >> So, that's the goal, right? China is a 266 00:14:45,440 --> 00:14:49,597 second. Doesn't have the capital markets, doesn't have the chips, and the 267 00:14:49,680 --> 00:14:52,077 other countries are not anywhere near. 268 00:14:52,160 --> 00:14:56,477 Saudi has done a good job of partnering with America. Uh, and the hyperscalers 269 00:14:56,560 --> 00:14:58,717 will be located here and in the UAE. 270 00:14:58,800 --> 00:15:03,277 That's a good strategy. So that's a a good example of how you partner. You 271 00:15:03,360 --> 00:15:04,797 figure out which side you're on. 272 00:15:04,880 --> 00:15:06,397 Hopefully it's the United States. And 273 00:15:06,480 --> 00:15:09,437 you work with the US firms. I do think 274 00:15:09,520 --> 00:15:12,957 countries all should invest invest in 275 00:15:13,040 --> 00:15:15,917 their own human capital, invest in 276 00:15:16,000 --> 00:15:19,117 partnerships and and invest in its own 277 00:15:19,200 --> 00:15:22,477 uh technological stack as well as the 278 00:15:22,560 --> 00:15:25,437 business ecosystem. This is uh as Eric 279 00:15:25,520 --> 00:15:28,317 said it depends on the strength and uh 280 00:15:28,400 --> 00:15:30,637 particularity of the different countries 281 00:15:30,720 --> 00:15:34,317 but I think not investing in AI it would 282 00:15:34,400 --> 00:15:37,677 be macroscopically the wrong thing to do. 283 00:15:37,760 --> 00:15:40,317 >> So under the thesis that that investment 284 00:15:40,400 --> 00:15:45,117 involves building out data centers in your nation. 285 00:15:45,200 --> 00:15:46,957 Do you think every country should be 286 00:15:47,040 --> 00:15:49,517 building out a data center that it has 287 00:15:49,600 --> 00:15:53,117 sovereign AI running on? 288 00:15:53,200 --> 00:15:55,277 >> Every country is a very sweeping 289 00:15:55,360 --> 00:16:00,317 statement. I I I do think um it depends. 290 00:16:00,400 --> 00:16:03,997 It depends. I I think obviously for a 291 00:16:04,080 --> 00:16:06,557 region like this absolutely where you 292 00:16:06,640 --> 00:16:10,157 know oil uh energy is cheaper and uh and 293 00:16:10,240 --> 00:16:12,397 such an important region in the world. 294 00:16:12,480 --> 00:16:14,637 But if we're talking about smaller 295 00:16:14,720 --> 00:16:17,357 countries, I don't know if every single 296 00:16:17,440 --> 00:16:19,517 country can afford to build data 297 00:16:19,600 --> 00:16:22,957 centers. But there are other um other 298 00:16:23,040 --> 00:16:26,637 areas of investment, right? >> But let me give you an example. Let's 299 00:16:26,720 --> 00:16:27,997 pick Europe. It's easy to pick on 300 00:16:28,080 --> 00:16:31,517 Europe. Energy costs are high, right? 301 00:16:31,600 --> 00:16:35,117 Financing costs are not low. So the odds 302 00:16:35,200 --> 00:16:37,197 of Europe being able to build very large 303 00:16:37,280 --> 00:16:40,077 data centers is extremely low. But they 304 00:16:40,160 --> 00:16:44,717 can partner with countries where they can do it. Uh France for example did a 305 00:16:44,800 --> 00:16:48,717 partnership with Abu Dhabi. So there's an there are examples of that. So I 306 00:16:48,800 --> 00:16:52,157 think if you take a global view and you figure out who your partners are, you 307 00:16:52,240 --> 00:16:54,717 have a better chance. The the one that I 308 00:16:54,800 --> 00:16:59,597 worry a lot about is Africa. And the reason is how does Africa benefit from 309 00:16:59,680 --> 00:17:03,837 this? So there's obviously some benefit of globalization, better crop yields and 310 00:17:03,920 --> 00:17:08,637 so forth. But without stable governments, strong universities, major 311 00:17:08,720 --> 00:17:10,717 industrial structures, which Africa with 312 00:17:10,800 --> 00:17:12,877 some exceptions lacks, it's going to 313 00:17:12,960 --> 00:17:15,757 lag. It's been lagging for years. How do 314 00:17:15,840 --> 00:17:17,837 we get ahead of that? I don't think that 315 00:17:17,920 --> 00:17:20,157 problem is solved. 316 00:17:20,240 --> 00:17:23,277 We've seen incredible progress with AI 317 00:17:23,360 --> 00:17:26,477 today effectively beginning what people 318 00:17:26,560 --> 00:17:29,517 call solving math. 319 00:17:29,600 --> 00:17:31,757 that potentially tips physics, 320 00:17:31,840 --> 00:17:34,397 chemistry, biology. 321 00:17:34,480 --> 00:17:37,277 And we have the potential, my time frame 322 00:17:37,360 --> 00:17:39,197 is the next 5 years, others may think 323 00:17:39,280 --> 00:17:42,317 longer to be in a position to solve 324 00:17:42,400 --> 00:17:45,997 everything where the level of discovery 325 00:17:46,080 --> 00:17:48,877 and the level of new product creation, 326 00:17:48,960 --> 00:17:52,557 new materials, uh, biological, uh, 327 00:17:52,640 --> 00:17:55,277 therapeutics and such begins to grow at 328 00:17:55,360 --> 00:18:00,237 a super exponential rate. How do you think about that world in five years, 329 00:18:00,320 --> 00:18:01,277 Eric? 330 00:18:01,360 --> 00:18:03,917 >> So, um, first I think it's likely to 331 00:18:04,000 --> 00:18:07,517 occur and the reason technically is that 332 00:18:07,600 --> 00:18:09,837 the all of the large language models are 333 00:18:09,920 --> 00:18:12,477 essentially doing next word prediction. 334 00:18:12,560 --> 00:18:15,437 And if you have a limited v vocabulary, 335 00:18:15,520 --> 00:18:17,677 which math is, and you have a and 336 00:18:17,760 --> 00:18:20,557 software is, and also cyber attacks are, 337 00:18:20,640 --> 00:18:23,757 I'm sorry to say, you can make progress 338 00:18:23,840 --> 00:18:28,877 because they're scale free. All you have to do is just do more. So if you do 339 00:18:28,960 --> 00:18:32,077 software, you can verify it. You can do more software. If you do math, you can 340 00:18:32,160 --> 00:18:34,077 verify it, do more math. You're not 341 00:18:34,160 --> 00:18:36,797 constrained by real reality, physics and 342 00:18:36,880 --> 00:18:39,277 biology. So it's likely in the next few 343 00:18:39,360 --> 00:18:43,837 years that in math and software, you'll see the greatest of gains. And we all 344 00:18:43,920 --> 00:18:45,997 understand your point that math is at 345 00:18:46,080 --> 00:18:48,237 the basis of everything else. I think 346 00:18:48,320 --> 00:18:51,357 it's a a there Fifi is the expert on the 347 00:18:51,440 --> 00:18:56,557 real world. there's probably a longer period of time to get the real world 348 00:18:56,640 --> 00:18:59,997 right, which is why she founded the company of which I'm an investor. Do you 349 00:19:00,080 --> 00:19:03,997 want to talk about that? >> Yeah. Um, well, first of all, I actually 350 00:19:04,080 --> 00:19:06,477 want to respectfully disagree. Okay. I 351 00:19:06,560 --> 00:19:09,037 do not think that we will solve all the 352 00:19:09,120 --> 00:19:14,397 problems, fundamental math and physics and chemistry problems in uh in five years. 353 00:19:14,480 --> 00:19:15,837 >> We're going to take a bet on that one. 354 00:19:15,920 --> 00:19:19,517 >> Yes. So, FII14. >> Okay, you got it. 355 00:19:19,600 --> 00:19:22,157 >> We should take a bet on that. Um, part 356 00:19:22,240 --> 00:19:25,837 of humanity's greatest capability is to 357 00:19:25,920 --> 00:19:28,237 actually come up with new problems. You 358 00:19:28,320 --> 00:19:30,957 know, as Albert Einstein said, uh, most 359 00:19:31,040 --> 00:19:33,837 of science is asking the right question 360 00:19:33,920 --> 00:19:35,917 and we will continue to find new 361 00:19:36,000 --> 00:19:40,957 questions to ask and there are so many fundamental questions that in our 362 00:19:41,040 --> 00:19:42,717 science and math that we haven't 363 00:19:42,800 --> 00:19:45,437 answered. Feet your new company World 364 00:19:45,520 --> 00:19:48,877 Labs uh creating extra extraordinary uh 365 00:19:48,960 --> 00:19:52,237 persistent uh you know photorealistic 366 00:19:52,320 --> 00:19:55,517 worlds. Are you expecting that we are 367 00:19:55,600 --> 00:19:59,357 going to be spending a lot more of our time in virtual worlds? I mean my 368 00:19:59,440 --> 00:20:02,877 14-year-old boys right now are spending way too much time in their virtual 369 00:20:02,960 --> 00:20:05,597 gaming worlds. But is this what we're 370 00:20:05,680 --> 00:20:09,837 going to do in a you know uh 10 20 years 371 00:20:09,920 --> 00:20:12,637 in a post ASI world where we don't have 372 00:20:12,720 --> 00:20:17,357 to work as much we have a lot more free time our robots maybe by then are 373 00:20:17,440 --> 00:20:20,077 serving us are we going to live in the virtual worlds 374 00:20:20,160 --> 00:20:22,477 >> great question so so what we are doing 375 00:20:22,560 --> 00:20:24,957 is building large world models that's a 376 00:20:25,040 --> 00:20:27,517 problem that's after large language 377 00:20:27,600 --> 00:20:30,477 models that humans have the ability to 378 00:20:30,560 --> 00:20:32,397 have the kind of spatial intelligence 379 00:20:32,480 --> 00:20:34,877 that we can understand the physical 3D 380 00:20:34,960 --> 00:20:39,597 world we can imagine um any kind of of 381 00:20:39,680 --> 00:20:42,477 uh 3D worlds and be able to reason and 382 00:20:42,560 --> 00:20:45,837 interact with it. So we do not yet up 383 00:20:45,920 --> 00:20:48,797 till what our company has been doing we 384 00:20:48,880 --> 00:20:51,197 do not have such a world model. So World 385 00:20:51,280 --> 00:20:53,837 Labs the company I'm uh I'm uh I 386 00:20:53,920 --> 00:20:56,957 co-founded and I'm CEO in is just 387 00:20:57,040 --> 00:21:00,797 created the first large world model. So 388 00:21:00,880 --> 00:21:03,277 the future I see I actually agree with 389 00:21:03,360 --> 00:21:08,877 you that we will be spending more time in um in the multiverse 390 00:21:08,960 --> 00:21:09,277 >> yes 391 00:21:09,360 --> 00:21:12,957 >> of uh of the virtual worlds. It doesn't 392 00:21:13,040 --> 00:21:15,917 mean that the reality the real world 393 00:21:16,000 --> 00:21:18,637 this world this physical world is gone. 394 00:21:18,720 --> 00:21:24,237 It's just so much of our productivity, our entertainment, our communication, 395 00:21:24,320 --> 00:21:25,117 >> our education, 396 00:21:25,200 --> 00:21:28,877 >> our education are going to be a hybrid 397 00:21:28,960 --> 00:21:31,997 of virtual and physical world. Think 398 00:21:32,080 --> 00:21:34,797 about uh you know, think about in 399 00:21:34,880 --> 00:21:38,157 medicine, you know, how we conduct 400 00:21:38,240 --> 00:21:41,037 surgery is very much going to be a 401 00:21:41,120 --> 00:21:44,077 hybrid world of augmented reality, 402 00:21:44,160 --> 00:21:46,237 virtual reality as well as physical 403 00:21:46,320 --> 00:21:48,717 reality. And we can do that in every 404 00:21:48,800 --> 00:21:51,917 single sector. So, humanity is using 405 00:21:52,000 --> 00:21:54,877 these large world models are going to 406 00:21:54,960 --> 00:21:58,637 enter the mo uh infinite universe. Um, 407 00:21:58,720 --> 00:22:01,197 >> and I had a chance to see uh your model 408 00:22:01,280 --> 00:22:03,037 backstage. It's amazing. If you haven't 409 00:22:03,120 --> 00:22:06,397 yet, go check out Fee's World Labs. Uh, 410 00:22:06,480 --> 00:22:08,557 the technology she's building is going 411 00:22:08,640 --> 00:22:12,397 to be world changing. So, uh, my last 412 00:22:12,480 --> 00:22:14,797 question here is about human capital. So 413 00:22:14,880 --> 00:22:16,957 super intelligence has been called the 414 00:22:17,040 --> 00:22:20,157 last invention humanity will ever make. 415 00:22:20,240 --> 00:22:24,477 Uh as it could automate eventually every process. We'll see if it automates 416 00:22:24,560 --> 00:22:26,157 discovery. We'll see how much of 417 00:22:26,240 --> 00:22:29,757 creation automates. But in a world where 418 00:22:29,840 --> 00:22:31,917 the best strategy, science and economic 419 00:22:32,000 --> 00:22:35,037 decisions are being made by machines at 420 00:22:35,120 --> 00:22:37,357 some point. What is the ultimate 421 00:22:37,440 --> 00:22:39,677 irreplaceable function of human 422 00:22:39,760 --> 00:22:43,117 intellect and leadership? What are 423 00:22:43,200 --> 00:22:46,637 humans innately going to be left with in 424 00:22:46,720 --> 00:22:48,637 10, 20 years? 425 00:22:48,720 --> 00:22:51,117 >> Well, in 20 years, we will enjoy 426 00:22:51,200 --> 00:22:53,837 watching each other compete in human 427 00:22:53,920 --> 00:22:56,957 sports, knowing that the robots can beat 428 00:22:57,040 --> 00:23:01,677 us 100% of the time. >> But if you go to Formula 1, you're going 429 00:23:01,760 --> 00:23:04,957 to want to see a human driver, not an automated car. 430 00:23:05,040 --> 00:23:07,997 >> Yes. So humans will always be interested 431 00:23:08,080 --> 00:23:11,037 in what other humans can do and we'll 432 00:23:11,120 --> 00:23:13,917 have our own contests and perhaps the 433 00:23:14,000 --> 00:23:15,357 supercomputers will have their own 434 00:23:15,440 --> 00:23:19,517 contests too. But the your reasoning 435 00:23:19,600 --> 00:23:22,717 presumes many many things. It presumes a 436 00:23:22,800 --> 00:23:24,797 breakout of intelligence in computers 437 00:23:24,880 --> 00:23:27,757 that's humanlike. Unlikely probably a 438 00:23:27,840 --> 00:23:32,237 different kind of intelligence. It it presumes that humans are largely not 439 00:23:32,320 --> 00:23:34,237 involved in that process. highly 440 00:23:34,320 --> 00:23:36,797 unlikely. All of the evidence and fee 441 00:23:36,880 --> 00:23:39,357 said this very well is going to be human 442 00:23:39,440 --> 00:23:42,477 and computer interaction that basically 443 00:23:42,560 --> 00:23:45,277 we will all have so going back to what 444 00:23:45,360 --> 00:23:47,597 you said about 8 8 million people 8 445 00:23:47,680 --> 00:23:49,917 billion people with smartphones with 446 00:23:50,000 --> 00:23:52,717 Einstein in their phone the smart people 447 00:23:52,800 --> 00:23:55,117 of which there's a lot will use that to 448 00:23:55,200 --> 00:23:57,677 make themselves more productive the win 449 00:23:57,760 --> 00:24:00,397 will be teameming between a human and 450 00:24:00,480 --> 00:24:02,637 their judgment and a supercomput and 451 00:24:02,720 --> 00:24:04,877 what it can think and remember that 452 00:24:04,960 --> 00:24:09,757 There is a limit to this craze that supercomputers and super intelligence 453 00:24:09,840 --> 00:24:13,357 need energy. >> So perhaps what will happen at some 454 00:24:13,440 --> 00:24:17,837 point is the supercomputers will say huh we need more energy and these humans are 455 00:24:17,920 --> 00:24:21,837 not building fusion fast enough. So we'll accelerate it. We'll come up with 456 00:24:21,920 --> 00:24:26,157 a new form of energy. Now this is science fiction but you could imagine at 457 00:24:26,240 --> 00:24:28,077 some point the objective function of the 458 00:24:28,160 --> 00:24:31,517 system says what do I need? I need more 459 00:24:31,600 --> 00:24:36,157 chips or more energy and I'll design it myself. Now, that would be a great 460 00:24:36,240 --> 00:24:37,517 moment to see. >> I agree. 461 00:24:37,600 --> 00:24:41,277 >> I I I do want to say it's so important 462 00:24:41,360 --> 00:24:44,957 as we talk about AGI at ASI that the 463 00:24:45,040 --> 00:24:46,957 most important thing that we keep in 464 00:24:47,040 --> 00:24:51,117 mind is human dignity and human agency. 465 00:24:51,200 --> 00:24:53,837 Our world, unless we are going to wipe 466 00:24:53,920 --> 00:24:56,317 out this species, which we're not, has 467 00:24:56,400 --> 00:24:58,557 to be human- centered. Whether it's 468 00:24:58,640 --> 00:25:01,357 automation or collaboration, it needs to 469 00:25:01,440 --> 00:25:04,077 put human agency and dignity and human 470 00:25:04,160 --> 00:25:06,557 well-being in the center of all this. 471 00:25:06,640 --> 00:25:08,477 Whether it's technology, business, 472 00:25:08,560 --> 00:25:11,037 product, policy or any of that. And I 473 00:25:11,120 --> 00:25:14,797 think we cannot lose our focus uh from 474 00:25:14,880 --> 00:25:18,317 that. >> Amen. Everybody, ladies and gentlemen, 475 00:25:18,400 --> 00:25:21,117 FA Fa, Eric Schmidt, thank you all. 476 00:25:21,200 --> 00:25:26,157 Thank you. >> Every week, my team and I study the top 477 00:25:26,240 --> 00:25:29,917 10 technology metat trends that will transform industries over the decade 478 00:25:30,000 --> 00:25:33,757 ahead. I cover trends ranging from humanoid robotics, AGI, and quantum 479 00:25:33,840 --> 00:25:37,917 computing to transport, energy, longevity, and more. There's no fluff, 480 00:25:38,000 --> 00:25:41,917 only the most important stuff that matters that impacts our lives, our 481 00:25:42,000 --> 00:25:45,277 companies, and our careers. If you want me to share these meta trends with you, 482 00:25:45,360 --> 00:25:49,117 I write a newsletter twice a week, sending it out as a short two-minute 483 00:25:49,200 --> 00:25:52,717 read via email. And if you want to discover the most important metat trends 484 00:25:52,800 --> 00:25:56,557 10 years before anyone else, this report's for you. Readers include 485 00:25:56,640 --> 00:26:00,797 founders and CEOs from the world's most disruptive companies and entrepreneurs 486 00:26:00,880 --> 00:26:04,637 building the world's most disruptive tech. It's not for you if you don't want 487 00:26:04,720 --> 00:26:08,797 to be informed about what's coming, why it matters, and how you can benefit from 488 00:26:08,880 --> 00:26:10,957 it. To subscribe for free, go to 489 00:26:11,040 --> 00:26:15,917 dmmandis.com/tatrends to gain access to the trends 10 years 490 00:26:16,000 --> 00:26:21,040 before anyone else. All right, now back to this episode.