WEBVTT 1 00:00:01.760 --> 00:00:05.360 Okay, all right, well, good evening, everybody. Welcome Elon, 2 00:00:05.519 --> 00:00:08.679 thanks for being here, Thank you for having me. We feel very prov 3 00:00:08.800 --> 00:00:10.960 we're excited to have you. Right, So I'm going to start with some 4 00:00:11.039 --> 00:00:13.400 questions and then we're going to open it up. Let me get straight into 5 00:00:13.439 --> 00:00:18.320 it. So Bill Gates said, there is no one in our time who 6 00:00:18.320 --> 00:00:21.960 has done more to push the bounds of science innovation than you. What's kind 7 00:00:21.960 --> 00:00:24.920 of going to say, Well, that's a nice thing to have anyone say 8 00:00:24.960 --> 00:00:28.359 about you. Nice coming from Bill Gates. But oddly enough, when it 9 00:00:28.399 --> 00:00:33.079 comes to AI, actually for around a decade you've almost been doing the opposite 10 00:00:33.119 --> 00:00:36.240 and saying, hang on, you need to think about what we're doing and 11 00:00:36.280 --> 00:00:38.679 what we're pushing here, and what do we do to make this safe? 12 00:00:38.719 --> 00:00:42.320 And actually maybe we shouldn't be pushing as faster or as hard as we are. 13 00:00:42.920 --> 00:00:44.840 I mean, you've been doing it for a decade, Like what was 14 00:00:44.880 --> 00:00:47.799 it that caused you to think about it that way? And you know, 15 00:00:47.799 --> 00:00:50.799 why do we need to be worried? Yeah? I've been somewhat of a 16 00:00:50.799 --> 00:00:54.960 Cassandra for quite a while where people would I would tell you, like we 17 00:00:55.000 --> 00:00:57.159 should really be con learned about aid, like, what are you talking about? 18 00:00:57.200 --> 00:01:00.960 Like, I've never really had any experience with AI, but since I 19 00:01:00.039 --> 00:01:04.439 was immersed in technology, I have been immersed in technology for a long time, 20 00:01:04.719 --> 00:01:11.760 I could see it coming. So but I think this year was that 21 00:01:11.760 --> 00:01:14.760 there have been a number of breakthroughs. I mean, you know, the 22 00:01:14.760 --> 00:01:19.280 point of which someone can see a dynamically created video of themselves, you know, 23 00:01:19.319 --> 00:01:23.120 like it sort you can make a video of you saying anything in real 24 00:01:23.159 --> 00:01:27.840 time or me, And so there's sort of the deep bake videos which are 25 00:01:27.840 --> 00:01:34.560 really incredibly good, in fact, sometimes more convincing than real ones and a 26 00:01:34.560 --> 00:01:42.519 bit deep real and and then and then obviously things like chat GPT we were 27 00:01:42.599 --> 00:01:46.519 quite remarkable. Now I saw GPT one, GPT two, GPT three, 28 00:01:46.599 --> 00:01:49.079 GPT four that you know, the whole sort of lead up to that, 29 00:01:49.200 --> 00:01:53.280 So it was easy for me to kind of see where it's going. If 30 00:01:53.319 --> 00:01:56.519 you just sort of extrapolate the points on a curve and as seeing that train 31 00:01:56.560 --> 00:02:00.560 will continue, then we will have profound, artificially intelligence and obviously at a 32 00:02:00.640 --> 00:02:07.879 level that far exceeds human intelligence. So I'm glad to see at this point 33 00:02:07.919 --> 00:02:14.680 that people are taking safety seriously and I'd like I say thank you for holding 34 00:02:14.719 --> 00:02:17.000 this AI safety conference. I think actually will go down in history as being 35 00:02:17.120 --> 00:02:25.120 very important. I think it's really quite profound, and I do think overall 36 00:02:25.159 --> 00:02:32.120 that the potential is there for artificial intelligence AI too have most likely a positive 37 00:02:32.120 --> 00:02:39.080 effect and to create a future of abundance where there is no scarcity of goods 38 00:02:39.080 --> 00:02:46.159 and services. But it is somewhat of the magic genie problem where if you 39 00:02:46.199 --> 00:02:51.400 have a magic genie that can grant all the wishes, usually those stories don't 40 00:02:51.479 --> 00:02:57.080 end well. Who care for what you wish for? Including wishes? Yeah, 41 00:02:57.960 --> 00:03:02.039 so you talked a little bit about the summit and thank you for being 42 00:03:02.039 --> 00:03:06.000 engaged in it, which has been great and people enjoyed having you there. 43 00:03:06.639 --> 00:03:09.280 Puspending in this dialogue now, one of the things that we achieved today in 44 00:03:09.800 --> 00:03:16.680 the meetings between the companies and the leaders was an agreement that externally, ideally, 45 00:03:16.759 --> 00:03:22.919 governments should be doing safety testing of models before they're released. I think 46 00:03:22.919 --> 00:03:24.159 this is something that you've spoken about a little bit. It was something we 47 00:03:24.199 --> 00:03:29.479 worked really hard on because you know, my job in government is to say, 48 00:03:29.479 --> 00:03:30.719 hang on, there is a potential risk here, not a not a 49 00:03:30.759 --> 00:03:36.120 definite risk, but a potential risk of something that could be bad. My 50 00:03:36.199 --> 00:03:39.560 job is to protect the country and we can only do that if we develop 51 00:03:39.639 --> 00:03:43.639 the capability we need in our safety Institute and then go in and make sure 52 00:03:43.639 --> 00:03:46.639 we can test the models before they are released. Delight of that that happened 53 00:03:46.639 --> 00:03:50.639 today. But you know, what's your view on what we should be doing, 54 00:03:50.800 --> 00:03:53.319 right you've talked about the potential risk? Right again, we don't know, 55 00:03:53.479 --> 00:03:55.159 but you know, what are the types of things governments like our should 56 00:03:55.199 --> 00:04:00.560 be doing to manage and mitigate against those risks. Well, generally think that 57 00:04:00.879 --> 00:04:04.639 it is good for government to play a role when the public safety is at 58 00:04:04.719 --> 00:04:09.599 risk. So you know, really, for the vast majority of software, 59 00:04:10.280 --> 00:04:13.719 the public safety is not at risk. I mean, if if the app 60 00:04:13.759 --> 00:04:17.319 crashes on your phone, your laptop, it's not a massive catastrophe. But 61 00:04:17.519 --> 00:04:24.319 when you're talking about digital superintelligence, I think, which does pose a risk 62 00:04:24.399 --> 00:04:28.279 to the public, then there is a role for government to play to safeguard 63 00:04:28.319 --> 00:04:30.959 the interest of the public. And this is of course true in many fields, 64 00:04:31.680 --> 00:04:35.920 you know, aviation, cars, you know, I deal with regulators 65 00:04:36.199 --> 00:04:44.160 throughout the world because of stalling being communications, rockets being aerospace, and cars 66 00:04:44.279 --> 00:04:46.759 you know, being tracked vehicle transport. So I'm very familiar with dealing with 67 00:04:46.759 --> 00:04:50.160 with regulators, and I actually agree with the vast majority of regulations. There's 68 00:04:50.160 --> 00:04:53.920 a few that I disagree with from time to time, but zero point one 69 00:04:53.959 --> 00:04:59.360 percent, probably less than one percent of regulations I disagree with. So there 70 00:04:59.399 --> 00:05:03.920 is some concern and from people in Silicon Valley who who've never dealt with regulators 71 00:05:03.920 --> 00:05:09.040 before, and they think that this is gonna just crush innovation and slow them 72 00:05:09.079 --> 00:05:15.800 down and be annoying. But and and it will be annoying. It's true, 73 00:05:15.879 --> 00:05:19.879 they're not wrong about that. But but I think there's we've learnt over 74 00:05:19.920 --> 00:05:26.480 the years that having a referee is a good thing. And if you look 75 00:05:26.519 --> 00:05:30.959 at any sports game, there's always a referee, and and nobody's suggesting I 76 00:05:30.959 --> 00:05:33.639 think to have a sports game with that one. And I think that's the 77 00:05:34.120 --> 00:05:39.399 right way to think about this, is for for governed to be a referee, 78 00:05:39.439 --> 00:05:46.560 to make sure the sportsmanlike conduct and that the public safety is you know, 79 00:05:47.360 --> 00:05:50.720 it is addressed that we care about the public safety. Because I think 80 00:05:50.759 --> 00:05:55.319 there might be at times too much optimism about technology. And I speak I 81 00:05:55.319 --> 00:06:00.319 say that as a technologist. I mean, so I ought to know and 82 00:06:00.079 --> 00:06:04.519 and like said, on balance, I think that that the AI will be 83 00:06:04.839 --> 00:06:10.000 a forcible good most likely, but the probability of it going bad it is 84 00:06:10.040 --> 00:06:14.079 not zero. Said, so we just need to mitigate the downside potential. 85 00:06:14.319 --> 00:06:17.000 And then how you talk about referee and that's what we're demonstrate. Yeah, 86 00:06:17.120 --> 00:06:19.560 well there we go. I mean, you know, and we talked about 87 00:06:19.560 --> 00:06:23.800 this and Demos and I discussed this a long time ago, like literally facing 88 00:06:23.920 --> 00:06:29.959 right and actually Demister his credit and the credit of people in the industry, 89 00:06:30.000 --> 00:06:31.800 did say that to us. I think you say, it's not right that 90 00:06:32.000 --> 00:06:34.959 Demis and his colleagues are marking their own homework, right, there needs to 91 00:06:34.959 --> 00:06:40.160 be someone independent, and that's why we've developed the Safety Institute here. I 92 00:06:40.199 --> 00:06:43.560 mean, do you think governments can develop the expertise. One of the things 93 00:06:43.560 --> 00:06:46.480 we need to do is say, hang on, you know, Demis and 94 00:06:46.600 --> 00:06:48.399 Sam all the others have got a lot of very smart people doing this. 95 00:06:48.879 --> 00:06:54.079 Governments need to quickly tool up capability wise, personnel wise, which is what 96 00:06:54.079 --> 00:06:56.560 we're doing. I mean, do you think it is possible for governments to 97 00:06:56.600 --> 00:07:00.079 do that fast enough, given how quickly the technology is developing, or what 98 00:07:00.079 --> 00:07:01.120 do we need to do to make sure we do do it quick enough. 99 00:07:02.360 --> 00:07:05.439 No, I think it's good. It's a great point you're making. The 100 00:07:05.800 --> 00:07:12.199 pace of AI is faster than any technology I've seen in history by far, 101 00:07:13.240 --> 00:07:17.839 and it seems to be growing in capability by at least fivefold, perhaps tenfold 102 00:07:17.879 --> 00:07:26.240 per year. It'll certainly grow by an automatitude next year. Yeah, so 103 00:07:26.240 --> 00:07:30.639 so, and government isn't used to moving at that speed. But I think 104 00:07:30.720 --> 00:07:35.439 even if there are not firm regulations, even if there's not even if there 105 00:07:35.480 --> 00:07:40.839 isn't an enforcement capability, so we're having insight and being able to highlight concerns 106 00:07:40.839 --> 00:07:46.040 to the public will be very powerful. So even if that's all that's accomplished, 107 00:07:46.079 --> 00:07:48.360 I think that will be very very good. Okay, yeah, well 108 00:07:48.439 --> 00:07:51.000 hopefully we can do better than that, hopefully. Yeah. Yeah, No, 109 00:07:51.079 --> 00:07:55.800 but that's how fast we were talking before. It was striking. You're 110 00:07:55.800 --> 00:08:00.600 someone who spent their life in technology. They're living More's law. And what 111 00:08:00.680 --> 00:08:03.560 it was interesting over the last couple of days talking to everyone who's doing the 112 00:08:03.560 --> 00:08:07.399 development of this, and I think you concur with this is is just the 113 00:08:07.439 --> 00:08:13.519 pace of advancement here is unlike anything all of you have seen in your careers 114 00:08:13.519 --> 00:08:16.439 and technology. Is that fair because you've got these kind of compounding effects from 115 00:08:16.480 --> 00:08:24.279 the hardware and the data and the personnel. Yeah, I mean the two 116 00:08:26.240 --> 00:08:30.120 currently the two leading centers for a development are the San Fransco Bay area and 117 00:08:30.439 --> 00:08:33.559 the sort of London area, And there are many other places where it's being 118 00:08:33.559 --> 00:08:37.200 done, but those are two leading areas. So I think if you know, 119 00:08:37.519 --> 00:08:43.840 if the United States and the UK and China are sort of aligned on 120 00:08:45.279 --> 00:08:48.360 safety, that's all going to be a good thing. That's really that's where 121 00:08:48.600 --> 00:08:52.159 that's where the leadership is generally. You actually you mentioned China there. So 122 00:08:52.240 --> 00:08:56.600 I took a decision to invite China to summit over the last three days, 123 00:08:56.799 --> 00:09:00.559 and it was not an easy decision. A lot of people criticizes me for, 124 00:09:01.039 --> 00:09:03.000 you know, my view as if you're going to try serious conversation, 125 00:09:03.080 --> 00:09:05.559 you need to. But what would your thoughts? You do business all around 126 00:09:05.559 --> 00:09:09.960 the world. You just talked about it there. Should we be engaging with 127 00:09:09.000 --> 00:09:11.559 them? Can we trust them? Is that the right thing to have done? 128 00:09:13.399 --> 00:09:16.960 If we don't, if China is not on board with AI safety, 129 00:09:16.240 --> 00:09:22.399 it's somewhat of a moot situation. The single biggest objection that I get to 130 00:09:22.679 --> 00:09:26.799 any kind of AI regulation or sort of safety controls are well, China is 131 00:09:26.840 --> 00:09:30.639 not going to do it, and therefore they will just jump into the lead 132 00:09:30.759 --> 00:09:37.679 and exceed us all But actually China is willing to participate in AI safety and 133 00:09:37.720 --> 00:09:41.480 thank you for inviting them, and they you know, I think we should 134 00:09:43.440 --> 00:09:48.919 thank China for attending. When I was when I was in China earlier this 135 00:09:48.000 --> 00:09:52.639 year, my main subject of discussion with this the leadership in China was a 136 00:09:52.919 --> 00:09:58.240 safety and saying that this this is really something that they they should care about, 137 00:09:58.480 --> 00:10:01.639 and they took it seriously and and and you are too, which is 138 00:10:01.799 --> 00:10:05.840 which is great. And having them here I think was essential. Really if 139 00:10:05.879 --> 00:10:11.279 the if they're if they're not participants, it's it's pointless. Yeah, that's 140 00:10:11.399 --> 00:10:13.960 and I think we were pleased. I think they were engaged yesterday in the 141 00:10:13.000 --> 00:10:18.399 discussions and actually ended up signing the same communicy that everyone else did, which 142 00:10:18.440 --> 00:10:20.120 is a good stop, right. And I said, if we need everyone 143 00:10:20.159 --> 00:10:24.320 to approach us in a similar way, if we're going to have a realistic 144 00:10:24.399 --> 00:10:26.799 chance of resolving it. I was going to tell you you talked about innovation 145 00:10:26.879 --> 00:10:31.240 earlier and regulation being annoying. There was a good debate today we had about 146 00:10:31.279 --> 00:10:37.200 open source, and I think you've kind of been a proponent of algorithmic transparency 147 00:10:37.279 --> 00:10:41.240 and making some of the ex algorithms public and you actually we were talking about 148 00:10:41.600 --> 00:10:46.120 every hint and on the way in. You know, he's particularly been very 149 00:10:46.159 --> 00:10:50.879 concerned about open source models being used by bad actors. You've got a group 150 00:10:50.919 --> 00:10:54.360 of people who say they are critical to innovation happening in that distributed way. 151 00:10:54.559 --> 00:10:56.960 Look, it's it's a trick. There's probably no perfect answer, and there 152 00:10:58.000 --> 00:11:01.720 is a tricky balance your thoughts on how we should approach this open source question 153 00:11:03.000 --> 00:11:07.919 or where should we be targeting whatever regulatory or monitoring that we're going to do 154 00:11:09.639 --> 00:11:16.240 well. The open source algorithms and data tend to lag the closed source by 155 00:11:16.360 --> 00:11:22.320 six to twelve months. So that but given the rate of improvement that there's 156 00:11:22.320 --> 00:11:26.879 actually therefore quite a big difference between the closed source and the and the open 157 00:11:26.840 --> 00:11:31.559 If things are improving by a factor of let's say five or more than being 158 00:11:31.559 --> 00:11:35.919 a year behind is you're five times worse. So it's a pretty big difference. 159 00:11:35.720 --> 00:11:41.399 And that might be actually an okay situation, but it certainly we'll get 160 00:11:41.440 --> 00:11:46.679 to the point where you've got open source AI that can do that, that 161 00:11:46.759 --> 00:11:52.559 will start to approach human level intelligence, will PHAPS succeeded. I don't know 162 00:11:52.639 --> 00:11:54.639 quite what to do about it. I think it's somewhat inevitable. There will 163 00:11:54.679 --> 00:11:58.759 be some amount of open source, and I guess I would have a slight 164 00:11:58.799 --> 00:12:03.240 bias towards open source because at least you can see what's going on. It 165 00:12:03.320 --> 00:12:05.559 wears closed source. You don't know what's going on. Now, it should 166 00:12:05.600 --> 00:12:09.799 be said with AI that even if it's an open source, do you actually 167 00:12:09.840 --> 00:12:13.840 know what's going on? Because if you've got a gigantic data file and you 168 00:12:13.879 --> 00:12:22.720 know sort of billions of data points or weights and parameters, you can't just 169 00:12:22.759 --> 00:12:26.039 read it and see what it's going to do. It's a gigantic file of 170 00:12:26.159 --> 00:12:30.559 inscrutable numbers. You can test it when you when you run it, you 171 00:12:30.600 --> 00:12:33.279 can test it. You can run a bunch of tests to see what it's 172 00:12:33.279 --> 00:12:39.159 going to do. But it's probabilistic as opposed to deterministic. It's not it's 173 00:12:39.159 --> 00:12:45.200 not like traditional programming, where you've got it you've got very discrete logic and 174 00:12:45.720 --> 00:12:48.200 the outcome is very predictable, and you can read each line and see what 175 00:12:48.279 --> 00:12:54.360 each line is going to do. A neural net is just a whole bunch 176 00:12:54.360 --> 00:12:58.519 of probabilities. I mean, it sort of ends up being a giant Comma 177 00:12:58.600 --> 00:13:01.759 separated value file. It's like our digital guide is a CSP file. Really 178 00:13:05.440 --> 00:13:09.799 okay, but that's kind of what it is. Yeah, Now, that 179 00:13:11.240 --> 00:13:13.080 point you've just made is one that we have been talking about a lot, 180 00:13:13.200 --> 00:13:18.639 because again conversations with the people are developing their technology make the point that you've 181 00:13:18.679 --> 00:13:24.639 just made. It is not like normal software where there's predictability about inputs improving 182 00:13:24.840 --> 00:13:28.200 leading to this particular output improving, and as the models iterate and improve, 183 00:13:28.720 --> 00:13:31.240 we don't quite know what's going to come out the other end. I think 184 00:13:31.240 --> 00:13:35.840 Demis would agree with that, which is why I think there is this bias 185 00:13:35.840 --> 00:13:37.200 for Look, we need to get in there while the training runs are being 186 00:13:37.240 --> 00:13:41.919 done, before the models are released, to understand what is this new iteration 187 00:13:41.840 --> 00:13:46.879 broad about in terms of capability, which it sounds like you would would agree 188 00:13:46.919 --> 00:13:48.840 with. I was going to shift Goz a little bit on. You know, 189 00:13:48.879 --> 00:13:52.519 you've talked a lot about human consciousness human agency, which actually might strike 190 00:13:52.559 --> 00:13:58.559 people as as strange given that you are known for being such a brilliant innovator 191 00:13:58.559 --> 00:14:01.639 and technologist, but it's it's quite heartfelt when I hear you talk about it 192 00:14:01.679 --> 00:14:07.320 and the importance of maintaining that agency in technology and preserving human consciousness. Now 193 00:14:07.480 --> 00:14:09.519 it kind of links to the thing I was going to ask, is when 194 00:14:09.559 --> 00:14:13.720 I do interviews or talk to people out and about in this job about AI, 195 00:14:13.799 --> 00:14:16.639 the thing that comes up most actually is it probably not so much the 196 00:14:16.679 --> 00:14:20.600 stuff we've been talking about, but jobs. It's what is AI mean for 197 00:14:20.720 --> 00:14:22.960 my job? Is it going to mean that I don't have a job or 198 00:14:24.120 --> 00:14:26.080 my kids are not going to have a job. Now, you know, 199 00:14:26.679 --> 00:14:31.120 my answer as a policymaker, as a leader is actually, AI is already 200 00:14:31.120 --> 00:14:35.519 creating jobs, and you can see that in the companies that are starting. 201 00:14:35.039 --> 00:14:39.240 Also, the way it's being used is a little bit more as a co 202 00:14:39.399 --> 00:14:43.840 pilot necessarily, versus replacing the person. There's still human agency, but it's 203 00:14:43.879 --> 00:14:46.919 helping you do your job better, which is a good thing. And as 204 00:14:46.960 --> 00:14:52.120 we've seen with technological revolutions in the past, clearly there's change in the labor 205 00:14:52.159 --> 00:14:56.320 market the amount of jobs. I was quoting an MIT study today that they 206 00:14:56.360 --> 00:14:58.399 did a couple of years ago is something like sixty percent of the jobs at 207 00:14:58.399 --> 00:15:01.919 that moment exists forty years ago, so hard to predict. And my job 208 00:15:01.960 --> 00:15:05.440 is to create an incredible education system, whether it's at school, whether it's 209 00:15:05.480 --> 00:15:09.759 retraining people at any point in their career, because ultimately, if we've got 210 00:15:09.759 --> 00:15:13.320 a skill population, they'll be able to keep up with the pace of change 211 00:15:13.320 --> 00:15:16.360 and have a good life. But you know that it's still a concern. 212 00:15:16.600 --> 00:15:22.399 And what would your kind of observation be on AI and the impact on labor 213 00:15:22.440 --> 00:15:26.159 markets and people's jobs and how they should feel about that as they think about 214 00:15:26.200 --> 00:15:31.600 this. Well, I think we are seeing the most disruptive force in history 215 00:15:31.600 --> 00:15:35.879 here. You know, we have for the first time, we will have 216 00:15:35.879 --> 00:15:41.919 for the first time something that is smarter than the smartest human and that I 217 00:15:41.960 --> 00:15:46.039 mean, it's hard to say exactly what that moment is, but there will 218 00:15:46.080 --> 00:15:50.960 come a point where no job is needed. You can have a job if 219 00:15:50.000 --> 00:15:54.919 you want to have a job for sort of personal satisfaction, but the AI 220 00:15:56.039 --> 00:16:00.320 will be able to do everything. So I don't know if that makes people 221 00:16:00.320 --> 00:16:06.679 comfortable uncomfortable, it's it's you know, That's why that's why I say, 222 00:16:06.720 --> 00:16:10.759 if you, if you, if you wish for a magic genie that gives 223 00:16:10.759 --> 00:16:11.960 you any wishes you want, and there's no limit. You don't have those 224 00:16:11.960 --> 00:16:15.759 three limits three wish limits, not since you just have many as many wishes 225 00:16:15.799 --> 00:16:22.320 as you want. So it's both good and bad. One of the challenges 226 00:16:22.320 --> 00:16:26.919 in the future will be how do we find meaning in life if if you 227 00:16:27.039 --> 00:16:32.080 have a magic genie that can do everything you want. I do think we's 228 00:16:32.200 --> 00:16:33.600 it's hard. You know, when there's when when there's new technology, it 229 00:16:33.639 --> 00:16:38.080 tends to have usually follow an S curve. In this case, we're going 230 00:16:38.120 --> 00:16:41.120 to be on the exponential portion of the S curve for a long time, 231 00:16:44.200 --> 00:16:47.720 and we have to like so we able to ask for anything. And it 232 00:16:47.759 --> 00:16:51.639 won't be we won't have universal basic income. We'll have universal high income. 233 00:16:52.559 --> 00:16:55.720 So in some in some sense it will be somewhat of a leveler or an 234 00:16:55.759 --> 00:17:02.600 equalizer, because really I think everyone will have access to this magic Jenny anyway, 235 00:17:02.600 --> 00:17:03.880 able to ask any question. It will certainly be good for education, 236 00:17:04.039 --> 00:17:07.400 you could it'll be the best tutor you could add them the most patient tutor, 237 00:17:08.200 --> 00:17:15.799 UH said there all day. And there will be no shortage of goods 238 00:17:15.799 --> 00:17:21.759 and services, will be an age of abundance. I think if i'd recommend 239 00:17:21.839 --> 00:17:27.400 people read in Banks, the Banks culture books are probably the best in visioning. 240 00:17:27.400 --> 00:17:30.880 In fact, not probably, they're definitely by far the best envisioning of 241 00:17:30.920 --> 00:17:36.039 an a future. There's nothing even close. So I'd recommend, really recommend 242 00:17:36.160 --> 00:17:41.440 Banks. I'm a very big fan. All these books are good. There's 243 00:17:41.480 --> 00:17:47.960 not Jay which one all of them. So so that's that'll give you a 244 00:17:47.960 --> 00:17:56.319 sense of what is a I guess a fairly utopian or pro topin future with 245 00:17:56.319 --> 00:18:02.240 with Ai, which is good for as you said, it's a universal high 246 00:18:02.240 --> 00:18:03.440 income, which is a nice phrase, and that's it's good from a kind 247 00:18:03.440 --> 00:18:08.759 of materialistic sense of abundance. Actually, that then leads to the question that 248 00:18:08.799 --> 00:18:11.200 you pose, right, I'm someone who believes, you know, work gives 249 00:18:11.200 --> 00:18:14.920 you meaning. I write a lot about that. As you know, I 250 00:18:14.960 --> 00:18:17.400 think work is a good thing. It, you know, gives people purpose 251 00:18:17.400 --> 00:18:19.720 in their lives. And if you then remove a large chunk of that, 252 00:18:21.200 --> 00:18:22.240 you know, what does that mean? And where do you get that? 253 00:18:22.759 --> 00:18:26.200 You know, where do you get that drive, that motivation, that purpose? 254 00:18:26.240 --> 00:18:27.319 I mean you're talking about it. You work a lot of hours, 255 00:18:29.319 --> 00:18:32.519 you know, as I was mentioning, when we were talking earlier. I 256 00:18:32.519 --> 00:18:37.160 have to somewhat engage in liberate suspension of disbelief because I'm putting so much blood, 257 00:18:37.160 --> 00:18:40.599 sweat and tears into a work project and burning the you know, three 258 00:18:40.640 --> 00:18:45.000 am oil. Then I'm like, wait, why am I doing this? 259 00:18:45.200 --> 00:18:48.920 I can just wait for the AI to do it. I'm just lashing myself 260 00:18:48.960 --> 00:18:56.359 for no reason. Must be a gutton for punishment or something. So we 261 00:18:56.440 --> 00:18:59.359 called called called demos and tell them to hurry up and then you can have 262 00:18:59.400 --> 00:19:03.160 a holiday. It's the plan. Yeah, No, it's a Look, 263 00:19:03.160 --> 00:19:04.920 it's a tricky. It's a tricky thing because I think you know, part 264 00:19:04.960 --> 00:19:10.200 of our job is to make sure that we can navigate to that very I 265 00:19:10.240 --> 00:19:15.279 think largely positive place that you're described people through it between now and then, 266 00:19:15.359 --> 00:19:18.160 because these things bring a lot aboutter change in in the labor market as we've 267 00:19:18.200 --> 00:19:22.240 seen. Yeah, I think it's probably is generally a good thing because you 268 00:19:22.279 --> 00:19:26.359 know, there are a lot of jobs that are uncomfortable or dangerous or which 269 00:19:26.400 --> 00:19:30.359 sort of tedious, and the computer will have no problem doing that. We're 270 00:19:30.400 --> 00:19:34.079 happy to do that all day long. So you know, it's it's fun 271 00:19:34.119 --> 00:19:37.880 to cook food, but it's not that fun to wash the dishes, but 272 00:19:37.920 --> 00:19:41.359 the computer is perfectly happy to watch the dishes. I guess there is. 273 00:19:42.599 --> 00:19:48.240 You know, we still have sports like where where humans compete and like the 274 00:19:48.279 --> 00:19:52.720 Olympics, and obviously a machine can can go faster than any human, but 275 00:19:52.759 --> 00:20:00.200 we still have we still humans race against each other and and have a you 276 00:20:00.200 --> 00:20:03.519 know, these sports competitions against each other where even though the machines are better, 277 00:20:03.519 --> 00:20:06.640 which I guess computnitity, who can be the best human at something? 278 00:20:07.279 --> 00:20:11.640 And people do find performing in that. So I guess that's perhaps a good 279 00:20:11.680 --> 00:20:14.799 example of how even when machines are faster than are stronger than us, we 280 00:20:14.839 --> 00:20:18.480 still find a way we saw enjoy computing against other humans, so at leaste, 281 00:20:18.480 --> 00:20:21.640 who was the best human? Yeah, that's a good it's a good 282 00:20:21.640 --> 00:20:25.480 analogy. And we've been talking a lot about managing the risks. Just before 283 00:20:25.519 --> 00:20:29.000 we move on and finish on AI is just talk a little bit about the 284 00:20:29.039 --> 00:20:33.880 opportunities. You know, you're engaged in lots of different companies. You're an 285 00:20:33.880 --> 00:20:38.359 obvious line which is doing which is doing some exciting stuff. You touched on 286 00:20:38.359 --> 00:20:41.920 the thing that I'm probably most excited about, which is an education. Yeah, 287 00:20:42.119 --> 00:20:47.200 and I think many people will have seen sal CON's video from earlier this 288 00:20:47.279 --> 00:20:48.799 year. Is Ted talk about as you talked about, it's like a personal 289 00:20:48.799 --> 00:20:52.599 tutor. Yeah, an amazing personal tutor, an amazing personal tutor. And 290 00:20:52.599 --> 00:20:57.359 we know the difference in learning having that personal tutor is incredible compared to class 291 00:20:57.400 --> 00:21:00.799 from learning. So if you can have every child, I'll have a personal 292 00:21:00.839 --> 00:21:03.559 tutor specifically for them that then just evolves with them over time. Yeah, 293 00:21:03.599 --> 00:21:07.119 it could be extraordinary. And so that you know, for me, I 294 00:21:07.119 --> 00:21:08.759 look at it, I think, gosh, that is within reach at this 295 00:21:08.880 --> 00:21:12.599 point. And that's one of the benefits I'm most excited about. When you 296 00:21:12.640 --> 00:21:15.880 look at the landscape of things that you see as possible, what is it 297 00:21:15.960 --> 00:21:22.680 that you know you are particularly excited about. I think certainly AI AI tutors 298 00:21:22.720 --> 00:21:27.279 are going to be amazing, perhaps already are. I think there's also perhaps 299 00:21:27.400 --> 00:21:32.480 companionship, which may seem odd because how can the computer really be your friend? 300 00:21:32.480 --> 00:21:36.279 But if you if you have an AI that has memory, you know, 301 00:21:36.359 --> 00:21:40.000 and remembers all of your interactions and has read everything, you've got to 302 00:21:40.000 --> 00:21:41.799 actually like give it permission to read everything you've ever done, so it really 303 00:21:41.799 --> 00:21:47.559 will know you better than anyone, perhaps even yourself, and and and where 304 00:21:47.559 --> 00:21:49.799 you can talk to it every day and those conversations spoiled upon each other, 305 00:21:51.119 --> 00:21:56.319 you will actually have a great friend as long as that friend can stay your 306 00:21:56.359 --> 00:22:00.400 friend and not get turned off or something. Don't turn off my friends. 307 00:22:03.759 --> 00:22:07.720 But I think that will actually be a real thing. And I have a 308 00:22:08.119 --> 00:22:12.640 one of my sons is sort of has some learning disabilities and has trouble making 309 00:22:12.680 --> 00:22:17.160 friends actually, and I was like, well, you know he AI friend 310 00:22:17.200 --> 00:22:21.319 would actually be a great fri enid Okay, you know's that was a surprising 311 00:22:21.359 --> 00:22:25.599 answer. That's actually it's worth worth reflecting on that. That's really interesting. 312 00:22:25.640 --> 00:22:30.000 I mean we're already seeing it actually as we deliver you know, psychotherapy anyway, 313 00:22:30.400 --> 00:22:33.359 now doing far more by digitally and by telephone to people, and it's 314 00:22:33.400 --> 00:22:36.799 making a huge difference. And you can see a world and which actually, 315 00:22:37.160 --> 00:22:41.319 you know, AI can provide that social benefit to people. Just a quick 316 00:22:41.400 --> 00:22:45.799 question on on X and then we should open it up to everybody. You 317 00:22:45.880 --> 00:22:51.640 made a change when in one of the made many changes, but one of 318 00:22:51.680 --> 00:22:56.279 the one of the changes, We've got a real thing about it. You 319 00:22:56.359 --> 00:23:00.920 really do, really do one of the changes which you know kind of you 320 00:23:00.960 --> 00:23:03.200 know, it goes into the space that you know, we have to operate 321 00:23:03.240 --> 00:23:07.720 in. And this, this balance between free speech and moderation is you know, 322 00:23:07.759 --> 00:23:11.240 we grapple with as politicians. You were grappling with your own version of 323 00:23:11.240 --> 00:23:15.839 that, and you you moved away from a kind of manual human way of 324 00:23:15.920 --> 00:23:19.960 doing it, the moderation to the community notes and yeah, and I think 325 00:23:21.000 --> 00:23:22.440 that's you know, it was an interesting change, right, It's not what 326 00:23:22.480 --> 00:23:26.680 everyone else has done. It would be good, you know, what's what 327 00:23:26.720 --> 00:23:29.200 was the reasoning behind that? And why do you think that is a better 328 00:23:29.240 --> 00:23:33.519 way to do that? Yeah? Part of the problem is if you if 329 00:23:33.519 --> 00:23:37.279 you empower people as sensors, then well there's going to be some amount of 330 00:23:37.279 --> 00:23:42.400 bias they have, and then whoever appoints the sensors is effectively in control of 331 00:23:42.480 --> 00:23:47.720 information. So then the idea behind community notice, well, how do we 332 00:23:47.920 --> 00:23:52.599 have a consensus driven I mean, so it's not really censoring it, but 333 00:23:52.680 --> 00:23:56.799 consensus driven approach to truth? How do we how do we how do we 334 00:23:56.839 --> 00:24:02.640 make things the least amount untrue? Like you say, like what you can't 335 00:24:03.319 --> 00:24:07.920 perhaps get to pure truth, but you can aspire to be more truthful So 336 00:24:08.079 --> 00:24:12.559 the thing about community notes is it doesn't actually delete anything. It simply adds 337 00:24:12.599 --> 00:24:15.519 context. And that context could be this thing is untrue for the following reasons. 338 00:24:18.400 --> 00:24:22.960 And but but importantly with community notes, everything is open source actually, 339 00:24:23.039 --> 00:24:27.759 so you can see the software, every line of the software. You can 340 00:24:27.799 --> 00:24:32.160 see all of the data that went into a community note, and and you 341 00:24:32.200 --> 00:24:36.000 can independently create that community note. So if you've got if you see a 342 00:24:36.039 --> 00:24:38.559 manipulations of data, you can actually highlight that and say well this, this, 343 00:24:40.079 --> 00:24:42.440 this, the appears to be some gaming of the system, and you 344 00:24:42.440 --> 00:24:48.599 can suggest improvement. So it's it's it's maximum transparency, which is i think 345 00:24:48.759 --> 00:24:52.279 combined with the kind of wisdom of the crowds and transparency to get to a 346 00:24:52.279 --> 00:24:56.119 better answer is you. And really one of the key elements of community notes 347 00:24:56.200 --> 00:24:59.960 is that in order for a note to be shown, people who have historic 348 00:25:00.279 --> 00:25:04.720 disagreed must agree. And there is a bit of AI usage here, so 349 00:25:04.759 --> 00:25:12.319 there's will populate a parameter space around each contributor to community notes, and then 350 00:25:12.720 --> 00:25:18.400 the parameter space so everyone's got basically these these vectors associated with them, which 351 00:25:18.519 --> 00:25:22.400 so it's it's not as simple as as right or left it's saying it's it's 352 00:25:22.400 --> 00:25:26.480 more, it's several hundred vectors. That that because things are more complicated than 353 00:25:26.519 --> 00:25:33.079 something right right or left and and and then will we'll do sort of inverse 354 00:25:33.119 --> 00:25:37.839 correlation, say like, okay, these these people generally disagree, but they 355 00:25:37.880 --> 00:25:41.119 agree about this note. Okay, so then that so then that that that 356 00:25:41.160 --> 00:25:45.519 gives the note credibility. Yeah, that's that's the that's the core of it. 357 00:25:45.559 --> 00:25:49.359 And it's working quite well. I get to see a note actually be 358 00:25:49.720 --> 00:25:53.079 be present for more than a few hours that that is incorrect. So the 359 00:25:53.079 --> 00:25:56.119 batting average is extremely good. And when I ask PEP, people say, 360 00:25:56.119 --> 00:26:00.720 oh, they're worried about community notes sort of being dis like, send me 361 00:26:00.759 --> 00:26:04.799 one and then they can't. So so I think it's I think it's quite 362 00:26:04.799 --> 00:26:10.640 good. I mean, the general aspiration is with the X platform is to 363 00:26:10.680 --> 00:26:14.839 inform and entertain the public and to be as accurate as possible and as truthful 364 00:26:14.880 --> 00:26:18.279 as possible, even if someone doesn't like the truth. You know, people 365 00:26:18.279 --> 00:26:26.400 don't always like the truth, not always, but that's that's that's the aspiration. 366 00:26:26.599 --> 00:26:29.880 And I think if we are if we stay true to the truth, 367 00:26:30.039 --> 00:26:33.400 then I think we'll find that people use the system to learn what is going 368 00:26:33.440 --> 00:26:41.759 on and to learn. I think actually truth pays, so I think it'll 369 00:26:41.759 --> 00:26:45.480 be what I mean, assuming you don't want to engage your self delusion, 370 00:26:45.599 --> 00:26:52.240 then I think it's it's the smart move. So excellent, very helpful. 371 00:26:52.279 --> 00:26:55.839 Right, let's open it up to all our guests here, and we've got 372 00:26:55.839 --> 00:26:57.319 some microphones that they'll come. Put your hands up, they'll come and find 373 00:26:57.359 --> 00:27:03.519 you. We got yes, go for it. Thank you, good evening. 374 00:27:03.720 --> 00:27:07.119 Alice Bentink from Entrepreneur. First, thank you for a fascinating conversation. 375 00:27:07.160 --> 00:27:11.279 I suppose a question for each of you, Prime Minister. The UK has 376 00:27:11.440 --> 00:27:15.640 some of the best universities in the world, we have the talent. What 377 00:27:15.680 --> 00:27:19.839 will it takes the UK to be a real breeding ground for unicorn companies? 378 00:27:21.680 --> 00:27:26.839 And elon being a founder in the UK is still a non obvious career choice 379 00:27:26.880 --> 00:27:30.319 for the most exceptional technical talent. What are the cultural elements that we need 380 00:27:30.319 --> 00:27:33.960 to put into place to change this? Thank you both. Well you want 381 00:27:33.960 --> 00:27:37.319 to go first, go for it? Sure? Well, you're right that 382 00:27:37.400 --> 00:27:44.880 there are cultural elements where you know, the culture should celebrate creating new companies 383 00:27:45.559 --> 00:27:53.720 and there should be a bias towards supporting small, small companies because they're the 384 00:27:53.759 --> 00:28:00.119 ones that need nurturing. The larger companies really don't need nurturing. So you 385 00:28:00.160 --> 00:28:02.319 know, just you can think of it's sort of like a garden. If 386 00:28:02.319 --> 00:28:04.200 it's a little sprout that needs needs nurturing, if it's a mighty oak, 387 00:28:04.240 --> 00:28:10.519 it doesn't need quite as much. So I think that that is a mindset 388 00:28:10.599 --> 00:28:17.599 change that is important. But but I should mention that London is you know, 389 00:28:18.000 --> 00:28:21.559 London and San Francisco or the Bay Area are really the two census for 390 00:28:21.960 --> 00:28:25.960 AI. So that so London is actually doing very well on that front. 391 00:28:26.240 --> 00:28:30.720 That the two I said, the two leading locations on Earth. You know, 392 00:28:30.640 --> 00:28:33.680 San Francisco's probably head of London, but London is really very strong. 393 00:28:33.960 --> 00:28:41.200 But London Area, greater London home counties, I guess, keep keep going. 394 00:28:42.400 --> 00:28:48.160 So I'm just saying objectively this is the case. And but you do 395 00:28:48.240 --> 00:28:52.240 need that. You need you need the infrastructure, need, you need landlords 396 00:28:52.240 --> 00:28:56.640 who are willing to rant to new companies, who you need law firms and 397 00:28:56.680 --> 00:29:00.160 accounts that are willing to support new companies. And it's generally it is a 398 00:29:00.200 --> 00:29:03.759 mindset change, and I think some of that is happening. But I think 399 00:29:04.119 --> 00:29:07.640 really it's just culturally people need to decide this is this is a good thing. 400 00:29:08.359 --> 00:29:11.400 Yeah yeah, actually, well thanks for what you said about the UK. 401 00:29:11.480 --> 00:29:14.960 It's something that we work hard on. Lots of people in the room 402 00:29:15.720 --> 00:29:19.519 are part of what makes this a fabulous place for companies, including Alice. 403 00:29:19.599 --> 00:29:25.000 So that's what i'd say is my job is to get all the you know, 404 00:29:25.000 --> 00:29:26.880 the nuts and bolts right, make sure that all of you are starting 405 00:29:26.880 --> 00:29:30.519 companies can raise a capital that you need everything from you know, your seed 406 00:29:30.559 --> 00:29:36.640 funding with our incredible you know eis tax release all the way through to you 407 00:29:36.680 --> 00:29:38.880 know, your late stage rounds. And we need reform of our pension funds 408 00:29:38.880 --> 00:29:44.680 and the Chancellor's got a bunch of incredible reforms to unlock capital from all the 409 00:29:44.720 --> 00:29:47.519 people who have it and deploy it into growth equity. Right, that is 410 00:29:47.519 --> 00:29:49.079 a work in progress. We're not there yet, but I think we're we're 411 00:29:49.079 --> 00:29:52.599 making We're making good progress. We need talent, we need people, you 412 00:29:52.599 --> 00:29:56.240 know, so that means an education system that prioritizes the things that matter. 413 00:29:56.319 --> 00:30:00.559 And you've seen my reforms. I go on about more matt, more maths, 414 00:30:00.599 --> 00:30:03.079 more maths, but I think that it's important, but also attracting the 415 00:30:03.079 --> 00:30:07.880 best and the brightest here. If you look at our fastest growing companies in 416 00:30:07.240 --> 00:30:11.319 this country, and I think it's probably the same in the US, over 417 00:30:11.400 --> 00:30:15.039 half of them have a non British founder, right, And so that tells 418 00:30:15.039 --> 00:30:18.920 you we've got to be a place that is open to the world's best and 419 00:30:18.960 --> 00:30:22.119 brightest entrepreneurial talent. So the visa regime that we've put in place, I 420 00:30:22.119 --> 00:30:25.839 think, does that makes it easy for those people to come here. And 421 00:30:25.880 --> 00:30:29.039 then actually it's the thing that we've spent the beginning of the session talking about 422 00:30:29.039 --> 00:30:33.200 the regulation, right, making sure that we've got a regulatory system that's pro 423 00:30:33.240 --> 00:30:37.519 innovation. That Yeah, of course we always need guardrails on the things that 424 00:30:37.680 --> 00:30:41.039 will worry us, but we've got to create a space for people to innovate 425 00:30:41.079 --> 00:30:44.000 and do different things. Now, those are all my jobs. The thing 426 00:30:44.039 --> 00:30:45.920 that is tougher is the thing that Elon talked about, which is culture. 427 00:30:47.480 --> 00:30:52.119 Right, It's how do you transpose that culture from places like Silicon Valley across 428 00:30:52.160 --> 00:30:57.039 the world where people are unafraid to give up the security of a regular paycheck 429 00:30:57.200 --> 00:31:02.000 to go and start something, be comfortable with failure, You talk about that 430 00:31:02.039 --> 00:31:03.480 a lot. I think you talk about it more and when you're playing games, 431 00:31:03.559 --> 00:31:07.559 right, but that you've got to be comfortable failing and knowing that that's 432 00:31:07.640 --> 00:31:11.559 just part of the process. And that is as a tricky cultural thing to 433 00:31:11.599 --> 00:31:15.319 do overnight, but it's an important part of I think creating that kind of 434 00:31:15.440 --> 00:31:18.839 environment. Yeah, if you don't succeed with your first startup, it shouldn't 435 00:31:18.839 --> 00:31:22.640 be a sort of a catastrophic career ending exactly, thing it should be you 436 00:31:22.680 --> 00:31:25.599 know, well good I think generally should like should be like, well, 437 00:31:25.720 --> 00:31:26.759 you know, you gave it a good shot, you know, and and 438 00:31:26.960 --> 00:31:33.119 and now try again and exactly. It's so one thing I'm going to mention 439 00:31:33.240 --> 00:31:34.759 is like, obviously creating a company is sort of a high risk, high 440 00:31:34.759 --> 00:31:41.839 reward situation. But I don't know quite what the how it works in the 441 00:31:41.960 --> 00:31:48.039 UK, I think probably better than continental Europe, but the stock options are 442 00:31:48.480 --> 00:31:52.279 very difficult in most parts of Europe. I'm not sure how does you give 443 00:31:52.319 --> 00:31:56.079 it. But if somebody's basically going to risk the sort of life savings and 444 00:31:56.279 --> 00:32:00.640 with and the vast majority of startups fail, So I mean you hear about 445 00:32:00.640 --> 00:32:07.000 the startups that succeed, but most companies are most startups consist of you know, 446 00:32:07.119 --> 00:32:13.359 a massive amount of work followed by failure. That's actually most most companies. 447 00:32:13.720 --> 00:32:16.119 And so it's a high risk kigh reward and so the higher reward part 448 00:32:16.279 --> 00:32:20.519 does need to be there for it to make sense. I think that was 449 00:32:20.559 --> 00:32:23.759 a very soft pitch for a tax policy that I need chance that. But 450 00:32:23.960 --> 00:32:28.119 actually I can tell you so like AI agree, and we have so we 451 00:32:28.200 --> 00:32:31.599 have, I think relative to certainly European countries, but certainly the US, 452 00:32:31.799 --> 00:32:37.039 definitely California a much lower rate of capital gains tax right. So for those 453 00:32:37.079 --> 00:32:39.599 people who are risking and growing something like, we think the reward should be 454 00:32:39.640 --> 00:32:44.680 there at the end. So it's twenty percent capital gains tax right. And 455 00:32:44.839 --> 00:32:47.359 on stock options, I don't know if we've got anyone from Index Ventures in 456 00:32:47.920 --> 00:32:52.440 the room. So Index one of our bleeding VC funds here. They do 457 00:32:52.559 --> 00:33:00.160 a regular report looking at most countries' tax treatment of stock options. And you 458 00:33:00.160 --> 00:33:02.960 know, when I was a Chancellor of Treasury Sexuary equivalent, you know, 459 00:33:04.000 --> 00:33:06.960 we were I think down at we were pretty good, but we were fourth 460 00:33:07.039 --> 00:33:08.839 or fifth. Then I said we need to for exactly the reason that you 461 00:33:08.920 --> 00:33:12.839 mentioned was like this has got to be the best place for renovative We need 462 00:33:12.880 --> 00:33:15.039 to move that up. And I think in the last iteration of that report 463 00:33:15.079 --> 00:33:19.640 we had because of the changes that Jeremy and I have made, we have 464 00:33:19.720 --> 00:33:22.799 moved up to I think second from memory. Hopefully that actually give you and 465 00:33:22.839 --> 00:33:27.680 everyone else some comfort that we recognize. That's important because when people work hard 466 00:33:27.680 --> 00:33:30.960 and risk things, yeah, they should be able to enjoy the rewards high 467 00:33:30.000 --> 00:33:32.480 reward. Yeah. And I think we have a we very much have a 468 00:33:32.519 --> 00:33:36.680 tax system that supports that. And those are the values that you know, 469 00:33:36.720 --> 00:33:38.119 I believe in, and I think most of us in this room probably do 470 00:33:38.240 --> 00:33:42.680 as well. Right, next, next question, I've got seven front of 471 00:33:42.720 --> 00:33:47.720 me, and then I'll come over here. Gone gone, So thanks very 472 00:33:47.799 --> 00:33:53.240 much. We've talked about some really big ideas, global changing ideas. I'm 473 00:33:53.279 --> 00:33:58.400 really interested, particularly in the Contexte creation of science and technology, superhubs and 474 00:33:58.400 --> 00:34:02.759 so on. How does that app onto the everyday lives of people living in 475 00:34:02.839 --> 00:34:07.160 say Austin, Texas, to choose around you and more in my case Nottingham, 476 00:34:07.240 --> 00:34:10.000 East Midlands. Uh, what what is? How do you see that 477 00:34:10.079 --> 00:34:17.920 evolving for people? You know every day sort of every day effective AI for 478 00:34:19.079 --> 00:34:23.000 context, Elon so SEB. SEB runs our equivalent of CDs, right or 479 00:34:23.000 --> 00:34:25.920 Walgreens, so you know when actually I visited, right, So he's got 480 00:34:25.960 --> 00:34:30.159 millions of people coming in the shops every day, and it's making sure how 481 00:34:30.199 --> 00:34:32.159 do we make this relevant? I think is your question, how is this 482 00:34:32.239 --> 00:34:36.159 relevant to that person? You know? Maybe actually let me go I'll go 483 00:34:36.239 --> 00:34:40.039 first on that because I think it's a fair a fair point. I was 484 00:34:40.079 --> 00:34:43.800 just going over with the team a couple of things that we're doing because I 485 00:34:43.800 --> 00:34:45.480 was saying, like, how are we doing AI right now that it's making 486 00:34:45.519 --> 00:34:50.239 a difference to people's lives. And we have this thing called gov dot UK, 487 00:34:50.480 --> 00:34:53.119 which is which actually when we when it happened several years ago, was 488 00:34:53.159 --> 00:34:58.840 a pioneering thing. All the government information brought together on one website, gov 489 00:34:58.880 --> 00:35:01.519 dot uk, so you need to get a driving license, passport, any 490 00:35:01.559 --> 00:35:06.800 interaction with government. It was centralized in a very easy, relatively easy to 491 00:35:06.880 --> 00:35:13.800 use way, better than most. So we're about so we're about to deploy 492 00:35:14.199 --> 00:35:16.360 AI across that PI form. So that is something that I think, you 493 00:35:16.400 --> 00:35:21.320 know, several million people a day use right, so a large chunk of 494 00:35:21.320 --> 00:35:24.400 the population is interacting with gov dot UK every single day to do all these 495 00:35:24.559 --> 00:35:28.440 day to day tasks. Right, every one of your customers is doing all 496 00:35:28.440 --> 00:35:31.960 those things, and so we're about to deploy AI into that to make that 497 00:35:32.079 --> 00:35:36.800 whole process so much easier. Because you know, some people will be like, 498 00:35:36.800 --> 00:35:39.760 look, well I'm currently here and I've lost my passport and my flights 499 00:35:39.760 --> 00:35:44.000 in five hours. You know at the moment that would require you know how 500 00:35:44.039 --> 00:35:45.840 many steps to figure out what you do? You know, well, actually, 501 00:35:45.880 --> 00:35:49.360 when we deploy the AI, it should be that you could just literally 502 00:35:49.599 --> 00:35:52.039 say that and boom boom boom, bom boom. This is what we're going 503 00:35:52.079 --> 00:35:54.599 to do walking through it, and that's going to benefit millions of millions of 504 00:35:54.639 --> 00:35:58.840 people every single day, right, Because that's a very practical way in my 505 00:35:59.119 --> 00:36:01.679 seat, that I can start using this technology to help people in their day 506 00:36:01.679 --> 00:36:07.039 to day lives, not just healthcare discoveries and everything else that we're also doing. 507 00:36:07.079 --> 00:36:09.880 But I thought that's quite a powerful demonstration of literally your day to day 508 00:36:10.000 --> 00:36:15.079 customer seeing actually their just day to day life get a little bit easier because 509 00:36:15.119 --> 00:36:16.760 of something that you know, Elon Demis and others in this room have helped 510 00:36:16.800 --> 00:36:21.800 create. Yeah, exactly, The most immediate thing is just being able to 511 00:36:21.840 --> 00:36:27.360 ask, like having a very smart friend that you can ask anything, you 512 00:36:27.400 --> 00:36:30.920 know, ask how to make something, how to solve any problem, and 513 00:36:30.960 --> 00:36:37.920 it'll tell you so. And obviously companies are going to adopt this, so 514 00:36:37.800 --> 00:36:40.920 I think maybe you'll have much better customer service. I guess essentially that'll probably 515 00:36:40.920 --> 00:36:47.960 be the first thing you notice. And then if we talked about education, 516 00:36:50.679 --> 00:36:55.079 so having a tutor, so if you're trying to understand a subject, like 517 00:36:55.159 --> 00:37:01.440 having a phenomenal tutor on any subject, is that that's really pretty much there 518 00:37:01.480 --> 00:37:05.679 already almost. I mean we need to obviously, I need to stop hallucinating 519 00:37:05.840 --> 00:37:09.199 before you know it can't give you. I mean that we still have a 520 00:37:09.239 --> 00:37:13.320 little bit of the problem where it can give you an answer that's competently wrong 521 00:37:14.280 --> 00:37:19.079 with great grammar and you know, in bullet points and everything in citations, 522 00:37:19.079 --> 00:37:22.800 it was not real. So that's be okay. We need to make sure 523 00:37:22.800 --> 00:37:28.800 it's it's not it's not it's not giving you competently wrong tunor answers. But 524 00:37:28.800 --> 00:37:34.800 but that's going to happen pretty quickly where it is actually correct. So yeah, 525 00:37:34.800 --> 00:37:37.639 I'm gonna say for any for any parent who was homeschooling during COVID and 526 00:37:37.800 --> 00:37:42.280 realizing what their kids needed to be helped with. That will come as an 527 00:37:42.360 --> 00:37:45.159 enormous relief. I think, yeah, very very good. Right have we 528 00:37:45.239 --> 00:37:47.880 got let's go questions over here? Who have we got? We only microphones 529 00:37:49.000 --> 00:37:53.360 or brand you they're perfect? Hi, Brent Herman. So you know you've 530 00:37:53.360 --> 00:37:58.880 spoken eloquently about abundance in the Age of abundance, So it feels obviously with 531 00:37:59.000 --> 00:38:05.440 AI it's everything else were everywhere all at once. But with robots, and 532 00:38:05.800 --> 00:38:07.039 to get the age of abundance, we'll need a lot of robots. I 533 00:38:07.119 --> 00:38:12.079 know you're working on robots as well. Are there sort of constraints that we 534 00:38:12.119 --> 00:38:15.559 should think of our politician should be thinking of that we might get one country 535 00:38:15.639 --> 00:38:19.320 might get heavily behind in robots. They can do all these things and end 536 00:38:19.360 --> 00:38:23.559 to the age of bunds and therefore be it a strategic disadvantage. Well, 537 00:38:25.719 --> 00:38:30.239 really, anything that can be actuated by a computer is effectively a robot. 538 00:38:30.719 --> 00:38:37.920 So you can think of frankly, Tesla cars or robots on wheels. Anything 539 00:38:37.920 --> 00:38:44.960 that's connected to the Internet is effectively an endpoint actuator for artificial intelligence. So 540 00:38:47.800 --> 00:38:51.840 you've got Bust and Dynamics. Obviously, they've been making impressive robots for a 541 00:38:51.920 --> 00:38:54.960 while. I think there are at this point mostly on by Hundai, So 542 00:38:55.119 --> 00:39:00.559 I guess who is probably gonna make robots of that are humanoid and and some 543 00:39:01.760 --> 00:39:05.639 interesting shapes that I wasn't just patting, like the one that looks like a 544 00:39:06.119 --> 00:39:08.280 has wheels and looks sort of like a kangaroo on wheels. I'm sure what 545 00:39:08.400 --> 00:39:16.960 that is, but it looks a little demanded, frankly. But but there's 546 00:39:16.960 --> 00:39:22.119 gonna be all sorts of all sorts of robots. You've got the company Dison 547 00:39:22.239 --> 00:39:29.320 in the UK, which I think there's some pretty impressive things. I think 548 00:39:29.320 --> 00:39:32.480 the UK will not be behind actually on that front. UK also has ARM, 549 00:39:32.639 --> 00:39:38.599 which is really the best one one of the best paths, the best 550 00:39:39.840 --> 00:39:45.119 in chip design in the world. Tees I use a lot of a lot 551 00:39:45.159 --> 00:39:51.280 of ARM technology. Almost everyone does, actually, so I think UK is 552 00:39:51.280 --> 00:39:55.599 in a strong position. Germany obviously makes a lot of robots, natural robots 553 00:39:57.000 --> 00:40:01.400 that I mean, I think generally countries that make robots of any kind, 554 00:40:01.880 --> 00:40:07.920 even if they seem somewhat conventional'll be We'll be fine. I do think there 555 00:40:08.000 --> 00:40:15.679 is a there is a safety concern, especially with humanoid robots, because at 556 00:40:15.760 --> 00:40:19.519 least the car can't chase you into this building not very easily, you know, 557 00:40:19.760 --> 00:40:22.039 or chase you over a tree, or you know, you can sort 558 00:40:22.039 --> 00:40:28.320 of run up a flight of stairs and get away from a tesla. I 559 00:40:28.360 --> 00:40:32.440 think it's a Stephen King movie about that if your car gets possessed. So, 560 00:40:34.800 --> 00:40:37.800 but if you have a humanoid robot, it can it can basically chase 561 00:40:37.840 --> 00:40:44.920 you anywhere. So I think we should have some kind of hardwired local cutoff 562 00:40:45.599 --> 00:40:50.039 that you can't update from the internet. So anything that can be software updated 563 00:40:50.079 --> 00:40:53.519 from the internet obviously can be overridden. But if you have a local sort 564 00:40:53.559 --> 00:40:57.840 of off switch where you have to say a keyword or something, and then 565 00:40:57.920 --> 00:41:04.239 that puts the robot into a safe state, some kind of localized safe state 566 00:41:04.320 --> 00:41:09.239 ability, an off switch, you know, where you don't have to get 567 00:41:09.239 --> 00:41:13.719 too close to the robot. I don't know, if we've got millions of 568 00:41:13.719 --> 00:41:15.920 these things going over the place, you're not selling it, just you know, 569 00:41:17.320 --> 00:41:22.280 I know, I'm saying this is something we should be quite concerned about 570 00:41:22.800 --> 00:41:28.559 because Robert Robert can follow you anywhere. Then you know what if they just 571 00:41:28.639 --> 00:41:32.119 one day get a software update, and they're not so friendly anymore. We've 572 00:41:32.199 --> 00:41:37.599 got a James Cameron movie on our house. It's actually that's it's funny you're 573 00:41:37.599 --> 00:41:42.719 saying that because in our session that we had today, you know, just 574 00:41:42.960 --> 00:41:45.360 I would say who they made exactly the same point right then, as we're 575 00:41:45.400 --> 00:41:50.440 talking about they're talking about movies actually about mentioning James Cameron. They're talking about 576 00:41:50.480 --> 00:41:53.559 James cameeron movies, and they're saying, if you think about it's not just 577 00:41:53.679 --> 00:41:59.639 those movies, but any of these movies, trains, subways, metro. 578 00:42:00.599 --> 00:42:04.800 They said, all these movies with the same plot fundamentally all end with the 579 00:42:04.920 --> 00:42:08.000 person turning it off right or finding a way to shut the thing down. 580 00:42:08.199 --> 00:42:12.960 And they were making the same point that you were about the importance of actual 581 00:42:13.320 --> 00:42:16.199 physical off switches and so all the technology is great, but undamentally, the 582 00:42:16.280 --> 00:42:20.719 same movie has played out fifty times, we've all watched it, and all 583 00:42:20.840 --> 00:42:23.079 fundamentally, you know, you know, the point I'm revenged right or ends 584 00:42:23.119 --> 00:42:27.480 in pretty much the same way with someone finding a way to just do the 585 00:42:27.559 --> 00:42:30.760 wrong. Which is kind of interesting that you've said a similar point. Right, 586 00:42:30.280 --> 00:42:34.480 it's not the it's not the obvious place you'd go to, but be 587 00:42:34.519 --> 00:42:37.159 One of the tests for the AI was just so like, blank is your 588 00:42:37.199 --> 00:42:47.840 favorite Dreamers camera movie? Excellent? Right, Yes, we got over there. 589 00:42:47.920 --> 00:42:53.400 Yeah, perfect, Hi question for you both. So I'm a founder 590 00:42:53.480 --> 00:42:58.079 of a AI and mL scale up in the third center for AI, which 591 00:42:58.119 --> 00:43:01.039 is leads in the North of England a bit biased. Since the launch of 592 00:43:01.360 --> 00:43:06.920 chat GBT three months after that, we saw a real increase in phishing attacks 593 00:43:07.000 --> 00:43:13.400 using much more sophisticated language patterns. What do we do to protect businesses consumers 594 00:43:13.480 --> 00:43:16.119 that they trust this technology better and how do we bring them along that journey 595 00:43:16.159 --> 00:43:22.559 with us? Well, I think we shouldn't trust it that much. Actually, 596 00:43:25.239 --> 00:43:29.199 it is actually quite quite a significant challenge because we're getting to the point 597 00:43:29.199 --> 00:43:32.840 where even open source AI can pass human capture tests. So you know this, 598 00:43:34.119 --> 00:43:37.679 are you a human identify all the traffic lights in this picture? You're 599 00:43:37.719 --> 00:43:40.880 like, Okay, it's going to have no problem doing that. In fact, 600 00:43:40.920 --> 00:43:45.599 it'll do it better than the human and faster than human. So we're 601 00:43:45.679 --> 00:43:47.559 like, how do you you know at the point which is better a better 602 00:43:47.639 --> 00:43:52.039 human? Better passing human tests than humans? Then well, what tests actually 603 00:43:52.159 --> 00:43:55.119 make sense? That is a real problem. I don't actually have a good 604 00:43:55.119 --> 00:43:59.440 solution to it. That's one of the things we're trying to figure out on 605 00:43:59.519 --> 00:44:02.920 the XP form is how to deal with that, because it really we really 606 00:44:02.920 --> 00:44:08.679 are at the point where even with with open source, you know, readily 607 00:44:08.719 --> 00:44:12.159 available AI, you don't need to be sort of leading in the field. 608 00:44:12.960 --> 00:44:16.079 You can actually be better than humans at passing these tests. And that's sort 609 00:44:16.119 --> 00:44:20.280 of why we think, well, perhaps we should sort of charge a dollar 610 00:44:20.400 --> 00:44:22.679 or a pound a year. It's a very tiny amount of money, but 611 00:44:22.800 --> 00:44:25.599 it's it's still it still makes it privatively expensive to make a million bots. 612 00:44:28.599 --> 00:44:31.679 So and especially if you need a million payment methods, then you run out 613 00:44:31.719 --> 00:44:39.400 of sort of stolen credit cards pretty quickly. So that that's that's sort of 614 00:44:39.440 --> 00:44:42.880 where we're thinking, like we might have to sort of just charge some very 615 00:44:42.920 --> 00:44:49.679 tiny amount of moneyo point three cents a day effectively to deal with the the 616 00:44:49.760 --> 00:44:54.079 onsought of AI powered bonds. And if if and that that is that is 617 00:44:54.119 --> 00:44:59.639 spilit growing problem, but it will be I think perhaps an insurmountable problem next 618 00:44:59.719 --> 00:45:04.199 year. So and then you have to worry about, well, manipulation of 619 00:45:05.400 --> 00:45:08.599 information is making something seem very popular when in fact it it's not because it's 620 00:45:08.639 --> 00:45:15.039 getting boosted by all these likes and reposts from AI powered barts. So that's 621 00:45:15.039 --> 00:45:21.039 why I sort of think somewhat inevitably it leads to some small payment in order 622 00:45:21.119 --> 00:45:28.719 to dramatically increase the cost of a bart. So I think I frankly, 623 00:45:28.719 --> 00:45:30.679 I think probably any social media system it doesn't do that. Also believe you 624 00:45:30.719 --> 00:45:37.880 over on my bart. You know, I think my general answer would be, 625 00:45:37.280 --> 00:45:40.320 you know, we need to show that we are on top of mitigating 626 00:45:40.360 --> 00:45:45.400 the risks right so people can trust the technology. That's what actually the last 627 00:45:45.400 --> 00:45:49.159 couple of days has been about on the Safety Summit is just showing, you 628 00:45:49.239 --> 00:45:52.719 know, we're investing in the Safety Institute, having the people who can do 629 00:45:52.800 --> 00:45:55.719 the research on these things to figure out how we mitigate against them, and 630 00:45:55.880 --> 00:46:00.360 we have to do it fast, and we have to keep iterating it because 631 00:46:00.719 --> 00:46:02.559 all of us probably in this room, believe that the technology can be incredibly 632 00:46:02.639 --> 00:46:06.599 powerful, but we've got to make sure we bring people along that journey with 633 00:46:06.719 --> 00:46:09.519 us, that we're handling the risks that are there and as there's a job 634 00:46:09.599 --> 00:46:13.719 to do, and the last couple of days I think we make good progress 635 00:46:13.840 --> 00:46:16.119 on it. Because we want to focus on the positives and manage these things. 636 00:46:16.199 --> 00:46:20.599 But that requires action, and that's what the last couple of days has 637 00:46:20.599 --> 00:46:23.960 been about your your story, you know, analogy. There was part of 638 00:46:24.000 --> 00:46:29.639 the research that actually, you know, the team working on the task force 639 00:46:29.719 --> 00:46:32.599 here published and presented yesterday. I don't know if you saw it was, 640 00:46:34.280 --> 00:46:39.000 which is essentially that it was a using AI to do to create a ton 641 00:46:39.079 --> 00:46:45.920 of fake profiles on social media and then infiltrate particular groups with particular information. 642 00:46:46.480 --> 00:46:49.360 And actually, at the moment that, as I said to your point, 643 00:46:49.519 --> 00:46:52.280 there's a cost free it's getting. It's getting at the point where it's like, 644 00:46:52.599 --> 00:46:54.639 really, you're going to have one hundred for a penny sort of thing. 645 00:46:54.719 --> 00:46:58.280 Ridiculous. And if you think about some of these social networks that quite 646 00:46:58.280 --> 00:47:01.039 a neighborhood or town levels, not that many big profiles that you need to 647 00:47:01.119 --> 00:47:06.559 quickly create. Suddenly they're everywhere and there's some local issue that might be of 648 00:47:06.639 --> 00:47:09.400 importance, and you know, the team have run versions of how that would 649 00:47:09.440 --> 00:47:13.920 look like, and suddenly they're interacting with everybody and then spreading and misinformation around. 650 00:47:15.880 --> 00:47:19.480 Were literally as part of the research that we published on this information yesterday. 651 00:47:19.679 --> 00:47:22.719 It's a it's a real challenge. Yeah, exactly to your point. 652 00:47:22.760 --> 00:47:27.360 I mean the images. You don't even need to steal somebody's picture because then 653 00:47:27.519 --> 00:47:30.480 that's traceable. But you can actually just say, create a new image of 654 00:47:30.519 --> 00:47:37.639 a person realistic looking but doesn't exist, and then create a biography realistic but 655 00:47:37.719 --> 00:47:40.480 doesn't exist, and do that en mass and practically the only way of Bill 656 00:47:40.559 --> 00:47:46.519 tells that the grammar is too good. Don't give away typho. Come on 657 00:47:49.360 --> 00:47:51.880 now, I'm getting waved at because I think we are out of time. 658 00:47:51.920 --> 00:47:54.639 I don't we take one very brief last question and let's make it go on? 659 00:47:54.760 --> 00:47:57.119 Yes, sir, going, you're right in front of me, go 660 00:48:00.679 --> 00:48:05.440 elon. Question for you related to the X platform. Are there simple things 661 00:48:05.480 --> 00:48:07.480 we can do, especially when it comes to visual media. You alluded to 662 00:48:07.559 --> 00:48:13.239 the fact that it's fairly straightforward and effectively free to make people like yourselves say 663 00:48:13.559 --> 00:48:15.840 and do things that you never said or did. Can we do something like 664 00:48:15.880 --> 00:48:20.679 cryptographically signed media. I'm from Adobe, We're working on this project. Twitter 665 00:48:20.840 --> 00:48:23.719 was a member. Love to see X come back. Okay, digitally signed 666 00:48:23.760 --> 00:48:28.039 media to indicate not only what was created by AI, but what came from 667 00:48:28.079 --> 00:48:30.559 a camera, what was real to imbue a sense of trust in media that 668 00:48:30.719 --> 00:48:35.039 can go viral. That sounds like a good idea. Actually, so if 669 00:48:36.559 --> 00:48:43.079 some way of authenticating would be would be good. So, yeah, that 670 00:48:43.159 --> 00:48:46.159 sounds like a good idea, which we'll probably do it. Yeah, there 671 00:48:46.199 --> 00:48:52.280 you go, actually on that on that point. So I've already and this 672 00:48:52.400 --> 00:48:55.000 is particularly pertinent for people in my job, right, and I've already had 673 00:48:55.039 --> 00:49:00.719 a situation happened to me with adopted image that goes everywhere negative. By the 674 00:49:00.800 --> 00:49:04.400 time everyone realizes, well that's fake and we should stop sending it, the 675 00:49:04.519 --> 00:49:08.280 damages the damage is done. And actually we were again reflecting today. If 676 00:49:08.320 --> 00:49:13.119 you think next year, you've got elections in you know, I think you 677 00:49:13.159 --> 00:49:17.280 know, the US, India, I think Indonesia probably here there you go, 678 00:49:20.000 --> 00:49:24.880 massive news. Actually you've got just an enormous junk of the world's population 679 00:49:25.199 --> 00:49:29.679 is voting next year, right, and you've got EU elections as well. 680 00:49:30.320 --> 00:49:32.199 You know. Actually, just these issues are right in front of ours. 681 00:49:32.840 --> 00:49:37.400 Next year is one of big elections across the globe, probably the first set 682 00:49:37.440 --> 00:49:40.800 of elections where this has been a real issue. So figuring out how we 683 00:49:42.440 --> 00:49:45.840 manage that is I think kind of mission critical for the people who want, 684 00:49:45.880 --> 00:49:47.639 you know, the integrity of our democracy. Yeah, I mean some of 685 00:49:47.679 --> 00:49:51.480 it it's quite entertaining, Like the pope and the puffa jacket. Have you 686 00:49:51.559 --> 00:49:54.760 seen that one? That's amazing, But I mean I saw ritter people who 687 00:49:55.280 --> 00:49:59.159 think that's real. I'm like, well, one of the answer is wearing 688 00:49:59.159 --> 00:50:04.000 a puffa jacket and lie, you know, be sweating. But it actually 689 00:50:04.079 --> 00:50:08.079 looks quite quite dashing of things. In fact, I think AI fashion is 690 00:50:08.239 --> 00:50:12.159 going to be a real thing. So do in glue like, we live 691 00:50:12.199 --> 00:50:14.920 in the most interesting times, and I think this is it is, you 692 00:50:15.000 --> 00:50:17.719 know, eight like eighty percent likely to be good and twenty percent bad. 693 00:50:17.760 --> 00:50:22.840 I think if we're cognizant and careful about the bad part, on balance, 694 00:50:22.000 --> 00:50:27.079 actually it will be the future that we want. Well for the future that 695 00:50:27.360 --> 00:50:31.519 is preferable, and it actually will be some somewhat of a leveler, an 696 00:50:31.559 --> 00:50:36.440 equalizer in the sense that you know, every I think everyone will have access 697 00:50:36.440 --> 00:50:39.440 to goods and services and education, and so you know, I think probably 698 00:50:39.480 --> 00:50:44.639 it leads to more human happiness. So I guess I'd probably leave on an 699 00:50:44.679 --> 00:50:49.199 optimistic note. Tough act. Yeah, that's well, that's that is a 700 00:50:49.239 --> 00:50:52.000 great note to end on. I think that you know we're all onant that 701 00:50:52.159 --> 00:50:54.360 better future. I think it's there. The promise of it is certainly there. 702 00:50:54.519 --> 00:50:58.119 Lots of people in this room, including yourselves, are working hard to 703 00:50:58.159 --> 00:51:00.760 make it happen. Our job in government is to make sure happened safely. 704 00:51:00.920 --> 00:51:04.159 But on the basis of this conversation in the last couple of days, I'm 705 00:51:04.159 --> 00:51:07.199 certainly leaving more confident that we can make that happen. It's been a huge 706 00:51:07.199 --> 00:51:09.719 privilege and pleasure to have you here. Thank you very much for having me. 707 00:51:13.679 --> 00:51:14.280 I think it's