马斯克与苏纳克对话 AI

0:01Okay, all right. Well good evening everybody welcome Elon. Thanks for being here. Thank you for having me

0:07We feel we feel very proud. We're excited to have you right so I'm gonna start with some questions and then we're gonna open it up

0:12Let me get straight into it. So

0:15Bill Gates said there is no one in our time who has done more to push the bounds of science innovation than you

0:21Well, it's kind of insane. Yeah. Well, that's it. That's a nice thing to have anyone say about you nice coming from Bill Gates

0:26but

0:27Oddly enough, when it comes to AI, actually, for around a decade,

0:31you've almost been in the opposite and saying, hang on,

0:34you need to think about what we're doing and what we're pushing here

0:37and what do we do to make this safe and actually we shouldn't be pushing as

0:40faster as hard as we are.

0:42You've been doing it for a decade.

0:44What was it that caused you to think about it that way and why do we need to be worried?

0:49I've been somewhat of a Cassandra for quite a while.

0:52Where people would tell people, we should really be concerned about AI,

0:56like what are you talking about?

0:57Like, if never really had any experience with AI.

1:00Since I was immersed in technology,

1:03I have been immersed in technology for a long time,

1:05I could see it coming.

1:10But I think this year was,

1:11there have been a number of breakthroughs.

1:13I mean, you know, the point of which someone can see

1:16a dynamically created video of themselves.

1:20Like, it's already created a video of you saying anything

1:22in real time, or me.

1:25And so the sort of the deep peak videos, which are really incredibly good, in fact, sometimes

1:30we're convincing the real ones.

1:32And deep reel.

1:37And then obviously things like Chai GBT were quite remarkable.

1:43Now I saw GPT-1, GPT-2, GPT-3, GPT-4, the whole sort of lead up to that.

1:49So it was easy for me to kind of see where it's going.

1:53If you just sort of extrapolate the points on a curve,

1:56and I see that trend will continue,

1:57then we will have profound artificial intelligence,

2:01and obviously at a level that far exceeds human intelligence.

2:04So, but I'm glad to see at this point that people

2:10are taking safety seriously, and I'd like to say thank you

2:14for holding this AI safety conference.

2:16I think actually it will go out on in history

2:17as being very important.

2:19I think it's really quite profound.

2:23And I do think overall that the potential is there for

2:27artificial intelligence, AI, to have most likely a positive effect.

2:35And to create a future of abundance where there is no scarcity of goods and services.

2:41But it is somewhat of the magic genie problem,

2:45if you have a magic genie that can grant all the wishes.

2:49Usually those stories don't end well.

2:52Be careful what you wish for, including wishes.

2:56Yeah.

2:57So you talked a little bit about the summit,

3:01and thank you for being engaged in it, which has been great,

3:04and people enjoyed having you there.

3:06Potter's been in this dialogue.

3:08Now, one of the things that we achieved today

3:10in the meetings between the companies and the leaders

3:13was an agreement that externally, ideally governments,

3:18should be doing safety testing of models

3:21before they're released.

3:22I think this is something that you've spoken about a little bit.

3:24It was something we worked really hard on,

3:26because my job in government is to say, hang on,

3:29there is a potential risk here.

3:31Not a definite risk, but a potential risk of something

3:34that could be bad.

3:36My job is to protect the country.

3:38And we can only do that if we develop the capability

3:41we need in our safety institute, and then go in

3:43and make sure we can test the models before they are released.

3:46Delighted that that happened today,

3:47but what's your view on what we should be doing?

3:51You've talked about the potential risk, right?

3:52Again, we don't know.

3:53But what are the types of things governments like us

3:56should be doing to manage and mitigate against those risks?

3:59Well, I generally think that it is good for government play

4:03a role when the public safety is at risk.

4:06So really, for the vast majority of software,

4:10the public safety is not at risk.

4:11I mean, if the app crashes on your phone, you're a laptop,

4:15it's not a massive catastrophe.

4:18But when you're talking about digital superintelligence,

4:21I think, which does pose a risk to the public,

4:25then there is a role for government

4:26to play to safeguard the interests of the public.

4:29And this is, of course, true in many fields, aviation, cars.

4:35And I deal with regulators throughout the world

4:38because of starting communications, rockets being aerospace, and cars being transport.

4:46So I'm very familiar with dealing with regulators.

4:49And I actually agree with the vast majority of regulations.

4:50There's a few that I disagree with from time to time, but 0.1% probably.

4:55Less than 1% of regulations I disagree with.

4:59And there is some concern from people in Silicon Valley who have never dealt with regulators

5:04before and they think that this is going to just crush innovation and slow them down and

5:09be annoying.

5:11And it will be annoying, it's true.

5:15They're not wrong about that.

5:18But I think there's, we've learned over the years that having a referee is a good thing.

5:26And if you look at any sports game, there's always a referee and nobody's suggesting I

5:30think to have a sports game without one.

5:34And I think that's the right way to think about this is for

5:37government to be a referee to make sure the sportsman like conduct and

5:43that the public safety is addressed, that we care about public safety.

5:51Because I think there might be at times too much optimism about technology.

5:54And I say that as a technologist, I mean, so I ought to know.

5:57And on balance, I think the AI will be a forceful good, most likely, but the probability of

6:09it going bad is not zero percent.

6:11So we just need to mitigate the downside potential.

6:14And then how you talk about referee, and that's what we're talking about.

6:18Yeah.

6:19Well, there we go.

6:20We talk about this, and Demis and I discussed this a long time ago.

6:23They're literally facing right end of it.

6:27Actually, you know, Demis to his credit and the credit of people in the industry did say that to him.

6:30I think you would say it's not right that Demis and his colleagues are marking their own homework, right?

6:35There needs to be someone independent and that's why we've developed the Safety Institute here.

6:39I mean, do you think governments can develop the expertise?

6:42One of the things we need to do is they hang on, you know, Demis and all the others have got a lot of very smart people doing this.

6:49Governments need to quickly tool up capability-wise personnel-wise, which is what we're doing.

6:54I mean, do you think it is possible for governments to do that fast enough, given how quickly

6:58the technology is developing, or what do we need to do to make sure we do it quick enough?

7:03No, I think it's a great point you're making.

7:06The pace of AI is faster than any technology I've seen in history by far, and it seems to

7:14be growing in capability by at least five full, perhaps ten full per year.

7:19It will certainly grow by an order of magnitude next year.

7:23So, and government isn't used to moving at that speed.

7:30But I think even if there are not firm regulations,

7:35even if there isn't an enforcement capability,

7:38so we're having insight

7:39and being able to highlight concerns to the public

7:41will be very powerful.

7:44So, even if that's all that's accomplished,

7:47I think that will be very, very good.

7:48Okay, well hopefully we can do better than that.

7:50Hopefully, yeah.

7:51No, but that's how fast you know. We were talking before it was striking you know

7:55you're someone who spent their life in technology living Moore's law and

8:00What was interesting over the last couple of days talking to everyone who's doing the development of this?

8:05And I think you can cover this is it's just the pace of advancement here is unlike anything

8:11Yeah, all of you have seen in your careers and technology is that fair because you've got these kind of compounding

8:16effects from the hardware and and the data and the personnel

8:20Yeah, I mean, the two, currently the two leading centers for AI development are the San Francisco

8:29Bay Area and the sort of London area.

8:31And there are many other places where it's being done, but those are the two leading areas.

8:34So I think if the United States and the UK and China are sort of aligned on safety, that's

8:45will be a good thing.

8:47So that's where the leadership is generally.

8:50And actually, you mentioned China there.

8:53So I took a decision to invite China

8:55to submit over the last few days.

8:57And it was not an easy decision.

8:59A lot of people criticized me for it.

9:01My view is if you're going to try and have serious conversation,

9:04you need to.

9:04But what were your thoughts?

9:05You did business all around the world.

9:06You just talked about it there.

9:08Should we be engaging with them?

9:10Can we trust them?

9:11Is that the right thing to have done?

9:13If China is not on board with AI safety,

9:17it's somewhat of a mood situation.

9:20The single biggest objection that I get to in AI regulation

9:24or safety controls are, well, China's not going to do it

9:28and therefore they will just jump into the lead

9:30and exceed us all.

9:32But actually, China is willing to debate in AI safety.

9:38And thank you for inviting them.

9:40And they, you know, I think we should thank China for attending.

9:46When I was in China earlier this year, my main subject of discussion with the leadership

9:53in China was AI safety and saying that this is really something that they should care

9:57about and they took it seriously and URT, which is great.

10:04And having them here, I think, was essential really.

10:07If they're not participants, it's pointless.

10:10Yes, yes.

10:11And I think we were pleased.

10:12They were engaged yesterday in the discussions,

10:14and actually ended up signing the same communique

10:16that everyone else did, which is a good start.

10:19As we need everyone to approach this in a similar way,

10:22if we're going to have a realistic chance of resolving it.

10:26You talked about innovation earlier and regulation

10:29being annoying.

10:30There was a good debate today we had about open source.

10:33And I think you've kind of been a proponent

10:35of algorithmic transparency and making some of the X algorithms

10:39public.

10:40And actually, we were talking about every hinted on the way in.

10:44He's particularly been very concerned

10:46about open source models being used by bad actors.

10:50You've got a group of people who say they are critical to innovation

10:53happening in that distributed way.

10:56It's probably no perfect answer, and there is a tricky balance.

10:59What are your thoughts on how we should approach this open source question?

11:02or where should we be targeting whatever regulatory

11:06or monitoring that we're going to do?

11:09Well, the open source algorithms and data

11:14tend to lag the closed source by six to 12 months.

11:19But given the rate of improvement,

11:21that this actually therefore quite a big difference

11:24between the closed source and the open.

11:27If things are improving by a factor of the take-by

11:30or more than being a year behind is your five times worst.

11:34So it's a pretty big difference.

11:36And that might be actually an okay situation.

11:41But it certainly will get to the point

11:42where you've got open source AI that can do,

11:47that will start to approach human level intelligence

11:49will perhaps succeed it.

11:52I don't know quite what to do about it.

11:53I think it's somewhat inevitable.

11:55There'll be some amount of open source tonight.

11:57I guess I would have a slight bias towards open source,

12:01because at least you can see what's going on.

12:03It was closed source, you don't know what's going on.

12:05Now, it should be said with AI that even if it's an open

12:08source, do you actually know what's going on?

12:10Because if you've got a gigantic data file and sort of

12:17billions of data points or weights and parameters, you

12:22can't just read it and see what it's going to do.

12:24It's a gigantic file of inscrutable numbers.

12:28You can test it when you run it.

12:30You can test it.

12:31You can run a bunch of tests to see what it's going to do.

12:33But it's probabilistic as opposed to deterministic.

12:39It's not like traditional programming

12:41where you've got to sort of discrete logic.

12:45And the outcome is very predictable.

12:46And you can read each line and see what each line's going to do.

12:51a neural net is just a whole bunch of probabilities.

12:56I mean, it sort of ends up being a giant

12:58comma separated value file.

13:00It's like our digital guide is a CSV file, really?

13:02Hahaha.

13:03Hahaha.

13:04Okay.

13:07But that is kind of what it is.

13:09Yeah.

13:10Now, at that point you've just made,

13:12it's one that we have been talking about a lot

13:14because, again, conversation with the people

13:16that are developing their technology

13:17make the point that you've just made.

13:19It is not like normal software where there's predictability

13:22about inputs improving,

13:24leading to this particular output improving.

13:27And as the models iterate and improve,

13:29we don't quite know what's gonna come out the other end

13:32that Demis would agree with that,

13:33which is why I think there is this bias

13:35that we need to get in there

13:37while the training runs are being done,

13:38before the models are released to understand

13:40what is this new iteration brought about

13:43in terms of capability,

13:44which it sounds like you would agree with.

13:47I was going to shift gears a little bit.

13:49You've talked a lot about human consciousness, human

13:51agency, which actually might strike people as strange,

13:55given that you are known for being such a brilliant

13:58innovator in technologists.

13:59But it's quite heartfelt when I hear you talk about it and

14:02the importance of maintaining that agency in technology and

14:06preserving human consciousness.

14:08Now, the thing I was going to ask is, when I do interviews

14:11or talk to people out and about in this job, that AI, the

14:13thing that comes up most, actually, is probably not so

14:16much the stuff we've been talking about, but jobs.

14:18It's what does AI mean for my job?

14:21Is it going to mean that I don't have a job,

14:23or my kids are not going to have a job?

14:25Now, my answer as a policy maker, as a leader,

14:30is actually AI is already creating jobs.

14:32And you can see that in the companies that are starting.

14:36Also, the way it's being used is a little bit more

14:38as a co-pilot, necessarily versus replacing the person.

14:42There's still human agency, but it's helping you

14:44your job better, which is a good thing.

14:47And as we've seen with technological revolutions in the past, clearly there's change in the

14:51labor market in the amount of jobs that I was quoting an MIT study today that they did

14:56a couple of years ago, something like 60% of the jobs at that moment didn't exist 40 years

15:00ago, so hard to predict.

15:02And my job is to create an incredible education system, whether it's at school, whether it's

15:06retraining people at any point in their career, because ultimately if we've got a skilled population

15:10and they ought to keep up with the pace of change

15:13and have a good life.

15:14But it's still a concern.

15:16And what would your observation be on AI

15:21and the impact on labor markets and people's jobs

15:23and how they should feel about that as they think about this?

15:28Well, I think we are seeing the most disruptive force

15:31in history here.

15:34We have for the first time, we will

15:36have the first time something that is smarter than the smartest

15:39human. And that, I mean, it's hard to say exactly what that moment is, but there will come a

15:47point where no job is needed. You can have a job if you want to have a job for sort of

15:52personal satisfaction, but the AI will be able to do everything. So, I don't know if

16:00that makes people comfortable, uncomfortable. Well, it's, it's, you know, that's why, that's

16:06That's why I say if you wish for a magic genie, that gives you any wishes you want.

16:12And there's no limit.

16:13You don't have those three-wish limits, if not since you just have many wishes you want.

16:18So it's both good and bad.

16:22One of the challenges in the future will be how do we find meaning in life if you have

16:28a magic genie that can do everything you want?

16:30I do think it's hard when this new technology

16:34tends to usually follow an S curve.

16:38In this case, we're going to be on the exponential portion

16:40of the S curve for a long time.

16:44And you're really likely to be able to ask for anything.

16:48And we want to have universal basic income.

16:50We'll have universal high income.

16:52So in some sense, it'll be somewhat of a leveler

16:55or equalizer.

16:57Because really, I think everyone will have access

16:59this magic genie.

17:02And you're able to ask any question.

17:03It'll be certainly good for education.

17:05It'll be the best tutor.

17:06And then the most patient tutor.

17:09So they're all there.

17:14And there will be no shortage of goods and services.

17:17There will be an age of abundance.

17:19I think if I'd recommend people read in banks,

17:25the banks culture books are probably the best envisioning.

17:28In fact, not probably.

17:29They're definitely by far the best envisioning

17:30of an AI future.

17:32There's nothing you're in close.

17:34So I'd really recommend banks.

17:36I'm a very big fan.

17:39All these books are good.

17:41There's not every one, all of them.

17:45So that'll give you a sense of what is a fairly utopian

17:53or pro-topian future with AI.

17:59which is good from a, as you said,

18:01it's a universal high income, which is a nice phrase,

18:03and it's good from a materialistic sense,

18:05and your abundance, actually,

18:07that it then leads to the question that you pose, right?

18:10I'm somebody who believes work gives you meaning,

18:12like I think a lot about that,

18:14as I think work is a good thing,

18:16it gives people purpose in their lives,

18:17and if you then remove a large chunk of that,

18:21what does that mean, and where do you get that?

18:23Where do you get that drive, that motivation, that purpose?

18:26I mean, you were talking about it,

18:27you work a lot of hours.

18:28I was mentioning when we were talking earlier, I have to somewhat engage in deliberate suspension

18:34of disbelief because I'm putting so much blood, sweat and tears into a work project and burning

18:39the 3 AM oil.

18:41Then I'm like, wait, why am I doing this?

18:45I can just wait for the AI to do it.

18:48I'm just lashing myself for no reason.

18:51I'm just being glad for punishment or something.

18:55So we call them, call Demes and tell them to hurry up and then you can have a holiday, right?

19:00It's a plug.

19:02Yeah, no, it's a, look, it's a tricky thing because I think, you know, part of our job

19:06is to make sure that we can navigate to that very, I think, largely positive place that

19:11you're describing.

19:12It is like, and how much people throw it between now and then because these things bring a

19:16lot about a change in the labor market as we've seen.

19:20Yeah.

19:21I think it's probably is generally a good thing because, you know, there are a lot of

19:23jobs that are uncomfortable or dangerous or which sort of tedious. And the computer will

19:30have no problem doing that. They're happy to do that all day long. So, you know, it's

19:34fun to cook food, but it's not that fun to wash the dishes. And like, but the computer's

19:38perfectly happy to wash the dishes. I guess there is, you know, we still have sports,

19:46like where humans compete in like the Olympics. And obviously, a machine can go faster than

19:52any human, but we still have humans race against each other.

19:57And have these sports competitions against each other,

20:02where even though the machines are better,

20:03we're still, I guess, competing to see

20:05who can be the best human at something.

20:08And people do find fulfillment in that.

20:10So I guess that's perhaps a good example

20:12of how even when machines are faster than us stronger than us.

20:14We still find a way.

20:15We still enjoy competing against other humans,

20:18so at least the human is the best human.

20:20That's a good analogy.

20:22And we've been talking a lot about managing the risks.

20:25Before we move on and finish on AI,

20:27it's just talking a little bit about the opportunities.

20:31You're engaged in lots of different companies,

20:33neuro-looking and obvious one.

20:36Which is doing some exciting stuff.

20:38You touched on the thing that I'm probably most excited

20:40about, which is an education.

20:43And I think many people will have seen

20:44Sal Khan's video from earlier this year

20:47has talked about, as you talked about,

20:49it's like a personal tutor at Edge.

20:50an amazing personal tutor.

20:52An amazing personal tutor.

20:53And we know the difference in learning,

20:55having that personal tutor is incredible

20:57compared to class from learning.

20:58So if you can have every child have a personal tutor,

21:01specifically for them,

21:02that then just evolves with them over time.

21:04That could be extraordinary.

21:05And so that, for me, I look at it, I think,

21:07gosh, that is within reach at this point.

21:10And that's one of the benefits I'm most excited about.

21:13When you look at the landscape of things

21:15that you see as possible,

21:16what is it that you are particularly excited about?

21:19I think certainly AI tutors are going to be amazing.

21:25Perhaps already are.

21:26I think there's also perhaps companionship, which may seem odd, because how can the computer

21:31really be your friend.

21:32But if you have an AI that has memory and remembers all of your interactions and has

21:38read everything, you can say like give it permission to read everything you've ever

21:42done.

21:43So it really will know you better than anyone, perhaps even yourself.

21:47where you can talk to it every day and those conversations

21:49hold upon each other, you will actually have a great friend.

21:54As long as that friend can stay, your friend

21:56did not get turned off or something.

21:58Don't turn off my friends.

22:00But I think that will actually be a real thing.

22:06And one of my sons has some learning disabilities

22:12and has trouble making friends actually.

22:14And I was like, well, you know, he,

22:16the AI friend would actually be grateful.

22:18Oh, okay.

22:19Yeah, that was a surprising answer,

22:21but that's actually worth reflecting on.

22:24So it's really interesting.

22:25I mean, we're already seeing it actually as we deliver,

22:28you know, psychotherapy anyway,

22:30now doing far more by digitally and by telephone to people.

22:34And it's making a huge difference,

22:35and you can see a world in which actually,

22:37you know, AI can provide that social benefit to people.

22:40And just a quick question on X, and then we should open it up to everybody.

22:46You made a change when you made many changes.

22:49But one of the changes-

22:50Yeah, quite a few.

22:51You love that letter.

22:52Yeah.

22:53I've got a real thing about it.

22:56You really do.

22:57You really do.

22:58One of the changes, which kind of, it goes into the space that we have to operate in,

23:03and this balance between free speech and moderation is, you know, we grapple with as politicians.

23:09you were grappling with your inversion of that.

23:11And you moved away from a manual human way of doing it,

23:17the moderation to the community notes.

23:20And I think it was an interesting change.

23:23It's not what everyone else has done.

23:25What was the reasoning behind that?

23:27Why do you think that is a better way to do that?

23:31Yeah, part of the problem is if you're in power people as sensors,

23:35then there's going to be some amount of bias.

23:39they have, and then whoever points the senses is effectively in control of information.

23:44So then the idea behind theory notes is how do we have a consensus driven, I mean, so

23:52it's not really censoring it, but consensus driven approach to truth.

23:55How do we make things the least amount untrue?

24:00You can say like, you can't perhaps get to pure truth, but you can aspire to be more

24:06truthful.

24:07So the thing about community notes is it doesn't actually

24:11delete anything, it simply adds context.

24:13Now that context could be, this thing is untrue

24:15for the following reasons.

24:16But importantly with community notes, everything is open source,

24:23actually, so you can see the software, every line of the software,

24:28you can see all of the data that went into a community note

24:31and you can independently create that community note.

24:34So if you've got, if you see manipulation of the data,

24:36You can actually highlight that and say, well, this,

24:39there appears to be some gaming of the system.

24:43And you can suggest improvements.

24:45So it's maximum transparency, which is combined with the

24:50kind of wisdom of the crowds and transparency to get to a

24:53better answer.

24:54And really, one of the key elements of community notes is

24:57that in order for a note to be shown, people who have

24:59historically disagreed must agree.

25:03And there is a bit of AI usage here.

25:05So this will populate a parameter space around each contributed

25:11to community notes.

25:12And then a parameter space.

25:15So everyone's got basically these vectors associated with them.

25:19Which is not as simple as Rijolift.

25:21It's saying it's several hundred vectors

25:25because things are more complicated than something

25:27right or left.

25:28And then we'll do sort of inverse correlation.

25:34say like, okay, these people generally disagree,

25:37but they agree about this note.

25:39And so then that gives the note credibility.

25:43Yeah, that's the core of it.

25:45And it's working quite well.

25:47I get to see a note actually be present

25:51for more than a few hours that is incorrect.

25:53So the batting average is extremely good.

25:55And when I ask people, so they're worried

25:57about community notes sort of being disinformation

26:00like send me one, and then they can't.

26:02So I think it's quite good.

26:05I mean, the general aspiration is what the X platform is to inform

26:10and entertain the public and to be as accurate as possible

26:14and as truthful as possible, even if someone doesn't like the truth.

26:18You know, people that always like the truth.

26:20No.

26:21It's not always.

26:24But that's the aspiration.

26:26And I think if we stay true to the truth, then I think we'll find

26:30that people use the system to learn what is going on and to learn. I think actually truth

26:38pays. So I think it will be, assuming you don't want to engage in self-tolution, then

26:45I think it's the smart move.

26:50Excellent. Very helpful. Right. Let's open it up to all our guests here. We've got some

26:55microphones, they'll come up, they'll come and find you. We've got yes, go for it.

27:00Thank you. Good evening, Alice Bentink from Entrepreneur First. Thank you for

27:06a fascinating conversation. I suppose a question for each of you.

27:10Prime Minister, the UK has some of the best universities in the world. We have the

27:14talent. What will it take for the UK to be a real breeding ground for unicorn

27:20companies? And Elon, being a founder in the UK is still a non-obvious career

27:26choice for the most exceptional technical talent.

27:28What are the cultural elements that we need to put into place

27:31to change this?

27:32Thank you both.

27:33Well, you want to go first?

27:34Go for it.

27:35Sure.

27:36You're right that there are cultural elements where the

27:42culture should celebrate creating new companies.

27:48And there should be a bias towards supporting small

27:53companies, because they're the ones that need nurturing the

27:55larger companies really don't need nurturing.

27:59So you can think of it sort of like a God,

28:01and if it's a little sprout that needs nurturing,

28:04if it's a mighty oak, it doesn't need quite as much.

28:07So I think that is a mindset change that is important.

28:12But I should mention that London and San Francisco

28:19or the Bay Area are really the two senses for AI.

28:22So London is actually doing very well on that front.

28:26The two leading locations on a,

28:29you know, San Francisco is probably head of London,

28:33but London's really very strong.

28:35Or London area, greater London, home counties, I guess.

28:38I'll keep going, keep going.

28:41So I'm just saying objectively this is the case.

28:48But you do need that, you need the infrastructure,

28:51you need landlords who are willing to rent to new companies,

28:55you need law firms and accountants that are willing

28:58to support new companies.

28:59And it's generally a mind, it is a mindset change.

29:03And I think some of that is happening,

29:04but I think really it's just culturally

29:06people need to decide this is a good thing.

29:09Yeah.

29:10No, actually, well, thanks for what you said about the UK.

29:12It's something that we work hard on,

29:14lots of people in the room are part of what makes this

29:17a fabulous place for any of the companies,

29:18including Alice.

29:20So, Alice White says, my job is to get all the nuts and bolts right, make sure that all

29:26of you starting companies can raise a capital that you need, everything from your seed funding

29:31with our incredible EIS tax reliefs all the way through to your late stage rounds and we

29:38need reform of our pension funds and the Chancellor's got a bunch of incredible reforms to unlock

29:43capital from all the people who have it and deploy it into growth equity, right?

29:47That is a work in progress.

29:48We're not there yet.

29:49I think we're making good progress.

29:51We need talent, we need people.

29:53So that means an education system that

29:55prioritizes the things that matter.

29:57And you've seen my reforms.

29:58I go on about more maths, more maths, more maths.

30:01But I think that is important.

30:03But also attracting the best and the brightest here.

30:05If you look at our fastest growing companies in this country,

30:08and I think it's probably the same in the US,

30:11over half of them have a non-British founder.

30:15And that tells you we've got to be a place that is open to the world's

30:18best and brightest entrepreneurial talent.

30:20So the visa regime that we've put in place, I think,

30:23does that.

30:24Makes it easy for those people to come here.

30:25And then actually, it's the thing that we spent

30:27the beginning of the session talking about, the regulation.

30:30Making sure that we've got a regulatory system that's

30:32pro-innovation, that, yeah, of course, we always need

30:35guardrails on the things that we'll worry us, but we've

30:38got to create a space for people to innovate and do

30:41different things.

30:42Those are all my jobs.

30:43The thing that is tougher is the thing that Elon talked

30:46about, which is culture.

30:47is how do you transpose that culture from places

30:50like Silicon Valley across the world

30:52where people are unafraid to give up the security

30:56of a regular paycheck to go and start something

30:59and be comfortable with failure.

31:01You talk about that a lot.

31:02I think you talk about it more in when you're playing games,

31:04right?

31:05But you've got to be comfortable failing

31:07and knowing that that's just part of the process.

31:09And that is a tricky cultural thing to do overnight,

31:12but it's an important part of, I think,

31:14creating that kind of environment.

31:16Yeah, if you don't succeed with your first startup,

31:18it shouldn't be a catastrophic career ending.

31:20Exactly.

31:21It should be, well, good.

31:23I think you're doing it.

31:25It should be like, well, you gave it a good shot.

31:27And now try again.

31:29Exactly.

31:30Yeah.

31:31So one thing I was going to mention is,

31:33obviously, creating a company is sort of a high risk, high

31:35reward situation.

31:38But I don't know quite what this works in the UK.

31:42I think it's probably better than Continental Europe.

31:45But the stock options are very difficult in most parts.

31:50You're not sure how to do it.

31:52But if somebody's basically going to risk

31:54the sort of life savings and the vast majority of startups

31:58fail.

31:59So I mean, you hear about the startups that succeed.

32:02But most companies are, most startups

32:05consist of a massive amount of work followed by failure.

32:11That's actually most companies.

32:13And so it's a high risk higher award.

32:15And so the higher award part does need to be there

32:18for it to make a sense.

32:19I think that was a very soft pitch for a tax policy.

32:23I mean, Charles, but actually I can tell you.

32:25So look, AI agree.

32:27So we have, I think, relative to certainly European countries,

32:31but certainly the US is definitely California,

32:34a much lower rate of capital gains tax.

32:36So for those people who are risking and growing something,

32:39we think the reward should be there at the end.

32:41So 20% capital gains tax rate.

32:44And on stock options, I don't know if we've got anyone

32:46from index ventures in the room.

32:49So index one of our leading VC funds here.

32:52They do a regular report looking at most countries' tax

32:57treatment of stock options.

32:59And when I was chancellor at Treasury Secretary equivalent,

33:03we were, I think, down at, we were pretty good,

33:06but we were fourth or fifth.

33:07And I said, we need to, for exactly the reason

33:09that you mentioned, I was like, this has got to be the best

33:11place for innovators.

33:12We need to move that up.

33:13And I think in the last iteration of that report,

33:15we had, because of the changes that Jeremy and I have made,

33:19we have moved up to, I think, second from memory.

33:22So I hope that actually give you and everyone else

33:24some comfort that we recognize that's important.

33:26Because when people work hard and risk things,

33:28yeah, they should be able to enjoy the rewards of that.

33:30I respect the rewards.

33:31And I think we very much have a tax system that supports that.

33:35And those are the values that I believe in,

33:37and I think most of us in this room probably do as well.

33:39Right, next question.

33:41I've got seven in front of me, and then I'll come over here.

33:43John, John, sir.

33:48Thanks very much.

33:49We've talked about some really big ideas,

33:52global changing ideas.

33:53I'm really interested, particularly in the context

33:55of creation of science and technology super hubs and so on,

33:59how does that map onto the everyday lives

34:01of people living in, say, Austin, Texas,

34:04to choose one around you more, in my case,

34:06Nottingham, East Midlands.

34:09How do you see that evolving for people every day?

34:12be sort of everyday effects of AI?

34:15Yeah.

34:18For context, Elon, so Seb runs our equivalent of CVS,

34:23right, or Walgreens, so, you know, when actually I visited,

34:26right, so he's got millions of people coming in

34:27in his shops every day and it's making sure,

34:29how do we make this relevant, I think Seb is your question,

34:32how is this relevant to that person?

34:34You know, actually, let me go first on that,

34:36because I think it's a fair point.

34:39I was just going over with the team,

34:41a couple of things that we're doing,

34:43because I was saying like, how are we doing AI right now

34:45that it's making a difference to people's lives?

34:48And we have this thing called Gov.UK,

34:50which is when it happened several years ago

34:53was a pioneering thing.

34:55All the government information brought together

34:57on one website, Gov.UK.

34:59So you need to get a driving license, passport,

35:02any interaction with government, it was centralized

35:04in a very relatively easy to use way.

35:07It's better than most, better than most.

35:10Yeah, so we are about to deploy

35:13AI across that platform. So that is something that I think several million people a day use,

35:19right? So a large chunk of the population is interacting with gov.uk every single day to

35:24do all these day-to-day tasks, right? Every one of your customers is doing all those things.

35:29And so we're about to deploy AI into that to make that whole process so much easier. Because

35:35some people will be like, well, I'm currently here and I've lost my passport and my flights

35:40in five hours, at the moment that would require how many steps to figure out what you do.

35:46Actually, when we deploy the AI, it should be that you could just literally say that

35:50and boom, boom, boom, boom, boom.

35:51This is what we're going to do, walk you through it.

35:53And that's going to benefit millions and millions of people every single day, right?

35:56Because that's a very practical way in my seat that I can start using this technology

36:01to help people in their day-to-day lives, not just healthcare discoveries and everything

36:05else that we're also doing.

36:07I thought that's quite a powerful demonstration of literally your day-to-day customer

36:11seeing actually their just day-to-day life get a little bit easier because of something

36:15that Elon Demis and others in this room have helped create.

36:18Yeah, exactly. The most immediate thing is just being able to ask, like having a very

36:23smart friend that you can ask anything, you know, ask how to make something, how to solve

36:29any problem and it'll tell you.

36:35And obviously companies are going to adopt this.

36:37So I think we'll have much better customer service, I guess,

36:40essentially.

36:41That'll probably be the first thing you notice.

36:44And then we talked about education.

36:50So having a tutor, so if you're trying to understand a

36:54subject like having a phenomenal tutor on any subject,

36:59that's really pretty much there already almost.

37:03And we need to, obviously AI needs to stop

37:05hallucinating before, you know, it can't give you,

37:08I mean, we still have a little bit of a problem

37:10where it can give you an answer that's

37:12confidently wrong with great grammar.

37:15And, you know, in a bullet points and everything in

37:18citations it was not real.

37:20So that's to be okay, we need to make sure it's,

37:24It's not giving you confidently wrong tutor answers.

37:28But that's going to happen pretty quickly,

37:30where it is actually correct.

37:34So yeah.

37:34I'm going to say for any parent who

37:36was homeschooling during COVID and realizing

37:39what their kids needed to be helped with,

37:41that will come as an enormous relief, I think.

37:43Yeah.

37:43Very, very good.

37:44Right.

37:45And we've got quite let's go questions over here.

37:47We've got, we've got any microphones or Brent, are you there?

37:50Perfect.

37:51Hi, Brent Hoebman.

37:53So you've spoken eloquently about abundance

37:56and the age of abundance.

37:57So it feels obviously with AI, it's everything everywhere,

38:00everywhere all at once.

38:02But with robots, and to get the age of abundance,

38:06we'll need a lot of robots.

38:07I know you're working on robots as well.

38:10Are there sort of constraints that we should think of

38:12and our politicians should be thinking of,

38:14that we might get, one country might get heavily behind

38:16in robots that can do all these things

38:19and enter the age of abundance,

38:19and therefore be at a strategic disadvantage?

38:24Well, really anything that can be actuated by a computer

38:29is effectively a robot.

38:31So you can think of, frankly, Tesla

38:33cause our robots on wheels.

38:37Anything that's connected to the internet

38:39is effectively an endpoint actuator

38:41for artificial intelligence.

38:44So you've got Boston Dynamics.

38:49Obviously, they've been making impressive robots for a while.

38:53I think they're at this point mostly owned by Hyundai.

38:55So I guess who is probably going to make robots

39:00that are humanoid and some rather interesting shapes

39:03that I wasn't just making the one that looks like it

39:06has wheels and looks sort of like a kangaroo on wheels.

39:08I'm not sure what that is.

39:14Looks a little demented, frankly.

39:17But there's going to be all sorts of robots.

39:20You have a company Dyson in the UK, which I think is some pretty impressive things.

39:28I think the UK will not be behind, actually, on that front.

39:32UK also has Arm, which is really the best, one of the best, perhaps the best in chip design

39:42in the world.

39:43It tells they use a lot of Arm technology, almost everyone does, actually.

39:49So I think UK is in a strong position.

39:53Germany obviously makes a lot of robots,

39:55industrial robots.

39:57I mean, I think generally countries

39:59that make robots of any kind, even if they seem somewhat

40:03conventional, will be fine.

40:08I do think there is a safety concern,

40:12especially with humanoid robots, because at least the car

40:16can't chase you into this building, not very easily.

40:19or chase you off a tree, or you can sort of run up a flight

40:23of stairs and get away from a Tesla.

40:28I think it was a Stephen King movie about that.

40:31A few targets possessed.

40:34But if you have a humanoid robot,

40:35it can basically chase you anywhere.

40:38So I think we should have some kind of hardwired local cut-off

40:45that you can't update from the internet.

40:48So anything that can be software updated from the internet,

40:50obviously, can be overridden.

40:52But if you have a local sort of off switch

40:55where you have say a keyword or something

40:57and then that puts the robot into a safe state,

41:01some kind of localized safe state ability,

41:06an off switch where you don't have to get too close

41:10to the robot.

41:11I don't know, so if you've got millions of these things

41:13going all over the place.

41:15You're not selling it.

41:16Just, you know, like so.

41:17I know. I'm saying, is this something we should be quite concerned about? Because Robert can

41:24follow you anywhere, then what if they just one day get a software update and they're

41:30not so friendly anymore? We've got a James Cameron movie on our hands.

41:37It's actually, it's funny you saying that because in our session that we had today,

41:42I would say, they made exactly the same point, right? They were talking about, they're talking

41:47talking about movies, actually, about mentioning James Cameron, they're talking about James

41:50Cameron movies.

41:51And so they're saying, if you think about it, it's not just those movies, but any of these

41:56movies trains, subways, metros, or as buses, they said all these movies with the same plot

42:03fundamentally all end with the person turning it off, right, or finding a way to shut the

42:08thing down and they were making the same point that you were about the importance of

42:12actual physical off switches.

42:14is. All the technology is great, but fundamentally this same movie has played out 50 times. We've

42:19all watched it and all fundamentally, you know, point of view, right? Or ends in pretty

42:24much the same way with someone finding a way to just do the thing. Which is kind of interesting

42:29that you've said a similar point, right? It's not the obvious place you'd go to.

42:33Maybe that could be one of the tests for the AI. We should say like, blank is your favorite

42:38of the James Cameron movie, film in the blank.

42:40Yeah, excellent, right.

42:46Yes, we got over there, yep, perfect.

42:51Hi, a question for you both.

42:53So I'm a founder of AI and ML scale up

42:56in the third centre for AI,

42:57which is Leeds in the North England, a bit biased.

43:00Since the launch of chat GPT, three months after that,

43:04we saw a real increase in phishing attacks

43:06using much more sophisticated language patterns.

43:10What do we do to protect businesses, consumers,

43:13that they trust this technology better?

43:15And how do we bring them along that journey with us?

43:20Well, I think we shouldn't trust it that much, actually.

43:25It is actually quite a significant challenge,

43:28because we're going to the point where even open source AI

43:30can pass human capture tests.

43:33So this is, are you a human?

43:35Identify all the traffic lights in this picture.

43:37like, okay, yeah, it's going to have no problem doing that. In fact, it'll do it better than

43:42the human and faster than the human. So we're like, how do you, you know, it's the point

43:46of which is better, a better human, better passing human tests than humans, then, well,

43:52what tests actually make sense? That is a real problem. I don't actually have a good solution

43:56to it. That's one of the things we're trying to figure out on the X platform is how to

44:02deal with that because it really are at the point where even with open source, you know,

44:09readily available AI, you don't need to be sort of leading the field, you can actually

44:13be better than humans at passing these tests. And that's sort of why we think, well, perhaps

44:18we should sort of charge a dollar or a pound a year. It's a very tiny amount of money, but

44:22it still makes it privately expensive to make a million bots. So, and especially if you

44:30a million payment methods, then you run out of sort of stolen credit cards pretty quickly.

44:38So that's sort of where we're thinking like we might have to sort of just charge some

44:42very tiny amount of money, 0.3 cents a day effectively, to deal with the onslaught of

44:50AI-powered bonds.

44:52And that is still a growing problem, but it will be I think perhaps an insurmountable

44:58problem next year.

44:59And then you have to worry about manipulation of information.

45:06Making something seem very popular when it's not because it's getting boosted by all these

45:11likes and repos from AI powered bots.

45:15So that's why I sort of think somewhat inevitably it leads to some small payment in order to

45:22dramatically increase the cost of a bot.

45:26So I think frankly, I think probably any social media system that doesn't do that will simply

45:31be overrun by bots.

45:34I think my general answer would be we need to show that we are on top of mitigating the

45:41risk, so people can trust the technology.

45:44That's what actually the last couple of days has been about on the safety summit is

45:48just showing we're investing in the safety institute, having the people who can do the

45:52research on these things to figure out how we mitigate against them, and we have to do

45:57it fast and we have to keep iterating it. Because all of us probably in this room believe that

46:02the technology can be incredibly powerful, but we've got to make sure we bring people along

46:05that journey with us that we're handling the risks that are there. And there's a job to

46:10do in the last couple of days. I think we make good progress on it because we want to focus

46:14on the positives and manage these things. But that requires action and that's what the last

46:19couple of days have been about. Your analogy there was part of the research that actually

46:26the team working on the task force here,

46:30published and presented yesterday,

46:32I don't know if you saw it,

46:33was which is essentially that.

46:35It was a using AI to create a ton of fake profiles

46:41on social media and then infiltrate particular groups

46:45with particular information.

46:47And at the moment, that is, as I said to your point,

46:49I know it's like cost free.

46:50It's getting to the point where it's like,

46:53really you're gonna have a hundred for a penny

46:55sort of thing, ridiculous.

46:56And if you think about some of these social networks

46:57at what a neighborhood or town level,

46:59it's not that many big profiles that you

47:01quickly create, suddenly they're everywhere,

47:04and there's some local issue that might be of importance,

47:07and the team have run versions of how that looked like,

47:10and suddenly they're interacting with everybody,

47:12and then spreading misinformation around.

47:13Yeah, that's a real challenge.

47:15We literally as part of the research

47:17that we published on misinformation yesterday,

47:19it's a real challenge.

47:21Yeah, exactly, a two point.

47:22I mean, the images, you don't even need to steal

47:25somebody's picture, because that's traceable, but you can actually just say create a new

47:30image of a person, realistic looking, it doesn't exist, and then create a biography,

47:36realistic, but doesn't exist, and do that en masse.

47:40And practically the only way a bill tells us that the grammar is too good.

47:44Take it away.

47:45So typos, come on.

47:47Now I'm getting waved at, because I think we are out of time.

47:51I don't think we take one very brief last question.

47:54And let's make a good one.

47:55Yes, sir.

47:56Go on your own front and be calm.

48:00Elon, question for you.

48:03Related to the X platform.

48:04Are there simple things we can do,

48:06especially when it comes to visual media?

48:07You alluded to the fact that it's fairly straightforward

48:10and effectively free to make people like yourselves say

48:13and do things that you never said or did.

48:16Can we do something like cryptographically signed media?

48:18I'm from Adobe.

48:18We're working on this project.

48:20Twitter was a member.

48:21Love to see X come back.

48:23digitally signed media to indicate not only what was created by AI, but what came from

48:27a camera, what was real to imbue a sense of trust in media that can go viral.

48:32That's something a good idea actually.

48:36So some way of dedicating would be good.

48:40So yeah, that sounds like a good idea, I wish we could probably do it.

48:45Yeah.

48:46There you go.

48:47Actually, on that point, I've already, so this is particularly pertinent for people

48:53in my job, right? And I've already had a situation happen to me with a doctored image

48:58that goes everywhere, negative. By the time everyone realizes, well, that's fake and we

49:03should stop sending it, the damage is done. And actually, we were, again, reflecting today.

49:09If you think next year, you've got elections in, you know, I think, you know, the U.S.,

49:13India, I think Indonesia, probably here. There you go. Massive news. And actually, you've

49:23You've got just an enormous chunk of the world's population is voting next year.

49:28You got EU elections as well.

49:30These issues are right in front of us.

49:33Next year is where big elections across the globe, probably the first set of elections

49:37where this has been a real issue.

49:40So figuring out how we manage that is, I think, kind of mission critical for people

49:45who want the integrity of our democracy.

49:47Yeah.

49:48I mean, some of it is quite interesting.

49:50Like the Pope in the Papa jacket, have you seen that one?

49:52That's amazing.

49:54But I mean, I still run into people who think that's real.

49:57I'm like, well, one of the odds he's

49:58wearing a puffer jacket in July in Rome.

50:02He'd be sweating.

50:03But it actually looked quite dashing, I think.

50:07In fact, I think AI fashion is going to be a real thing.

50:09So I've been doing it clearly.

50:11We live in the most interesting times.

50:12And I think this is 80% likely to be good.

50:17And 20% bad.

50:18I think if we're cognizant and careful about the bad part,

50:22on balance actually, it will be the future that we want.

50:25Or the future that is preferable.

50:28And it actually will be somewhat of a leveler,

50:32an equalizer in the sense that,

50:35I think everyone will have access to goods and services

50:37and education.

50:38And so, I think probably it leads to more human happiness.

50:42So I guess I'd probably leave on an optimistic note.

50:45Yeah.

50:46Yeah.

50:47Well, that is a great note to end on.

50:50I think we're all on that better future.

50:53I think it's there.

50:54The promise of it is certainly there.

50:55Lots of people in this room, including yourselves,

50:57are working hard to make it happen.

50:58Our job in government is to make sure it happens safely.

51:00But on the basis of this conversation

51:02in the last couple of days,

51:03I'm certainly leaving more confident

51:05that we can make that happen.

51:07It's been a huge privilege and a pleasure to have you here.

51:09Well, thank you very much for having me.

51:10Thank you.

51:11Thank you.

51:12Thank you.

51:13I think previously...