1
00:00:00,160 --> 00:00:02,240
Speaker 1: Latest interview of Elon Musk.

2
00:00:02,120 --> 00:00:04,639
Speaker 2: Bill Gates said there is no one in our time

3
00:00:04,679 --> 00:00:07,280
who has done more to push the bounds of science.

4
00:00:06,960 --> 00:00:07,879
Speaker 3: Innovation than you.

5
00:00:08,119 --> 00:00:10,160
Speaker 2: What's kind of going to say, well, that's a nice

6
00:00:10,160 --> 00:00:12,080
thing to have anyone say about you. Nice coming from

7
00:00:12,080 --> 00:00:15,560
Bill Gates. But oddly enough, when it comes to AI, actually,

8
00:00:15,640 --> 00:00:18,600
for around a decade, you've almost been doing the opposite

9
00:00:18,679 --> 00:00:21,199
and saying, hang on, you need to think about what

10
00:00:21,199 --> 00:00:23,199
we're doing and what we're pushing here, and what do

11
00:00:23,239 --> 00:00:25,199
we do to make this safe? And actually maybe we

12
00:00:25,199 --> 00:00:27,879
shouldn't be pushing as faster or as hard as we are.

13
00:00:28,480 --> 00:00:29,920
I mean, you've been doing it for a decade, Like

14
00:00:30,000 --> 00:00:32,280
what was it that caused you to think about it

15
00:00:32,280 --> 00:00:34,079
that way? And you know, why do we need to

16
00:00:34,119 --> 00:00:34,799
be worried?

17
00:00:34,880 --> 00:00:36,799
Speaker 4: Yeah? I've been somewhat of a Cassandra for quite a

18
00:00:36,880 --> 00:00:39,119
while where people would I tell you, like, we should

19
00:00:39,119 --> 00:00:40,399
really be concerned about AI, and the'd be like, what

20
00:00:40,399 --> 00:00:42,119
are you talking about? Like I've never really had any

21
00:00:42,119 --> 00:00:46,079
experience with it since I was immersed in technology. I

22
00:00:46,079 --> 00:00:48,320
have been immersed in technology for a long time. I

23
00:00:48,320 --> 00:00:50,840
could see it coming, so, but I think this year

24
00:00:51,039 --> 00:00:53,280
was that there's been a number of breakthroughs. I mean,

25
00:00:53,719 --> 00:00:55,240
you know, at the point in which someone can see

26
00:00:55,359 --> 00:00:58,880
a dynamically created video of themselves, you know, like so

27
00:00:58,960 --> 00:01:00,840
you can make a video of you say anything in

28
00:01:00,880 --> 00:01:03,159
real time or me, and so there's sort of the

29
00:01:03,399 --> 00:01:05,879
deep videos, which are really incredibly good, in fact, sometimes

30
00:01:05,920 --> 00:01:08,959
more convincing than real ones and deep real and then

31
00:01:09,040 --> 00:01:12,560
and then obviously things like chat GPT were quite remarkable,

32
00:01:12,560 --> 00:01:15,480
and now I saw a GPT one, GPT two, GPT three,

33
00:01:15,560 --> 00:01:17,680
GPD four that you know, the whole sort of lead

34
00:01:17,760 --> 00:01:20,159
up to that, So it was easy for me to

35
00:01:20,239 --> 00:01:21,920
kind of see where it's going. If you just sort

36
00:01:21,959 --> 00:01:24,480
of extrapolate the points on a curve and as seeing

37
00:01:24,480 --> 00:01:28,359
that tread will continue, then we will have profound artificial intelligence,

38
00:01:28,719 --> 00:01:31,319
and obviously at the level that far exceeds human intelligence.

39
00:01:31,560 --> 00:01:34,120
So but I'd see at this point that people are

40
00:01:34,159 --> 00:01:36,879
taking safety seriously, and I like, I say thank you

41
00:01:36,920 --> 00:01:39,239
for holding this AI safety conference. I think actually it

42
00:01:39,239 --> 00:01:41,159
will go down in history as being very important. I

43
00:01:41,200 --> 00:01:44,439
think it's really quite profound, and I do think overall

44
00:01:44,439 --> 00:01:47,480
that the potential is there for artificial intelligence AI to

45
00:01:48,079 --> 00:01:51,359
have most likely a positive effect and to create a

46
00:01:51,400 --> 00:01:54,120
future of abundance where there is no scarcity of goods

47
00:01:54,159 --> 00:01:57,239
and services. But it is somewhat of the magic gene

48
00:01:57,359 --> 00:01:59,719
problem where if you have a magic gene that can

49
00:01:59,719 --> 00:02:02,959
grab all the wishes, usually those stories don't end well.

50
00:02:03,079 --> 00:02:05,280
Who careful what you wish for? Including wishes?

51
00:02:05,400 --> 00:02:07,879
Speaker 2: So you talked a little bit about the summit and

52
00:02:08,199 --> 00:02:10,280
thank you for being engaged in it, which has been

53
00:02:10,319 --> 00:02:13,800
great and people enjoyed having you there pulling in this dialogue. Now,

54
00:02:13,800 --> 00:02:16,240
one of the things that we achieved today in the

55
00:02:16,280 --> 00:02:20,960
meetings between the companies and the leaders was an agreement

56
00:02:21,080 --> 00:02:25,840
that externally, ideally, governments should be doing safety testing of

57
00:02:26,000 --> 00:02:28,280
models before their relief. I think this is something that

58
00:02:28,319 --> 00:02:30,000
you've spoken about a little bit. It was something we

59
00:02:30,039 --> 00:02:32,639
worked really hard on because you know, my job in

60
00:02:32,759 --> 00:02:35,719
government is to say, hang on, there is a potential risk,

61
00:02:35,759 --> 00:02:38,039
and not a definite risk, but a potential risk of

62
00:02:38,159 --> 00:02:40,800
something that could be bad. And my job is to

63
00:02:40,879 --> 00:02:43,520
protect the country and we can only do that if

64
00:02:43,599 --> 00:02:46,159
we develop the capability we need in our safety institute

65
00:02:46,199 --> 00:02:47,639
and then go in and make sure we can test

66
00:02:47,680 --> 00:02:50,759
the models before they are released. Delight, did that happened today.

67
00:02:50,759 --> 00:02:52,680
But you know, what's your view on what we should

68
00:02:52,719 --> 00:02:55,240
be doing. Right, You've talked about the potential risk. Right again,

69
00:02:55,280 --> 00:02:57,199
we don't know, but you know, what are the types

70
00:02:57,199 --> 00:02:59,759
of things governments like ours should be doing to manage

71
00:02:59,800 --> 00:03:01,319
and de gate against those risks.

72
00:03:01,479 --> 00:03:03,639
Speaker 4: Well, I generally think that it is good for government

73
00:03:03,680 --> 00:03:07,319
to play a role when the public safety is at risk. So,

74
00:03:07,719 --> 00:03:10,080
you know, really, for the vast majority of software, the

75
00:03:10,080 --> 00:03:12,840
public safety is not at risk. If the app crashes

76
00:03:12,840 --> 00:03:15,719
on your phone, your laptop, it's not a massive catastrophe.

77
00:03:15,879 --> 00:03:20,319
But when talking about digital superintelligence, I think, which does

78
00:03:20,400 --> 00:03:22,960
pose a risk to the public, then there is a

79
00:03:23,039 --> 00:03:25,560
role for government to play to safeguard the interest of

80
00:03:25,599 --> 00:03:28,800
the public. And this is of course true in many fields,

81
00:03:29,080 --> 00:03:32,159
you know, aviation, cars, you know, I deal with regulators

82
00:03:32,360 --> 00:03:37,039
throughout the world because of stalling being communications, rockets being aerospace,

83
00:03:37,319 --> 00:03:40,159
and cars, you know, being vehicle transport. So I'm very

84
00:03:40,159 --> 00:03:42,599
familiar with dealing with with regulators, and I actually agree

85
00:03:42,599 --> 00:03:44,319
with the vast majority of regulations. There's a few that

86
00:03:44,360 --> 00:03:46,000
I disagree with from time to time, but zero point

87
00:03:46,080 --> 00:03:49,199
one percent, probably less than one percent of regulations I

88
00:03:49,240 --> 00:03:52,120
disagree with. So there is some concern from people in

89
00:03:52,120 --> 00:03:55,240
Silicon Valley who who have never dealt with regulators before,

90
00:03:55,280 --> 00:03:57,680
and they think that this is going to just crush

91
00:03:57,680 --> 00:04:00,719
innovation and slow them down and be annoying. But and

92
00:04:00,879 --> 00:04:03,159
it will be annoying, It's true. They're not wrong about that.

93
00:04:03,400 --> 00:04:06,159
But I think there's we've learnt over the years that

94
00:04:06,639 --> 00:04:09,680
having a referee is a good thing. And if you

95
00:04:09,719 --> 00:04:12,599
look at any sports game, there's always a referee, and

96
00:04:12,680 --> 00:04:15,159
nobody's suggesting, I think, to have a sports game without one.

97
00:04:15,240 --> 00:04:16,959
And I think that's the right way to think about this,

98
00:04:17,120 --> 00:04:19,800
is for governed to be a referee, to make sure

99
00:04:20,000 --> 00:04:24,360
the sportsmanlike conduct and that the public safety is you know,

100
00:04:24,680 --> 00:04:26,959
is addressed. That we care about the public safety because

101
00:04:26,959 --> 00:04:29,040
I think there might be at times too much optimism

102
00:04:29,079 --> 00:04:31,480
about technology. And I speak I say that as a technologist.

103
00:04:31,480 --> 00:04:33,519
I mean, so I ought to know and like I

104
00:04:33,560 --> 00:04:36,360
said that, on balance, I think that the AI will

105
00:04:36,519 --> 00:04:40,480
be a forcible good most likely, but the probability of

106
00:04:40,480 --> 00:04:42,959
it going bad it's not zero percent. Yeah, so we

107
00:04:43,040 --> 00:04:45,079
just need to mitigate the downside potential.

108
00:04:45,279 --> 00:04:47,639
Speaker 3: And then how you talk about referee and that's what

109
00:04:47,639 --> 00:04:48,439
we're demonstrate.

110
00:04:48,519 --> 00:04:50,279
Speaker 2: The yeah, well there we go, I mean, you know,

111
00:04:50,360 --> 00:04:52,199
and we talked about this in Demos and I discussed

112
00:04:52,199 --> 00:04:53,000
this a long time.

113
00:04:52,839 --> 00:04:56,079
Speaker 4: Ago, like literally facing right and actually.

114
00:04:56,000 --> 00:04:58,079
Speaker 3: Demoster his credit and the credit of people in the industry,

115
00:04:58,079 --> 00:04:58,839
did say that to us.

116
00:04:59,160 --> 00:05:01,439
Speaker 2: You say it's not right. Yeah, that Demis and his

117
00:05:01,519 --> 00:05:02,959
colleagues are marking their own homework.

118
00:05:03,040 --> 00:05:03,199
Speaker 4: Right.

119
00:05:03,240 --> 00:05:05,680
Speaker 2: There needs to be someone independent, and that's why we've

120
00:05:05,680 --> 00:05:08,839
developed the safety Institute here. I mean, do you think

121
00:05:09,040 --> 00:05:11,439
governments can develop the expertise. One of the things we

122
00:05:11,480 --> 00:05:14,079
need to do is they hang on you know, Demis

123
00:05:14,120 --> 00:05:15,920
and Exam all the others have got a lot of

124
00:05:16,000 --> 00:05:19,279
very smart people doing this. Governments need to quickly tool

125
00:05:19,360 --> 00:05:22,399
up capability wise, personnel wise, which is what we're doing.

126
00:05:22,759 --> 00:05:24,160
Speaker 3: I mean, do you think it is possible.

127
00:05:23,800 --> 00:05:25,959
Speaker 2: For governments to do that fast enough, given how quickly

128
00:05:26,040 --> 00:05:28,240
the technology is developing, or what do we need to

129
00:05:28,240 --> 00:05:29,680
do to make sure we do it quick enough?

130
00:05:29,800 --> 00:05:31,680
Speaker 4: No, I think it's a great point you're making. The

131
00:05:31,720 --> 00:05:35,000
pace of AI is faster than any technology I've seen

132
00:05:35,040 --> 00:05:37,399
in history by far, and it seems to be growing

133
00:05:37,399 --> 00:05:41,240
in capability by at least fivefold peraps tenfold per year.

134
00:05:41,360 --> 00:05:43,680
It'll certainly grow by an automatitude next year.

135
00:05:43,759 --> 00:05:44,600
Speaker 3: Yeah.

136
00:05:44,319 --> 00:05:47,800
Speaker 4: So and government isn't used to moving at that speed.

137
00:05:47,920 --> 00:05:50,839
But I think even if there are not firm regulations,

138
00:05:51,120 --> 00:05:53,959
even if there's not even if there isn't an enforcement capability,

139
00:05:54,160 --> 00:05:56,879
so we're having insight and being able to highlight concerns

140
00:05:56,920 --> 00:05:59,879
to the public will be very powerful. So even if

141
00:06:00,000 --> 00:06:02,040
that's all that's accomplished, I think that will be very good.

142
00:06:02,160 --> 00:06:04,279
Speaker 3: Okay, y, well, hopefully we can do better than that.

143
00:06:04,319 --> 00:06:04,680
Speaker 4: Hopefully.

144
00:06:04,759 --> 00:06:07,000
Speaker 2: Yeah, yeah, no, but that's how fair. Actually we were

145
00:06:07,040 --> 00:06:09,639
talking before it was striking. You know, you're someone who

146
00:06:09,639 --> 00:06:13,519
spent their life in technology. They're living More's law. And

147
00:06:13,959 --> 00:06:16,000
what it was interesting over the last couple of days

148
00:06:16,040 --> 00:06:18,680
talking to everyone who's doing the development of this, and

149
00:06:18,720 --> 00:06:21,000
I think you concur with this is just the pace

150
00:06:21,040 --> 00:06:24,600
of advancement here is unlike anything all of you have

151
00:06:24,680 --> 00:06:26,480
seen in your careers and technology.

152
00:06:26,519 --> 00:06:28,360
Speaker 3: Is that fair because you've got these kind of.

153
00:06:28,279 --> 00:06:32,040
Speaker 2: Compounding effects from the hardware and the data and the personnel.

154
00:06:32,279 --> 00:06:35,000
Speaker 4: Yeah, I mean the two currently the two leading centers

155
00:06:35,000 --> 00:06:37,639
for a development are the San Francisco Bay area and

156
00:06:37,680 --> 00:06:39,800
the sort of London area, And there are many other

157
00:06:39,800 --> 00:06:41,160
places where it's being done, but those are the two

158
00:06:41,240 --> 00:06:43,600
leading areas. So I think if you know, if the

159
00:06:43,680 --> 00:06:46,920
United States and the UK and China are sort of

160
00:06:47,000 --> 00:06:49,959
aligned on safety, that's all going to be. That's really

161
00:06:50,240 --> 00:06:52,199
that's where that's where the leadership is general.

162
00:06:52,279 --> 00:06:54,240
Speaker 3: And you actually you mentioned China there.

163
00:06:54,319 --> 00:06:57,560
Speaker 2: So I took a decision to invite China to summit

164
00:06:57,600 --> 00:07:00,560
over the last three days, and it was not easy decision.

165
00:07:00,560 --> 00:07:03,040
A lot of people criticize me for you know, my viewers,

166
00:07:03,040 --> 00:07:05,439
if you're going to try it essential serious conversation, you

167
00:07:05,439 --> 00:07:07,399
need to. But what would your thoughts? You do business

168
00:07:07,439 --> 00:07:09,319
all around the world. You just talked about it there.

169
00:07:10,160 --> 00:07:12,360
Should we be engaging with them? Can we trust them?

170
00:07:12,480 --> 00:07:13,560
Is that the right thing to have done?

171
00:07:14,040 --> 00:07:15,959
Speaker 4: If we don't, If China is not on board with

172
00:07:16,120 --> 00:07:19,439
AI safety, it's somewhat of a moot situation. The single

173
00:07:19,439 --> 00:07:22,560
biggest objection that I get to any kind of AI

174
00:07:22,680 --> 00:07:25,600
regulation or sort of safety controls are well, China is

175
00:07:25,600 --> 00:07:27,360
not going to do it, and therefore they will just

176
00:07:27,639 --> 00:07:30,040
jump into the lead and exceed us all. But actually

177
00:07:30,160 --> 00:07:33,759
China is willing to participay in safety and thank you

178
00:07:33,800 --> 00:07:36,079
for inviting them, and they you know, I think we

179
00:07:36,079 --> 00:07:38,920
should thank China for for attending. When I was in China,

180
00:07:38,920 --> 00:07:42,600
earlier this year. My main subject of discussion with the

181
00:07:42,639 --> 00:07:45,040
leadership in China was AI safety and saying that this

182
00:07:45,160 --> 00:07:47,439
is really something that they should care about, and they

183
00:07:47,480 --> 00:07:50,399
took it seriously and you are too, which is great.

184
00:07:50,560 --> 00:07:52,959
And having them here I think was essential. Really, if

185
00:07:52,959 --> 00:07:55,279
they're not participants, it's pointless.

186
00:07:55,319 --> 00:07:57,199
Speaker 3: Yeah, that's and I think we were pleased.

187
00:07:57,199 --> 00:07:59,360
Speaker 2: I think they were engaged yesterday in the discussions and

188
00:07:59,360 --> 00:08:02,920
actually ended up signing the same communicy that everyone else did. Great,

189
00:08:02,920 --> 00:08:04,959
which is a good stop, right, And I said, if

190
00:08:05,000 --> 00:08:07,319
we need everyone to approach us in a similar way,

191
00:08:07,360 --> 00:08:10,360
if we're going to have a realistic chance of resolving it.

192
00:08:10,600 --> 00:08:12,839
I was going to You talked about innovation earlier and

193
00:08:12,959 --> 00:08:14,040
regulation being annoying.

194
00:08:14,279 --> 00:08:16,680
Speaker 3: There was a good debate today we had about.

195
00:08:16,319 --> 00:08:18,399
Speaker 2: Open source and I think you've kind of been a

196
00:08:18,439 --> 00:08:21,920
proponent of algorithmic transparency and making some of the ex

197
00:08:21,959 --> 00:08:25,399
algorithms public and actually we were talking about every hint

198
00:08:25,399 --> 00:08:28,519
and on the way in. He's particularly been very concerned

199
00:08:28,560 --> 00:08:31,079
about open source models being used.

200
00:08:30,879 --> 00:08:31,800
Speaker 3: By bad actors.

201
00:08:32,159 --> 00:08:33,799
Speaker 2: You've got a group of people who say they are

202
00:08:34,120 --> 00:08:37,200
critical to innovation happening in that distributed way.

203
00:08:37,440 --> 00:08:38,000
Speaker 3: Look, it's a trick.

204
00:08:38,039 --> 00:08:40,879
Speaker 2: There's probably no perfect answer, and there is a tricky balance.

205
00:08:40,960 --> 00:08:42,799
What are your thoughts on how we should approach this

206
00:08:42,840 --> 00:08:46,799
open source question or where should we be targeting whatever

207
00:08:46,960 --> 00:08:48,759
regulatory or monitoring that we're going.

208
00:08:48,759 --> 00:08:52,080
Speaker 4: To do well. The open source algorithms and data tend

209
00:08:52,080 --> 00:08:54,600
to lag the closed source by six to twelve, but

210
00:08:55,600 --> 00:08:57,759
given the rate of improvement that there's actually there for

211
00:08:57,919 --> 00:09:00,440
quite a big difference between the closed source in the

212
00:09:00,480 --> 00:09:02,799
open If things are improving by a factor of let's

213
00:09:02,840 --> 00:09:05,639
say five or more than being a year behind is

214
00:09:05,759 --> 00:09:07,960
you're five times worse. So it's a pretty big difference.

215
00:09:08,080 --> 00:09:10,320
And that might be actually an okay situation, but it

216
00:09:10,399 --> 00:09:12,480
certainly we'll get to the point where you've got open

217
00:09:12,519 --> 00:09:15,159
source AI that can do that, that will start to

218
00:09:15,200 --> 00:09:17,960
approach human level intelligence or PAPS succeeded. I don't know

219
00:09:18,039 --> 00:09:19,799
quite what to do about it. I think it's somewhat

220
00:09:19,799 --> 00:09:21,879
in neeverment, there'll be some amount of open source, and

221
00:09:22,000 --> 00:09:24,360
I guess I would have a slight bias towards open

222
00:09:24,399 --> 00:09:26,200
source because at least you can see what's going on.

223
00:09:26,320 --> 00:09:28,360
It wears a closed source. You don't know what's going

224
00:09:28,360 --> 00:09:30,480
on now. It should be said with AI that even

225
00:09:30,480 --> 00:09:32,440
if it's an open source, do you actually know what's

226
00:09:32,480 --> 00:09:35,039
going on. Because if you've got a gigantic data file

227
00:09:35,200 --> 00:09:37,759
and you know sort of billions of days of weights

228
00:09:37,759 --> 00:09:39,879
and parameters, you can't just read it and see what

229
00:09:39,879 --> 00:09:43,320
it's going to do. It's a gigantic file of inscrutable numbers.

230
00:09:43,559 --> 00:09:45,159
You can test it when you run it. You can

231
00:09:45,159 --> 00:09:47,200
test it. You can run a bunch of tests to

232
00:09:47,240 --> 00:09:49,639
see what it's going to do. But it's probabilistic as

233
00:09:49,679 --> 00:09:52,840
opposed to deterministic. It's not like traditional programming where you've

234
00:09:52,879 --> 00:09:55,440
got you've got very discrete logic and the outcome is

235
00:09:55,519 --> 00:09:57,360
very predictable and you can read each line and see

236
00:09:57,360 --> 00:09:59,279
what each line's going to do. A neural net is

237
00:09:59,759 --> 00:10:02,039
just a much of probabilities. I mean, it sort of

238
00:10:02,120 --> 00:10:04,840
ends up being a giant Comma separated value file. It's

239
00:10:04,879 --> 00:10:07,840
like our digital guide is a CSP file. Really, Okay,

240
00:10:08,080 --> 00:10:09,000
that is kind of what it is.

241
00:10:09,120 --> 00:10:11,600
Speaker 2: Yeah, Now at that point you've just made is one

242
00:10:11,600 --> 00:10:14,720
that we have been talking about a lot, because again

243
00:10:15,080 --> 00:10:17,120
conversation with the people are developing their technology.

244
00:10:17,159 --> 00:10:19,120
Speaker 3: Make the point that you've just made. It is not

245
00:10:19,320 --> 00:10:20,159
like normal.

246
00:10:19,879 --> 00:10:24,240
Speaker 2: Software where there's predictability about inputs improving leading to this

247
00:10:24,240 --> 00:10:27,960
particular output improving, and as the models iterate and improve,

248
00:10:28,480 --> 00:10:30,200
we don't quite know what's going to come out the

249
00:10:30,240 --> 00:10:32,039
other end. I think we'd agree with that, which is

250
00:10:32,039 --> 00:10:34,120
why I think there is this bias that we need

251
00:10:34,159 --> 00:10:35,840
to get in there while the training runs are being

252
00:10:35,919 --> 00:10:38,840
done before the models are released, to understand what is

253
00:10:38,840 --> 00:10:41,919
this new iteration broad about in terms of capability, which

254
00:10:41,960 --> 00:10:43,559
it sounds like you would agree with.

255
00:10:44,080 --> 00:10:45,639
Speaker 3: I was going to shift goes a little bit on.

256
00:10:46,240 --> 00:10:48,879
Speaker 2: You know, you've talked a lot about human consciousness human agency,

257
00:10:48,919 --> 00:10:52,039
which it actually might strike people as strange given that

258
00:10:52,080 --> 00:10:55,360
you are known for being such a brilliant innovator and technologist,

259
00:10:55,480 --> 00:10:57,840
but it's quite heartfelt when I hear you talk about

260
00:10:57,879 --> 00:11:01,919
it and the importance of maintaining that agency in technology and.

261
00:11:01,879 --> 00:11:04,480
Speaker 3: Preserving human consciousness. Now it kind of links.

262
00:11:04,679 --> 00:11:06,039
Speaker 2: The thing I was going to ask is when I

263
00:11:06,159 --> 00:11:08,080
do interviews or talk to people out and about in

264
00:11:08,120 --> 00:11:10,279
this job about AI, the thing that comes up most

265
00:11:10,320 --> 00:11:12,480
actually is it probably not so much the stuff we've

266
00:11:12,519 --> 00:11:15,440
been talking about, but jobs. It's what is AI mean

267
00:11:15,480 --> 00:11:17,799
for my job? Is it going to mean that I

268
00:11:17,840 --> 00:11:19,759
don't have a job or my kids are not going

269
00:11:19,759 --> 00:11:20,519
to have a job.

270
00:11:20,799 --> 00:11:21,000
Speaker 4: Now.

271
00:11:21,159 --> 00:11:24,639
Speaker 2: Answer as a policymaker, as a leader is actually, AI

272
00:11:24,720 --> 00:11:27,240
is already creating jobs, and you can see that in

273
00:11:27,279 --> 00:11:28,159
the companies.

274
00:11:27,799 --> 00:11:28,519
Speaker 3: That are starting.

275
00:11:28,759 --> 00:11:31,440
Speaker 2: Also, the way it's being used is a little bit

276
00:11:31,519 --> 00:11:35,039
more as a co pilot necessarily versus replacing the person.

277
00:11:35,080 --> 00:11:37,279
There's still human agency, but it's helping you do your

278
00:11:37,360 --> 00:11:40,840
job better, which is a good thing. And as we've

279
00:11:40,840 --> 00:11:44,320
seen with technological revolutions in the past, clearly there's change

280
00:11:44,320 --> 00:11:45,159
in the labor market.

281
00:11:45,559 --> 00:11:46,440
Speaker 3: The amount of jobs.

282
00:11:46,720 --> 00:11:49,200
Speaker 2: I was quoting an MIT study today that they did

283
00:11:49,200 --> 00:11:51,279
a couple of years ago, something like sixty percent of

284
00:11:51,279 --> 00:11:53,840
the jobs at that moment didn't exist forty years ago,

285
00:11:53,960 --> 00:11:55,679
So hard to predict. And my job is to create

286
00:11:55,679 --> 00:11:58,840
an incredible education system, whether it's at school, whether it's

287
00:11:58,840 --> 00:12:01,639
retraining people at any point in their career, because ultimately,

288
00:12:01,720 --> 00:12:03,759
if we've got a skill population, they'll be able to

289
00:12:03,840 --> 00:12:06,000
keep up with the pace of change and have a

290
00:12:06,000 --> 00:12:08,720
good life. But you know that it's still a concern,

291
00:12:09,000 --> 00:12:12,440
and you know, what would your kind of observation be

292
00:12:12,559 --> 00:12:15,600
on AI and the impact of labor markets and people's

293
00:12:15,679 --> 00:12:17,960
jobs and how they should feel about that as they

294
00:12:18,039 --> 00:12:18,840
think about.

295
00:12:18,559 --> 00:12:20,720
Speaker 4: This, well, I think we are seeing the most disruptive

296
00:12:21,039 --> 00:12:24,080
force in history here. You know, where we have for

297
00:12:24,120 --> 00:12:26,360
the first time, we will have the first time something

298
00:12:26,360 --> 00:12:29,080
that is smarter than the smartest human and that I

299
00:12:29,080 --> 00:12:31,080
mean it's hard to say exactly what that moment is,

300
00:12:31,159 --> 00:12:34,159
but there will come a point where no job is needed.

301
00:12:34,360 --> 00:12:35,879
You can have a job if you want to have

302
00:12:35,919 --> 00:12:39,759
a job for sort of personal satisfaction, but the AI

303
00:12:39,919 --> 00:12:42,000
will be able to do everything. So I don't know

304
00:12:42,000 --> 00:12:44,440
if that makes people comfortable uncomfortable. You know. That's why

305
00:12:44,480 --> 00:12:46,559
I say, if you wish for a magic genie that

306
00:12:46,639 --> 00:12:48,639
gives you any wishes you want and there's no limit,

307
00:12:48,639 --> 00:12:50,639
you don't have those three limits three wish limits, not

308
00:12:50,720 --> 00:12:53,840
since you have many wishes as you want. So it's

309
00:12:53,919 --> 00:12:55,559
both good and bad. One of the challenges in the

310
00:12:55,559 --> 00:12:58,240
future will be how do we find meaning in life

311
00:12:57,879 --> 00:13:00,399
if you have a magic genie that can do everything

312
00:13:00,440 --> 00:13:02,960
you want. I do think we's it's hard, you know.

313
00:13:03,000 --> 00:13:05,759
Whene's when when there's new technology, it tends to have

314
00:13:06,080 --> 00:13:08,159
usually follow an S curve. In this case, we're going

315
00:13:08,240 --> 00:13:09,919
to be on the exponential portion of the S curve.

316
00:13:10,000 --> 00:13:13,200
For a long time and we have to ask for anything,

317
00:13:13,240 --> 00:13:15,840
and it won't be if we won't have universal basic income,

318
00:13:15,840 --> 00:13:18,240
we'll have universal high income. So in some in some sense,

319
00:13:18,240 --> 00:13:20,440
it will be somewhat of a leveler or an equalizer,

320
00:13:20,519 --> 00:13:22,480
because really, I think everyone will have access to this

321
00:13:22,559 --> 00:13:24,720
magic gene and you're able to ask any question. It

322
00:13:24,720 --> 00:13:26,919
will certainly be good for education. It'll be the best

323
00:13:26,960 --> 00:13:30,039
tutor you could, the most patient tutor sit there all day,

324
00:13:30,320 --> 00:13:32,279
and there will be no shortage of goods and services.

325
00:13:32,279 --> 00:13:34,440
Will be an age of abundance. I think if I'd

326
00:13:34,440 --> 00:13:38,320
recommend people read in Banks, the Banks culture books are

327
00:13:38,799 --> 00:13:40,879
probably the best in visioning. In fact, not probably, they're

328
00:13:40,879 --> 00:13:43,240
definitely by far the best envisioning of an a future.

329
00:13:43,360 --> 00:13:46,519
There's nothing even close. So I'd recommend, really recommend Banks.

330
00:13:46,519 --> 00:13:48,600
I'm a very big fan. All these books are good.

331
00:13:48,799 --> 00:13:51,399
There's not Jay which one all of them. So that's

332
00:13:51,559 --> 00:13:54,279
that'll give you a sense of what is a I

333
00:13:54,279 --> 00:13:58,559
guess a fairly utopian pro toopian future with with Ai, Yeah,

334
00:13:58,600 --> 00:13:58,960
which is.

335
00:13:58,960 --> 00:14:01,720
Speaker 2: Good from an as you said, it's a universal high income,

336
00:14:01,720 --> 00:14:03,360
which is a nice phrase, and that's it's good for

337
00:14:03,440 --> 00:14:07,559
a materialistic sense of abundance. Actually, that kind of then

338
00:14:07,759 --> 00:14:08,679
leads to the question.

339
00:14:08,519 --> 00:14:10,559
Speaker 3: That you pose. Right, I'm someone who believes, you know,

340
00:14:10,600 --> 00:14:12,639
work gives you meaning. I write a lot about that.

341
00:14:12,799 --> 00:14:14,240
As you know, I think work is a good thing.

342
00:14:14,320 --> 00:14:16,440
Speaker 2: It, you know, gives people purpose in their lives. And

343
00:14:16,759 --> 00:14:19,639
if you then remove a large chunk of that, you know,

344
00:14:19,679 --> 00:14:20,440
what does that mean?

345
00:14:20,519 --> 00:14:21,519
Speaker 3: And where do you get that?

346
00:14:21,679 --> 00:14:23,480
Speaker 2: You know, where do you get that drive, that motivation,

347
00:14:23,559 --> 00:14:25,519
that purpose. I mean, you're talking about it. You work

348
00:14:25,559 --> 00:14:27,840
a lot of our you know, as I was.

349
00:14:27,759 --> 00:14:29,879
Speaker 4: Mentioning when we were talking earlier, I have to somewhat

350
00:14:29,919 --> 00:14:32,639
engage in the liberty suspension of disbelief because I'm putting

351
00:14:32,639 --> 00:14:34,919
so much blood, sweat and tears into a work project

352
00:14:35,320 --> 00:14:38,480
and burning the you know, three am oil. Then I'm like, wait,

353
00:14:38,519 --> 00:14:40,039
why am I doing this? I just wait for the

354
00:14:40,080 --> 00:14:42,559
AI to do it. I'm just lashing myself for no reason.

355
00:14:43,159 --> 00:14:45,080
It must be a glutton for punishment.

356
00:14:45,080 --> 00:14:46,759
Speaker 3: Called demos and tell them to hurry up and then

357
00:14:46,799 --> 00:14:49,159
you can have a holiday. Right, it's a plan. Yeah, No,

358
00:14:49,279 --> 00:14:50,440
it's a Look, it's a tricky.

359
00:14:50,519 --> 00:14:52,679
Speaker 2: It's a tricky thing because I think, you know, part

360
00:14:52,679 --> 00:14:54,639
of our job is to make sure that we can

361
00:14:55,000 --> 00:14:58,720
navigate to that very I think largely positive place and

362
00:14:58,799 --> 00:15:01,399
help people through it between now and then, because these

363
00:15:01,399 --> 00:15:03,759
things bring a lot about change in the labor market

364
00:15:03,799 --> 00:15:04,440
as we've seen.

365
00:15:04,600 --> 00:15:06,720
Speaker 4: Yeah, I think it's probably is generally a good thing

366
00:15:06,759 --> 00:15:08,360
because you know, there are a lot of jobs that

367
00:15:08,399 --> 00:15:11,879
are uncomfortable or dangerous or which sort of tedious, and

368
00:15:12,399 --> 00:15:14,399
the computer will have no problem doing that. We're happy

369
00:15:14,440 --> 00:15:16,159
to do that all day long. So you know, it's

370
00:15:16,360 --> 00:15:17,879
fun to cook food, but it's not that fun to

371
00:15:17,919 --> 00:15:20,240
wash the dishes, but the computer is perfectly happy to

372
00:15:20,240 --> 00:15:22,840
watch the dishes. I guess there is. You know, we

373
00:15:22,879 --> 00:15:25,879
still have a sports like where where humans compete and

374
00:15:25,919 --> 00:15:29,039
like the Olympics, and obviously a machine can can go

375
00:15:29,159 --> 00:15:31,399
faster than any human, but we still have we saw

376
00:15:31,600 --> 00:15:34,159
humans race against each other and have all, you know,

377
00:15:34,399 --> 00:15:36,679
have these sports competitions against each other where even though

378
00:15:36,720 --> 00:15:39,000
the machines are better, which there's all I guess compunity

379
00:15:39,080 --> 00:15:41,240
who can be the best human at something, and people

380
00:15:41,240 --> 00:15:43,240
do find pulforment in that. So I guess that's perhaps

381
00:15:43,320 --> 00:15:45,960
a good example of how even when machines are faster than.

382
00:15:45,960 --> 00:15:47,639
Speaker 3: Us, stronger than us, we still find a way.

383
00:15:47,559 --> 00:15:49,840
Speaker 4: We still enjoy competing against other humans, so at least

384
00:15:49,919 --> 00:15:50,679
who's the best human.

385
00:15:50,720 --> 00:15:52,639
Speaker 2: It's a good it's a good analogy, and we've been

386
00:15:52,679 --> 00:15:55,600
talking a lot about managing the risks. Just before we

387
00:15:55,639 --> 00:15:57,799
move on and finish on AI is just talk a

388
00:15:57,840 --> 00:16:01,320
little bit about the opportunities. You know, you're engaged in

389
00:16:01,320 --> 00:16:04,200
lots of different companies. You're being an obvious one, which

390
00:16:04,240 --> 00:16:06,360
is doing some exciting stuff. You touched on the thing

391
00:16:06,360 --> 00:16:09,840
that I'm probably most excited about, which is an education. Yeah,

392
00:16:09,919 --> 00:16:12,679
and I think many people will have seen sous video

393
00:16:12,679 --> 00:16:14,720
from earlier this year. Is ted talk about as you

394
00:16:14,799 --> 00:16:16,159
talked about it's like a personal tutor.

395
00:16:16,279 --> 00:16:18,720
Speaker 4: Yeah, an amazing personal tutor.

396
00:16:18,480 --> 00:16:19,519
Speaker 3: An amazing personal tutor.

397
00:16:19,559 --> 00:16:22,399
Speaker 2: And we know the difference in learning having that personal

398
00:16:22,559 --> 00:16:24,960
his tutor is incredible compared to class from learning. So

399
00:16:25,000 --> 00:16:27,240
if you can have every child have a personal tutor

400
00:16:27,519 --> 00:16:30,200
specifically for them that then just evolves with them over time,

401
00:16:30,320 --> 00:16:32,320
that could be extraordinary. So that you know, for me,

402
00:16:32,360 --> 00:16:34,440
I look at it, I think, gosh, that is within

403
00:16:34,519 --> 00:16:36,720
reach at this point, and that's one of the benefits

404
00:16:36,759 --> 00:16:39,639
I'm most excited about. When you look at the landscape

405
00:16:39,679 --> 00:16:41,759
of things that you see as possible, what is it

406
00:16:41,840 --> 00:16:44,559
that you know you are particularly excited about.

407
00:16:44,840 --> 00:16:47,200
Speaker 4: I think certainly AI tutors are going to be the

408
00:16:47,279 --> 00:16:51,360
amazing perhaps already are. I think there's also perhaps companionship,

409
00:16:51,440 --> 00:16:53,799
which may seem odd because how can the computer really

410
00:16:53,799 --> 00:16:56,519
be your friend? But if you haven't that has memory,

411
00:16:56,639 --> 00:16:59,360
you know, and remembers all of your interactions and has

412
00:16:59,399 --> 00:17:01,720
read every you're going to actually like give it permission

413
00:17:01,720 --> 00:17:03,639
to read everything you've ever done, so really will know

414
00:17:03,679 --> 00:17:06,240
you better than anyone, perhaps even yourself. And where you

415
00:17:06,279 --> 00:17:08,559
can talk to it every day and those conversations spoiled

416
00:17:08,599 --> 00:17:11,319
upon each other, you will actually have a great friend

417
00:17:11,480 --> 00:17:13,359
as long as that friend can stay your friend and

418
00:17:13,400 --> 00:17:15,799
not get turned off or something. Don't turn off my friends.

419
00:17:16,160 --> 00:17:17,759
But I think that will actually be a real thing.

420
00:17:18,440 --> 00:17:20,839
And I one of my sons is sort of has

421
00:17:20,880 --> 00:17:23,960
some learning disabilities and has trouble making friends actually, and

422
00:17:24,000 --> 00:17:26,799
I was like, well, you know, he AI friend would

423
00:17:26,799 --> 00:17:27,720
actually be a great friend.

424
00:17:28,519 --> 00:17:31,519
Speaker 2: Okay, you know that was a surprising answer. That's actually

425
00:17:31,599 --> 00:17:34,480
it's worth worth reflecting on that. That's really interesting. I

426
00:17:34,480 --> 00:17:38,440
mean we're already seeing it actually as we deliver psychotherapy anyway,

427
00:17:38,839 --> 00:17:42,319
now doing far more by digitally and by telephone to

428
00:17:42,319 --> 00:17:44,039
people and it's making a huge difference, and you can

429
00:17:44,079 --> 00:17:46,359
see a world in which actually, you know, AI can

430
00:17:46,440 --> 00:17:50,599
provide that social benefit to people. Just a quick question

431
00:17:50,720 --> 00:17:53,319
on X and then we should open it up to everybody.

432
00:17:54,119 --> 00:17:56,599
You made a change when in one of the made

433
00:17:56,640 --> 00:17:58,440
many changes, but one of the one of.

434
00:17:58,400 --> 00:18:01,559
Speaker 4: The changes that letter. Yet thing about it, you really.

435
00:18:01,319 --> 00:18:03,839
Speaker 2: Do, really do one of the changes which you know

436
00:18:03,960 --> 00:18:05,799
kind of you know, it goes into the space that

437
00:18:05,880 --> 00:18:08,240
you know we have to operate in. And this balance

438
00:18:08,279 --> 00:18:12,079
between free speech and moderation is you know, we grapple.

439
00:18:11,720 --> 00:18:12,839
Speaker 3: With as politicians.

440
00:18:13,279 --> 00:18:15,039
Speaker 2: You were grappling with your own version of that, and

441
00:18:15,200 --> 00:18:18,519
you moved away from a kind of manual human.

442
00:18:18,720 --> 00:18:21,519
Speaker 3: Way of doing it the moderation to the community and that.

443
00:18:22,000 --> 00:18:24,640
Speaker 2: Yeah, and I think that's it was an interesting change, right,

444
00:18:24,680 --> 00:18:27,400
It's not what everyone else has done. It would be good,

445
00:18:27,480 --> 00:18:29,519
you know what's what was the reasoning behind that and

446
00:18:29,519 --> 00:18:31,200
why do you think that is a better way to

447
00:18:31,240 --> 00:18:31,519
do that?

448
00:18:31,759 --> 00:18:33,440
Speaker 4: Yeah. Part of the problem is if you if you

449
00:18:33,559 --> 00:18:36,720
empower people as sensors, then well there's going to be

450
00:18:36,720 --> 00:18:39,039
some amount of bias that they have. And then whoever

451
00:18:39,160 --> 00:18:42,440
appoints the sensors is effectively in control of information. So

452
00:18:42,480 --> 00:18:44,559
then the idea behind community notice, well, how do we

453
00:18:44,759 --> 00:18:47,200
have a consensus driven I mean, so it's not really

454
00:18:47,200 --> 00:18:50,000
censoring it, but consensus driven approach to truth. How do

455
00:18:50,039 --> 00:18:53,200
we how do we make things the least amount untrue?

456
00:18:53,319 --> 00:18:56,480
Could you say, like you can't perhaps get to pure truth,

457
00:18:56,519 --> 00:18:59,319
but you can aspire to be more truthful. So the

458
00:18:59,359 --> 00:19:02,359
thing about comurnity notes, it doesn't actually delete anything, It

459
00:19:02,400 --> 00:19:04,720
simply adds context. Now that context could be this thing

460
00:19:04,759 --> 00:19:08,519
is untrue for the following reasons. But importantly, with community notes,

461
00:19:08,599 --> 00:19:12,240
everything is open source actually, so you can see the software,

462
00:19:12,400 --> 00:19:14,119
every line of the software, you can see all of

463
00:19:14,119 --> 00:19:16,759
the data that went into a community note, and you

464
00:19:16,799 --> 00:19:19,839
can independently create that community note. So if you've got

465
00:19:20,160 --> 00:19:22,480
if you see manipulations of data, you can actually highlight

466
00:19:22,519 --> 00:19:24,759
that and say, well this this appears to be some

467
00:19:24,880 --> 00:19:28,440
gaming of the system, and you can suggest improvement. So

468
00:19:28,480 --> 00:19:30,880
it's maximum transparency, which.

469
00:19:30,720 --> 00:19:32,640
Speaker 2: Is I think combined with the kind of wisdom of

470
00:19:32,680 --> 00:19:34,920
the crowds and trying to get to a better answer.

471
00:19:35,240 --> 00:19:37,359
Speaker 4: And really one of the key elements of community notes

472
00:19:37,400 --> 00:19:38,799
is that in order for a note to be shown,

473
00:19:39,079 --> 00:19:42,240
people who have historically disagreed must agree, and there is

474
00:19:42,240 --> 00:19:45,039
a bit of AI usage here. So this will populate

475
00:19:45,079 --> 00:19:48,720
a parameter space around each contributor to the community notes,

476
00:19:49,039 --> 00:19:52,559
and then a parameter space so everyone's got basically these

477
00:19:52,640 --> 00:19:55,559
vectors associated with them, which so it's not as simple

478
00:19:55,599 --> 00:19:58,279
as right or left, it's saying, it's more it's several

479
00:19:58,400 --> 00:20:01,599
hundred vectors. That that because things are more complicated than

480
00:20:01,599 --> 00:20:04,319
something right, right or left. And then we'll do sort

481
00:20:04,359 --> 00:20:07,759
of inverse correlation, say like, okay, these people generally disagree,

482
00:20:07,799 --> 00:20:10,319
but they agree about this note, so then that so

483
00:20:10,440 --> 00:20:13,440
then that that that gives the note credibility. Okay, yeah,

484
00:20:13,799 --> 00:20:15,839
that's that's the core of it, and it's working quite well.

485
00:20:16,359 --> 00:20:18,680
I get to see a note actually be present for

486
00:20:18,720 --> 00:20:21,039
more than a few hours, but that that is incorrect.

487
00:20:21,079 --> 00:20:23,200
So the batting average is extremely good. And when I

488
00:20:23,240 --> 00:20:25,880
ask people say, oh, they're worried about community notes sort

489
00:20:25,880 --> 00:20:28,039
of being disinformation, like send me one, and then they can't,

490
00:20:28,359 --> 00:20:29,960
so so I think it's I think it's quite good.

491
00:20:29,960 --> 00:20:33,240
I mean, the general aspiration is with the X platform

492
00:20:33,400 --> 00:20:36,279
is to inform and entertain the public and to be

493
00:20:36,319 --> 00:20:39,000
as accurate as possible and as truthful as possible, even

494
00:20:39,039 --> 00:20:41,559
if someone doesn't like the truth. You know, people don't

495
00:20:41,559 --> 00:20:44,720
always like the truth always, but that's the aspiration. And

496
00:20:44,759 --> 00:20:47,480
I think if we are, if we stay true to

497
00:20:47,519 --> 00:20:50,480
the truth, then I think we'll find that people use

498
00:20:50,519 --> 00:20:53,160
the system to learn what is going on and to

499
00:20:53,400 --> 00:20:56,799
I think actually truth pays, so I think it'll be well,

500
00:20:56,839 --> 00:20:59,279
I mean, assuming you don't want to engage your self delusion,

501
00:20:59,319 --> 00:21:01,319
then I think it's the smart move.

502
00:21:01,519 --> 00:21:04,440
Speaker 3: Excellent, very helpful. Right, let's open it up to all

503
00:21:04,480 --> 00:21:05,440
our guests here, and.

504
00:21:05,359 --> 00:21:07,799
Speaker 2: We've got some microphones that'll company I find you have,

505
00:21:07,880 --> 00:21:09,519
We've got yes, go for it.

506
00:21:09,799 --> 00:21:10,839
Speaker 3: Thank you, good evening.

507
00:21:11,039 --> 00:21:14,440
Speaker 5: Alice Bentink from Entrepreneur. First, thank you for a fascinating conversation.

508
00:21:14,519 --> 00:21:17,000
I suppose a question for each of you, Prime Minister.

509
00:21:17,119 --> 00:21:19,759
The UK has some of the best universities in the world,

510
00:21:19,839 --> 00:21:22,119
we have the talent. What will it takes the UK

511
00:21:22,240 --> 00:21:26,000
to be a real breeding with unicorn companies? Being a

512
00:21:26,000 --> 00:21:28,640
founder in the UK is still a non obvious career

513
00:21:28,720 --> 00:21:31,440
choice for the most exceptional technical talent. What are the

514
00:21:31,480 --> 00:21:33,519
cultural elements that we need to put into place to

515
00:21:33,599 --> 00:21:33,960
change this?

516
00:21:34,119 --> 00:21:35,400
Speaker 3: Thank you? You want to go first?

517
00:21:35,440 --> 00:21:37,279
Speaker 4: Go for it? Sure? Well, You're right that there are

518
00:21:37,640 --> 00:21:41,640
cultural elements where you know, the culture should celebrate creating

519
00:21:41,720 --> 00:21:45,440
new companies and there should be a bias towards supporting

520
00:21:45,680 --> 00:21:48,079
small companies because the ones that need nurturing. The larger

521
00:21:48,119 --> 00:21:51,039
companies really don't need nurturing. So you know, just you

522
00:21:51,079 --> 00:21:52,680
can think of it's sort of like a garden. If

523
00:21:52,720 --> 00:21:55,000
it's a little sprout that needs needs nurturing, if it's

524
00:21:55,000 --> 00:21:57,400
a mighty ogre, doesn't need quite as much. So I

525
00:21:57,400 --> 00:22:00,240
think that is a mindset change that is important. But

526
00:22:00,400 --> 00:22:04,160
I should mention that London is you know, London and

527
00:22:04,279 --> 00:22:06,599
San Francisco or the Bay Area are really the two

528
00:22:06,640 --> 00:22:09,559
centers for AI. So that so London is actually doing

529
00:22:09,640 --> 00:22:12,440
very well on that front. That the two I said,

530
00:22:12,440 --> 00:22:15,519
the two leading locations on Earth. You know, San Francisco's

531
00:22:15,559 --> 00:22:17,960
probably head of London, but London's really very strong. But

532
00:22:18,079 --> 00:22:21,319
London Area greater London home counties, I guess. So I'm

533
00:22:21,319 --> 00:22:23,559
just saying objectively this is the case. But you do

534
00:22:23,640 --> 00:22:25,680
need that. You need the infrastructure, you need the landlords

535
00:22:25,680 --> 00:22:28,200
who are willing to rent to new companies. You need

536
00:22:28,319 --> 00:22:30,680
more firms and accounts that are willing to support new companies.

537
00:22:30,799 --> 00:22:33,519
And it's generally it is a mindset change, and I

538
00:22:33,519 --> 00:22:35,640
think some of that is happening, but I think really

539
00:22:35,640 --> 00:22:37,799
it's just culturally people need to decide this is a

540
00:22:37,799 --> 00:22:38,680
good thing. Yeah.

541
00:22:38,839 --> 00:22:40,799
Speaker 2: Yeah, no, Actually, well thanks for what you said about

542
00:22:40,799 --> 00:22:42,480
the UK. It's something that we work hard on. Lots

543
00:22:42,480 --> 00:22:44,559
of people in the room are part of what makes

544
00:22:44,559 --> 00:22:48,279
this a fabulous place for companies, including Alice. So that's

545
00:22:48,279 --> 00:22:50,079
what i'd say is my job is to get all

546
00:22:50,119 --> 00:22:52,400
the you know they're nuts and volts right, make sure

547
00:22:52,440 --> 00:22:55,519
that all of you are starting companies can raise the capital.

548
00:22:55,240 --> 00:22:56,559
Speaker 3: That you need everything from.

549
00:22:56,680 --> 00:22:59,400
Speaker 2: You know, you'll see funding with our incredible you know

550
00:22:59,599 --> 00:23:01,920
EI tax release all the way through.

551
00:23:01,720 --> 00:23:03,839
Speaker 3: To your late stage rounds.

552
00:23:03,880 --> 00:23:05,960
Speaker 2: And we need reform of our pension funds and the

553
00:23:06,039 --> 00:23:09,559
Chancellor's got a bunch of incredible reforms to unlock capital

554
00:23:09,680 --> 00:23:11,920
from all the people who have it and deploy it

555
00:23:11,920 --> 00:23:12,839
into growth equity.

556
00:23:12,920 --> 00:23:13,079
Speaker 4: Right.

557
00:23:13,079 --> 00:23:14,960
Speaker 2: That is a work in progress. We're not there yet,

558
00:23:15,000 --> 00:23:17,720
but I think we're making good progress. We need talent,

559
00:23:17,839 --> 00:23:20,759
we need people, so that means an education system that

560
00:23:20,920 --> 00:23:22,480
prioritizes the things that matter.

561
00:23:22,559 --> 00:23:23,720
Speaker 3: And you've seen my reforms.

562
00:23:23,759 --> 00:23:26,000
Speaker 2: I go on about more maths, more maths, more maths,

563
00:23:26,279 --> 00:23:28,440
but I think that it's important but also attracting the

564
00:23:28,440 --> 00:23:30,519
best and the brightest fit. If you look at our

565
00:23:30,640 --> 00:23:33,279
fastest growing companies in this country, and I think it's

566
00:23:33,279 --> 00:23:36,000
probably the same in the US, over half of them

567
00:23:36,119 --> 00:23:39,079
have a non British founder, right, And so that tells

568
00:23:39,119 --> 00:23:40,920
you we've got to be a place that is open

569
00:23:41,000 --> 00:23:44,279
to the world's best and brightest entrepreneurial talent. So the

570
00:23:44,359 --> 00:23:46,680
visa regime that we've put in place, I think, does

571
00:23:46,720 --> 00:23:48,720
that makes it easy for those people to come here.

572
00:23:48,839 --> 00:23:50,720
And then actually it's the thing that we've spent the

573
00:23:50,759 --> 00:23:53,599
beginning of the session talking about the regulation, right, making

574
00:23:53,640 --> 00:23:56,759
sure that we've got a regulatory system that's pro innovation.

575
00:23:57,240 --> 00:23:57,319
Speaker 4: That.

576
00:23:57,440 --> 00:23:59,839
Speaker 2: Yeah, of course we always need guardrails on the things

577
00:24:00,160 --> 00:24:02,240
that will worry us, but we've got to create a

578
00:24:02,279 --> 00:24:05,079
space for people to innovate and do different things. Now,

579
00:24:05,079 --> 00:24:07,200
those are all my jobs. The thing that is tougher

580
00:24:07,279 --> 00:24:09,759
is the thing that Elon talked about, which is culture. Right,

581
00:24:09,839 --> 00:24:12,680
It's how do you transpose that culture from places like

582
00:24:12,720 --> 00:24:17,240
Silicon Valley across the world where people are unafraid to

583
00:24:17,440 --> 00:24:20,240
give up the security of a regular paycheck to go

584
00:24:20,279 --> 00:24:22,680
and start something and be comfortable with failure.

585
00:24:22,759 --> 00:24:23,920
Speaker 3: You talk about that a lot.

586
00:24:23,960 --> 00:24:25,359
Speaker 2: I think you talked about it more in when you're

587
00:24:25,400 --> 00:24:29,119
playing games, right, But you've got to be comfortable failing

588
00:24:29,440 --> 00:24:30,079
and knowing that.

589
00:24:29,960 --> 00:24:31,160
Speaker 3: That's just part of the process.

590
00:24:31,200 --> 00:24:33,920
Speaker 2: And that is as a tricky cultural thing to do overnight,

591
00:24:34,039 --> 00:24:36,720
but it's an important part of I think creating that.

592
00:24:36,720 --> 00:24:37,480
Speaker 3: Kind of environment.

593
00:24:37,640 --> 00:24:40,039
Speaker 4: Yeah, if you don't succeed with your first startup, it

594
00:24:40,039 --> 00:24:43,240
shouldn't be a sort of a catastrophic career ending exactly thing,

595
00:24:43,279 --> 00:24:46,000
it should be you know, well good I think should

596
00:24:46,039 --> 00:24:47,960
like should be like, well, you know, you gave it

597
00:24:48,000 --> 00:24:51,079
a good shot, you know, in now try again and exactly. Yeah.

598
00:24:51,160 --> 00:24:53,720
And it's so one thing I'm going to mention is,

599
00:24:53,720 --> 00:24:55,559
like I've seen, creating a company is sort of a

600
00:24:55,599 --> 00:24:59,000
high risk, high reward situation. But I don't know quite

601
00:24:59,000 --> 00:25:01,160
what the how works In the UK, I think probably

602
00:25:01,160 --> 00:25:04,039
better than contenttle Europe, but I'm not sure how does

603
00:25:04,039 --> 00:25:06,160
the new give it. But if somebody's basically going to

604
00:25:06,200 --> 00:25:09,160
risk the life savings and with the beast story of

605
00:25:09,160 --> 00:25:12,119
startups fail, so I mean you hear about the startups

606
00:25:12,160 --> 00:25:15,559
that succeed, but most companies are most startups consist of

607
00:25:16,000 --> 00:25:18,680
you know, a massive amount of work followed by failure.

608
00:25:18,720 --> 00:25:21,640
But that's actually most companies, and so it's a high risk,

609
00:25:21,680 --> 00:25:23,960
high reward and so the higher reward part does need

610
00:25:24,000 --> 00:25:25,519
to be there for it to make sense.

611
00:25:25,680 --> 00:25:28,200
Speaker 2: I think that was a very soft pitch for a

612
00:25:28,279 --> 00:25:31,279
tax policy. I mean, but actually I can tell you

613
00:25:31,279 --> 00:25:33,880
so like AI agree, and we have so we have

614
00:25:34,319 --> 00:25:37,440
I think relative to certainly European countries, but certainly the

615
00:25:37,559 --> 00:25:40,559
US is definitely California a much lower rate of capital

616
00:25:40,599 --> 00:25:43,319
gains tax, right, So for those people who are risking

617
00:25:43,359 --> 00:25:45,519
and growing something like, we think the reward should be

618
00:25:45,519 --> 00:25:47,960
there at the end. So it's twenty percent capital gains

619
00:25:47,960 --> 00:25:48,440
tax right.

620
00:25:48,720 --> 00:25:50,000
Speaker 3: And on stock options, I.

621
00:25:49,960 --> 00:25:52,720
Speaker 2: Don't know if we've got anyone from Index Ventures in

622
00:25:52,759 --> 00:25:55,920
the room. So Index one of our bleeding VC funds here, Okay,

623
00:25:56,079 --> 00:25:59,799
they do a regular report looking at most countries tax

624
00:26:00,000 --> 00:26:02,880
moment of stock options. And you know, when I was

625
00:26:03,039 --> 00:26:05,960
a Chancellor of Treasury sexual equivalent, you know, we were

626
00:26:06,079 --> 00:26:08,200
I think down we were pretty good, but we were

627
00:26:08,200 --> 00:26:10,319
fourth or fifth, and I said we need to for

628
00:26:10,359 --> 00:26:12,440
exactly the reason that you mentioned. I was like, this

629
00:26:12,480 --> 00:26:14,119
has got to be the best place for renovative. We

630
00:26:14,160 --> 00:26:15,759
need to move that up and I think in the

631
00:26:15,799 --> 00:26:18,039
last iteration of that report we had because of the

632
00:26:18,119 --> 00:26:20,559
changes that Jeremy and I had made, we have moved

633
00:26:20,640 --> 00:26:23,680
up to I think second from memory. Hopefully that actould

634
00:26:23,680 --> 00:26:26,240
give you and everyone else some comfort that we recognize.

635
00:26:26,279 --> 00:26:29,160
That's important because when people work hard and risk things, yeah,

636
00:26:29,200 --> 00:26:29,880
they should be able to.

637
00:26:29,880 --> 00:26:31,920
Speaker 4: Enjoy the rewards and pigh award. Yeah.

638
00:26:31,920 --> 00:26:33,680
Speaker 2: And I think we have a We very much have

639
00:26:33,680 --> 00:26:36,079
a tax system that supports that. And those are the

640
00:26:36,119 --> 00:26:38,000
values that you know, I believe in, and I think

641
00:26:38,000 --> 00:26:40,960
most of us in this room probably do as well. Right, next,

642
00:26:41,119 --> 00:26:42,920
next question, I've got seven front of me, and then

643
00:26:42,920 --> 00:26:43,559
I'll come over here.

644
00:26:43,640 --> 00:26:44,880
Speaker 4: Gone, thanks so much.

645
00:26:45,240 --> 00:26:48,480
Speaker 6: We've talked about some really big ideas, global changing ideas.

646
00:26:48,720 --> 00:26:51,240
I'm really interested, particularly in the context of creation of

647
00:26:51,319 --> 00:26:53,920
science and technology superhubs and so on. How does that

648
00:26:53,960 --> 00:26:58,519
map onto the everyday lives of people living in say Austin, Texas,

649
00:26:58,519 --> 00:27:00,799
to choose not aroundin the more in my case Nottingham,

650
00:27:00,880 --> 00:27:03,960
East Midlands. What is how do you see that evolving

651
00:27:03,960 --> 00:27:05,160
for people you know every.

652
00:27:05,000 --> 00:27:07,000
Speaker 4: Day sort of every day effective AI for.

653
00:27:07,079 --> 00:27:11,119
Speaker 2: Context, Elon so Seib runs are equivalent of CDs, right

654
00:27:11,240 --> 00:27:13,720
or Walgreens. So you know, as I visited, right, so

655
00:27:13,799 --> 00:27:16,119
he's got millions of people coming in the shops every day,

656
00:27:16,160 --> 00:27:18,400
and it's making sure how do we make this relevant?

657
00:27:18,400 --> 00:27:21,000
I think that your question how is this relevant to

658
00:27:21,039 --> 00:27:23,279
that person? You know, maybe actually let me go I'll

659
00:27:23,319 --> 00:27:25,440
go first on that because I think it's a fair point.

660
00:27:25,599 --> 00:27:27,079
I was just going over with the team a couple

661
00:27:27,119 --> 00:27:29,119
of things that we're doing because I was saying, how

662
00:27:29,119 --> 00:27:31,279
are we doing AI right now that it's making a

663
00:27:31,279 --> 00:27:34,160
difference to people's lives. And we have this thing called

664
00:27:34,200 --> 00:27:36,839
gov dot which is which actually when we when it

665
00:27:36,880 --> 00:27:39,880
happened several years ago, was a pioneering thing all the

666
00:27:39,920 --> 00:27:43,440
government information together on one website, so you need to

667
00:27:43,440 --> 00:27:46,480
get a driving license, passport, any interaction with government. It

668
00:27:46,599 --> 00:27:50,279
was centralized in a very easy, relatively easy to use way.

669
00:27:50,319 --> 00:27:50,960
Speaker 3: Better than most.

670
00:27:51,319 --> 00:27:54,400
Speaker 2: So we're about to deploy AI across that platform. So

671
00:27:54,440 --> 00:27:57,160
that is something that I think you know, several million

672
00:27:57,200 --> 00:27:59,599
people a day use, right, So a large chunk of

673
00:27:59,599 --> 00:28:02,359
the popular is interacting with gov dot UK every single

674
00:28:02,440 --> 00:28:04,960
day to do all these day to day tasks. Right,

675
00:28:04,960 --> 00:28:06,960
every one of your customers is doing all those things,

676
00:28:07,240 --> 00:28:10,440
and so we're about to deploy AI into that to

677
00:28:10,519 --> 00:28:13,920
make that whole process so much easier, because you know,

678
00:28:13,960 --> 00:28:16,319
some people will be like, look, well I'm currently here

679
00:28:16,519 --> 00:28:19,599
and I've lost my passport and my flights in five hours.

680
00:28:19,680 --> 00:28:21,759
At the moment, that would require you know how many

681
00:28:21,799 --> 00:28:24,200
steps to figure out what you do. Yeah, Actually, when

682
00:28:24,240 --> 00:28:25,880
we deploy the AI, it should be that you could

683
00:28:25,920 --> 00:28:28,519
just literally say that and boom, this is what we're

684
00:28:28,519 --> 00:28:30,359
going to do, walk you through it, and that's going

685
00:28:30,400 --> 00:28:33,000
to benefit millions and millions of people every single day, right,

686
00:28:33,039 --> 00:28:36,000
Because that's a very practical way in my seat, that

687
00:28:36,079 --> 00:28:38,440
I can start using this technology to help people in

688
00:28:38,480 --> 00:28:41,680
their day to day lives, not just healthcare discoveries and

689
00:28:41,720 --> 00:28:43,839
everything else that we're also doing. But I thought that's

690
00:28:43,880 --> 00:28:47,000
quite a powerful demonstration of literally your day to day

691
00:28:47,160 --> 00:28:50,160
customer seeing actually their just day to day life get

692
00:28:50,200 --> 00:28:52,119
a little bit easier because of something that you know,

693
00:28:52,160 --> 00:28:54,039
Elon Demis and others in this room of help.

694
00:28:53,920 --> 00:28:57,279
Speaker 4: Create Yeah, exactly. The most immediate thing is just being

695
00:28:57,319 --> 00:28:59,720
able to ask, like having a very smart friend that

696
00:28:59,720 --> 00:29:02,880
you can ask anything, you know, ask how to make something,

697
00:29:02,960 --> 00:29:05,279
how to solve any problem, and it'll tell you so.

698
00:29:05,720 --> 00:29:08,519
And obviously companies are going to adopt this, so I

699
00:29:08,519 --> 00:29:10,559
think maybe you'll have much better customer service. I guess

700
00:29:10,680 --> 00:29:12,799
essentially that'll probably be the first thing you notice. And

701
00:29:13,359 --> 00:29:16,880
then if we talked about education, so having a tutor,

702
00:29:16,880 --> 00:29:20,160
so if you're trying to understand a subject, like having

703
00:29:20,599 --> 00:29:23,240
a phenomenal tutor on any subject, is that that's really

704
00:29:23,319 --> 00:29:26,480
pretty much there already almost. I mean we need to obviously,

705
00:29:26,599 --> 00:29:29,480
I need to stop hallucinating before you know it can't

706
00:29:29,480 --> 00:29:31,720
give you. I mean that we still have a little

707
00:29:31,720 --> 00:29:33,319
bit of the problem where it can give you an

708
00:29:33,359 --> 00:29:36,319
answer that's confidently wrong with great grammar, you know, bullet

709
00:29:36,319 --> 00:29:38,839
points and everything in citations. It was not real. So

710
00:29:38,880 --> 00:29:40,319
that's to be Okay, we need to make sure it's

711
00:29:40,359 --> 00:29:43,359
not it's not it's not giving you confidently wrong tutor answers.

712
00:29:43,440 --> 00:29:45,839
But but that's going to happen pretty quickly where it

713
00:29:46,079 --> 00:29:47,119
is actually correct.

714
00:29:47,440 --> 00:29:50,640
Speaker 2: Yeah, say for any parent who was homeschooling and realizing

715
00:29:50,720 --> 00:29:53,480
what their kids needed to be helped with, yeah, that will.

716
00:29:53,279 --> 00:29:54,400
Speaker 3: Come as an enormous relief.

717
00:29:54,440 --> 00:29:55,279
Speaker 4: I think they're very good.

718
00:29:55,359 --> 00:29:57,720
Speaker 2: Right, have we got let's go questions over here, who

719
00:29:57,799 --> 00:30:01,319
have got only microphones or brand that perfect.

720
00:30:01,359 --> 00:30:04,759
Speaker 7: Brent Herman, So you know you've spoken eloquently about abundance

721
00:30:04,759 --> 00:30:06,960
in the Age of abundance, so it feels obviously with

722
00:30:07,039 --> 00:30:11,160
AI it's everything everywhere, all at once, but with robots,

723
00:30:11,359 --> 00:30:13,160
and to get the age of abundance will need a

724
00:30:13,160 --> 00:30:15,039
lot of robots. I know you're working out on robots

725
00:30:15,079 --> 00:30:17,680
as well. Are there sort of constraints that we should

726
00:30:17,680 --> 00:30:19,920
think of our politician should be thinking of that we

727
00:30:20,000 --> 00:30:23,079
might get one country might get heavily behind in robots

728
00:30:23,119 --> 00:30:24,240
that can do all these things and end to the

729
00:30:24,279 --> 00:30:26,519
age of the bunds and therefore be it a strategic disadvantage.

730
00:30:26,599 --> 00:30:29,119
Speaker 4: Well, really, anything that can be actuated by a computer

731
00:30:29,519 --> 00:30:32,720
is effectively a robot. So you can think of frankly

732
00:30:32,720 --> 00:30:36,519
Tesla cars or robots on wheels. Anything that's connected to

733
00:30:36,559 --> 00:30:40,519
the Internet is effectively an endpoint actuator for artificial intelligence.

734
00:30:40,720 --> 00:30:43,400
So you've got Bust and Dynamics. Obviously they've been making

735
00:30:43,519 --> 00:30:45,960
impressive robots for a while. I think they are at

736
00:30:45,960 --> 00:30:48,759
this point mostly owned by Hundai, So I guess I

737
00:30:48,799 --> 00:30:51,519
was probably gonna make robots that are humanoid and some

738
00:30:52,200 --> 00:30:54,160
interesting shapes that I wasn't just beating, like the one

739
00:30:54,160 --> 00:30:56,599
that looks like a has wheels and looks sort of

740
00:30:56,599 --> 00:30:58,440
like a kangaroo on wheels. I'm sure what that is,

741
00:30:58,480 --> 00:31:00,640
but it's a little demanded, frankly, but there's going to

742
00:31:00,680 --> 00:31:03,079
be all sorts of robots. You've got the company Dison,

743
00:31:03,440 --> 00:31:07,200
in which I think there's some pretty impressive things, and

744
00:31:07,279 --> 00:31:09,440
I think the UK will not be behind. Actually on

745
00:31:09,799 --> 00:31:12,640
that front, the UK also has ARM, which is really

746
00:31:12,680 --> 00:31:15,319
the best one, one of the best peraps, the best

747
00:31:15,480 --> 00:31:18,400
in trip design in the world. Tesla uses a lot

748
00:31:18,400 --> 00:31:21,920
of ARM technology. Almost everyone is actually, so I have

749
00:31:21,960 --> 00:31:24,319
the UK is in a strong position. Germany obviously makes

750
00:31:24,359 --> 00:31:26,960
a lot of robots, industrial robots. I mean, I think

751
00:31:27,119 --> 00:31:30,279
generally countries that make robots of any kind, even if

752
00:31:30,319 --> 00:31:32,960
they seem somewhat conventional, will be fine. I do think

753
00:31:33,039 --> 00:31:38,160
there is a safety concerned, especially with humanoid robots, because

754
00:31:38,680 --> 00:31:41,039
at least the car can't chase you into this building,

755
00:31:41,160 --> 00:31:43,240
not very easily, you know, or chase you over a tree,

756
00:31:43,359 --> 00:31:44,960
or you know, you can sort of run up a

757
00:31:44,960 --> 00:31:46,559
flight of stairs and get away from a tesla. I

758
00:31:46,559 --> 00:31:48,279
think it's a Stephen King movie about that. If your

759
00:31:48,319 --> 00:31:51,000
car gets possessed so, but if you have a humanoid robot,

760
00:31:51,039 --> 00:31:53,680
it can basically chase you anywhere. So I think we

761
00:31:53,720 --> 00:31:57,480
should have some kind of hardwired local cutoff that you

762
00:31:57,519 --> 00:31:59,839
can't update from the internet. So anything that could be

763
00:32:00,000 --> 00:32:02,559
after updated from the internet obviously could be overridden. But

764
00:32:02,599 --> 00:32:04,920
if you have a local sort of off switch where

765
00:32:04,920 --> 00:32:06,880
you have to say a keyword or something, and then

766
00:32:06,920 --> 00:32:09,240
that puts the robot into a safe state, it's some

767
00:32:09,319 --> 00:32:12,920
kind of localized safe state ability and off switch, you know,

768
00:32:13,039 --> 00:32:14,599
where you don't have to get too close to the robot.

769
00:32:14,839 --> 00:32:16,680
I don't know. So we've got millions of these things

770
00:32:16,720 --> 00:32:17,440
going over the place.

771
00:32:17,599 --> 00:32:20,319
Speaker 3: You're not selling it, just you know, I know.

772
00:32:20,440 --> 00:32:22,400
Speaker 4: I'm saying this is something we should be quite concerned

773
00:32:22,440 --> 00:32:25,039
about because Robert can follow you anywhere. Then you know

774
00:32:25,279 --> 00:32:27,759
what if they just one day get a software update

775
00:32:27,880 --> 00:32:30,000
and they're not so friendly anymore. We've got a James

776
00:32:30,039 --> 00:32:31,039
Cameron movie on our house.

777
00:32:31,079 --> 00:32:33,160
Speaker 2: It's actually, that's it's funny you're saying that, because in

778
00:32:33,200 --> 00:32:34,279
our session.

779
00:32:33,880 --> 00:32:37,799
Speaker 3: That we had today, I just I would say, who they.

780
00:32:37,519 --> 00:32:39,960
Speaker 2: Made exactly the same point right then, So we're talking

781
00:32:40,039 --> 00:32:42,920
about they talking about movies, actually about mentioning James Cameron.

782
00:32:42,920 --> 00:32:45,160
They're talking about James cameraon movies the same. If you

783
00:32:45,160 --> 00:32:48,160
think about it, it's not just those movies, but any of

784
00:32:48,240 --> 00:32:53,279
these movies, trains, subways, metros. They said, all these movies

785
00:32:53,279 --> 00:32:55,920
with the same plot fundamentally all and with the person

786
00:32:56,480 --> 00:32:59,000
turning it off right or finding a way to shut

787
00:32:59,039 --> 00:33:01,079
the thing down. And they were making the same point

788
00:33:01,200 --> 00:33:03,640
that you were about the importance of actual.

789
00:33:03,559 --> 00:33:04,759
Speaker 3: Physical off switches.

790
00:33:05,000 --> 00:33:07,319
Speaker 2: Yeah, and so all the technology is great, but undamentally,

791
00:33:07,400 --> 00:33:10,039
the same movie has played out fifty times. We've all

792
00:33:10,079 --> 00:33:12,480
watched it and it's all fundamentally you know, you know

793
00:33:12,519 --> 00:33:14,880
the point I'm revenge right, all ends in pretty much

794
00:33:14,920 --> 00:33:16,680
the same way, with someone finding their way to just

795
00:33:17,400 --> 00:33:19,200
do the wrong. Which is kind of interesting that you've

796
00:33:19,200 --> 00:33:21,839
said a similar point. Right, it's not the it's not

797
00:33:21,880 --> 00:33:22,960
the obvious place you'd go to.

798
00:33:23,039 --> 00:33:24,880
Speaker 4: But I be one of the tests for the AI,

799
00:33:25,079 --> 00:33:29,000
which is so like blank is your favorite Gamers camera movie? Blank?

800
00:33:29,519 --> 00:33:32,680
Speaker 3: Excellent? Right, Yes, we got over there, yep, perfect, Hi.

801
00:33:32,920 --> 00:33:33,720
Speaker 4: Question for you both.

802
00:33:33,920 --> 00:33:36,440
Speaker 8: So I'm a founder of a AI and mL scale

803
00:33:36,480 --> 00:33:38,599
up in the third Center for AI which is leads

804
00:33:38,599 --> 00:33:40,519
in the North of England and bit biased. Since the

805
00:33:40,599 --> 00:33:43,839
launch of chat GBT. Three months after that, we saw

806
00:33:43,839 --> 00:33:47,240
a real increase in phishing attacks using much more sophisticated

807
00:33:47,279 --> 00:33:51,400
language patterns. What do we do to protect businesses consumers

808
00:33:51,599 --> 00:33:54,079
they trust this technology better, and how do we bring

809
00:33:54,119 --> 00:33:55,319
them along that journey with us?

810
00:33:55,400 --> 00:33:58,039
Speaker 4: Well, I think we shouldn't trusted that much. Actually, it

811
00:33:58,079 --> 00:34:00,559
is actually quite quite a significant allenge because we're getting

812
00:34:00,559 --> 00:34:02,880
to the point where even open source AI can pass

813
00:34:03,319 --> 00:34:06,119
human capture tests. So you know, this is are you're

814
00:34:06,240 --> 00:34:10,239
human identify all the traffic lights in this picture. You're like, okay, yeah,

815
00:34:10,280 --> 00:34:12,719
it's going to have no problem doing that. In fact,

816
00:34:12,760 --> 00:34:14,679
it'll do it better than the human and faster than human.

817
00:34:14,920 --> 00:34:16,519
So we're like, how do you know it's the point

818
00:34:16,599 --> 00:34:19,840
which is a better human better passing human tests than humans?

819
00:34:20,119 --> 00:34:22,840
Then well, what tests actually make sense? That is a

820
00:34:22,880 --> 00:34:24,840
real problem. I don't actually have a good solution to it.

821
00:34:25,199 --> 00:34:25,239
Speaker 7: That.

822
00:34:25,480 --> 00:34:27,079
Speaker 4: One of the things we're trying to figure out on

823
00:34:27,199 --> 00:34:30,239
the X platform is how to deal with that, because

824
00:34:30,519 --> 00:34:32,760
it really we really are at the point where even

825
00:34:32,880 --> 00:34:35,519
with open source you know, readily available AI, you don't

826
00:34:35,559 --> 00:34:37,360
need to be sort of leading in the field. You

827
00:34:37,440 --> 00:34:40,039
can actually be better than humans at passing these tests,

828
00:34:40,480 --> 00:34:42,239
and that's sort of why we think, well, perhaps we

829
00:34:42,280 --> 00:34:44,239
should sort of charge a dollar or a pound a year.

830
00:34:44,480 --> 00:34:46,800
It's a very tiny amount of money, but it's still

831
00:34:46,840 --> 00:34:49,559
makes it privatively expensive to make a million bots, So

832
00:34:50,440 --> 00:34:52,199
especially if you need a million payment methods, then you

833
00:34:52,360 --> 00:34:54,760
run out of sort of stolen credit cards pretty quickly.

834
00:34:55,280 --> 00:34:57,199
So that's that's sort of where we're thinking, like we

835
00:34:57,280 --> 00:34:59,039
might have to sort of just charge some very tiny

836
00:34:59,079 --> 00:35:02,719
amount of money three cents a day effectively to deal

837
00:35:02,800 --> 00:35:07,199
with the onslought of AI power advance. And that is

838
00:35:07,280 --> 00:35:08,800
not a growing problem, but it will be I think,

839
00:35:08,920 --> 00:35:11,679
perhaps an insurmountable problem next year. So and then you

840
00:35:11,760 --> 00:35:15,039
have to worry about, well, manipulation of information is making

841
00:35:15,119 --> 00:35:17,079
something seem very popular when in fact it it's not,

842
00:35:17,559 --> 00:35:20,960
because it's getting boosted by all these likes and reposts

843
00:35:21,000 --> 00:35:23,719
from AI powered barts. So that's why I sort of

844
00:35:23,760 --> 00:35:27,480
think somewhat inevitably it leads to some small payment in

845
00:35:27,639 --> 00:35:30,639
order to dramatically increase the cost of a bot. So

846
00:35:31,360 --> 00:35:33,679
I frankly I think probably any social media system that

847
00:35:33,719 --> 00:35:35,519
doesn't do that will simply be over OneD by bot.

848
00:35:35,679 --> 00:35:38,559
Speaker 2: You know, I think my general answer would be you know,

849
00:35:38,840 --> 00:35:40,880
we need to show that we are on top of

850
00:35:41,000 --> 00:35:44,519
mitigating the risks right so people can trust the technology.

851
00:35:44,599 --> 00:35:46,320
That's what actually the last couple of days has been

852
00:35:46,360 --> 00:35:49,039
about on the Safety Summit is just showing, you know,

853
00:35:49,119 --> 00:35:52,239
we're investing in the Safety Institute, having the people who

854
00:35:52,280 --> 00:35:54,320
can do the research on these things to figure out

855
00:35:54,320 --> 00:35:57,119
how we mitigate against them, and we have to do

856
00:35:57,199 --> 00:36:00,159
it fast and we have to keep iterating it is.

857
00:36:00,519 --> 00:36:02,079
All of us probably in this room, believe that the

858
00:36:02,119 --> 00:36:04,800
technology can be incredibly powerful, but we've got to make

859
00:36:04,800 --> 00:36:06,679
sure we bring people along that journey with us, that

860
00:36:06,800 --> 00:36:09,960
we're handling the risks that are there and as there's

861
00:36:09,960 --> 00:36:12,159
a job to do, and the last couple of days,

862
00:36:12,400 --> 00:36:14,280
I think we make good progress on it because we

863
00:36:14,360 --> 00:36:16,880
want to focus on the positives and manage these things.

864
00:36:16,960 --> 00:36:19,960
But that requires action, and that's what the last couple

865
00:36:19,960 --> 00:36:22,400
of days has been about. Your story, you know, analogy.

866
00:36:22,480 --> 00:36:25,400
There was part of the research that actually, you know,

867
00:36:25,480 --> 00:36:28,880
the team working on the task force here published and

868
00:36:29,039 --> 00:36:31,320
presented yes day. I don't know if you saw it was,

869
00:36:31,480 --> 00:36:34,920
which is essentially that it was using AI to do

870
00:36:35,719 --> 00:36:38,960
to create a ton of fake profiles on social media

871
00:36:39,480 --> 00:36:43,679
and then infiltrate particular groups with particular information. And actually,

872
00:36:43,840 --> 00:36:45,119
at the moment that, as I said, to your point,

873
00:36:45,159 --> 00:36:46,679
and there's a cost free.

874
00:36:47,199 --> 00:36:48,920
Speaker 4: It's getting to the point where it's like, really, you're

875
00:36:48,920 --> 00:36:51,199
going to have one hundred for a penny sort of thing. Ridiculous.

876
00:36:51,239 --> 00:36:52,920
Speaker 2: And if you think about some of these social networks

877
00:36:52,920 --> 00:36:55,199
at quite a neighborhood or town level, it's not that

878
00:36:55,320 --> 00:36:59,239
many fake profiles. Quickly suddenly they're everywhere, and there's some

879
00:36:59,360 --> 00:37:01,559
local issue that might be of importance, and you know,

880
00:37:01,639 --> 00:37:04,239
the team of run versions of how that would look like,

881
00:37:04,280 --> 00:37:06,480
and suddenly they're interacting with everybody and then spreading and

882
00:37:06,519 --> 00:37:09,760
misinformation around. Yeah, challenge we literally as part of the

883
00:37:09,800 --> 00:37:12,880
research that we published on this information yesterday. It's a

884
00:37:12,960 --> 00:37:13,559
real challenge.

885
00:37:13,719 --> 00:37:17,039
Speaker 4: Yeah, exactly to your point. I mean, the images. You

886
00:37:17,039 --> 00:37:20,320
don't even need to steal somebody's picture because then that's traceable.

887
00:37:20,360 --> 00:37:23,039
But you can actually just say, create a new image

888
00:37:23,039 --> 00:37:26,559
of a person, realistic looking it doesn't exist, and then

889
00:37:26,840 --> 00:37:30,360
create a biography realistic but doesn't exist, and do that

890
00:37:30,639 --> 00:37:32,480
en mass and practically the only way of able to

891
00:37:32,480 --> 00:37:34,039
tell us that the grammar is too good. Don't give

892
00:37:34,039 --> 00:37:35,840
away typhos come on.

893
00:37:36,079 --> 00:37:37,960
Speaker 3: Now, I'm getting waved at because I think we are

894
00:37:38,079 --> 00:37:39,599
out of time. I don't we take one.

895
00:37:39,800 --> 00:37:42,960
Speaker 2: Very brief last question and let's make it go on? Yes, sir, going,

896
00:37:43,000 --> 00:37:44,480
you're right in front of me, go on.

897
00:37:44,639 --> 00:37:47,039
Speaker 9: Question for you related to the X platform. Are there

898
00:37:47,079 --> 00:37:49,039
simple things we can do, especially when it comes to

899
00:37:49,119 --> 00:37:51,800
visual media. You alluded to the fact that it's fairly

900
00:37:51,840 --> 00:37:55,039
straightforward and effectively free to make people like yourselves say

901
00:37:55,360 --> 00:37:57,199
and do things that you never said or did. Can

902
00:37:57,239 --> 00:37:59,719
we do something like cryptographically signed media? I'm from Adobe

903
00:38:00,039 --> 00:38:02,679
working on this project. Twitter was a member. Love to

904
00:38:02,719 --> 00:38:05,360
see X come back. Digitally sign media to indicate not

905
00:38:05,480 --> 00:38:07,400
only what was created by AI, but what came from

906
00:38:07,440 --> 00:38:09,760
a camera, what was real to imview, a sense of

907
00:38:09,840 --> 00:38:11,719
trust in media that can go viral.

908
00:38:11,880 --> 00:38:13,960
Speaker 4: That sounds like a good idea. Actually, so if some

909
00:38:14,079 --> 00:38:17,519
way of authenticating would be good. So, yeah, that sounds

910
00:38:17,519 --> 00:38:18,559
like a good idea, which probably do it.

911
00:38:18,760 --> 00:38:21,760
Speaker 2: There you go, and actually on that on that point already,

912
00:38:21,800 --> 00:38:24,119
And this is particularly partner for people in my job,

913
00:38:24,360 --> 00:38:27,039
right and I've already had a situation happened to me

914
00:38:27,159 --> 00:38:30,480
with adopted image that goes everywhere negative.

915
00:38:30,159 --> 00:38:31,960
Speaker 3: By the time everyone realizes well.

916
00:38:31,880 --> 00:38:35,480
Speaker 2: That's fake and we should stop sending it the damages damage.

917
00:38:35,480 --> 00:38:37,920
And actually we were again reflecting today. If you think

918
00:38:38,039 --> 00:38:41,000
next year, you've got elections in you know, I think

919
00:38:41,119 --> 00:38:44,320
you know, the US, India, I think Indonesia probably here

920
00:38:44,519 --> 00:38:48,119
there you go for massive news and actually you've got

921
00:38:48,440 --> 00:38:51,280
just an enormous junk of the world's population is voting

922
00:38:51,360 --> 00:38:54,320
next year, right, and you've got EU elections as well.

923
00:38:54,920 --> 00:38:55,079
Speaker 4: You know.

924
00:38:55,159 --> 00:38:57,880
Speaker 2: Actually, just these issues are right in front of ours.

925
00:38:58,440 --> 00:39:00,760
Next year is where a big election across the globe,

926
00:39:00,760 --> 00:39:03,239
probably the first set of elections where this has been

927
00:39:03,280 --> 00:39:06,519
a real issue. Yeah, So figuring out how we manage

928
00:39:06,559 --> 00:39:09,000
that is I think kind of mission critical for the

929
00:39:09,000 --> 00:39:11,199
people who want the integrity of our democracy.

930
00:39:11,400 --> 00:39:13,800
Speaker 4: Yeah, I mean some of it it's quite entertaining, Like

931
00:39:14,000 --> 00:39:15,760
the pope and the puffer jacket. Have you seen that one?

932
00:39:16,280 --> 00:39:18,199
That's amazing. But I mean I still write too people

933
00:39:18,199 --> 00:39:20,360
who think that's real. I'm like, well, one of the

934
00:39:20,400 --> 00:39:23,800
answer is wearing a puffer jacket in July, right, be sweating,

935
00:39:23,840 --> 00:39:26,960
but it actually looks quite quite dashing of things. In fact,

936
00:39:27,000 --> 00:39:28,719
I think AI fashion is going to be a real thing.

937
00:39:28,760 --> 00:39:30,760
So a bit of doing gloom, like we learn in

938
00:39:30,800 --> 00:39:32,559
the most interesting times. And I think this is it is,

939
00:39:33,039 --> 00:39:35,119
you know, like eighty percent likely to be good and

940
00:39:35,199 --> 00:39:38,679
twenty percent bad. I think if we're cognizant and careful

941
00:39:38,679 --> 00:39:41,000
about the bad part, on balance, actually it will be

942
00:39:41,119 --> 00:39:43,199
the future that we want or for the future that

943
00:39:43,519 --> 00:39:46,840
is preferable, and it actually will be somewhat of a leveler,

944
00:39:47,199 --> 00:39:49,159
an equalizer in the sense that you know, I think

945
00:39:49,199 --> 00:39:51,559
everyone will have access to goods and services and education,

946
00:39:52,199 --> 00:39:54,639
and so you know, I think probably it leads to

947
00:39:54,840 --> 00:39:57,440
more human happiness. So I guess i'd probably leave on

948
00:39:57,480 --> 00:39:58,400
an optimistic note.

949
00:39:58,440 --> 00:39:59,920
Speaker 3: Tough act. Yeah, well, that's it.

950
00:40:00,239 --> 00:40:01,840
Speaker 2: That is a great note to end on. I think

951
00:40:01,880 --> 00:40:04,960
that we all want that better future. I think it's

952
00:40:05,000 --> 00:40:06,800
that the promise of it is certainly there. Lots of

953
00:40:06,840 --> 00:40:09,199
people in this room, including yourselves, are working hard to

954
00:40:09,239 --> 00:40:10,880
make it happen. Our job in government is to make

955
00:40:10,920 --> 00:40:13,239
sure it happens safely. But on the basis of this

956
00:40:13,320 --> 00:40:15,920
conversation in the last couple of days, I'm certainly leaving

957
00:40:16,039 --> 00:40:18,800
more confident that we can make that happen. It's been

958
00:40:18,920 --> 00:40:20,480
a huge privilege and pleasure to have you here.

959
00:40:20,920 --> 00:40:21,480
Speaker 4: Thank you very much.

960
00:40:21,480 --> 00:40:35,760
Speaker 1: For me, Thanks for listening, See you in the next episode.