In this episode #16 of Are You Human, I have a pleasure to speak with Joanna Bryson, the Professor of Ethics and Technology at the Hertie School. Her research focuses on the impact of technology on human cooperation, and AI/ICT governance. We discuss a broad range of topics, exploring her expertise in ethics and technology; her childhood interests in animal behaviour and how it influenced her path into AI and behavioural sciences. Joanna shares the risks of anthropomorphising AI and the potential dangers of giving AI citizen rights, which could lead to corruption in our justice system. She stresses the importance of ensuring that the benefits of AI are widely distributed, not just concentrated among a powerful few, and highlighted the need for holding stakeholders accountable for AI’s actions and consequences. We also touch on the existential risks associated with AI, such as the concentration of power, lack of responsibility, and the prevalence of surveillance. Joanna shares her views on the current state and future of AGI, emphasising the combinatorial power of human-AI collaboration. I hope you’ll enjoy this episode as much as I did.

Transcript

1
00:00:00,319 –> 00:00:05,160
AGI it depends what you define how you

2
00:00:02,760 –> 00:00:06,759
mean it um when I first heard the term

3
00:00:05,160 –> 00:00:09,320
it was people that were saying oh you

4
00:00:06,759 –> 00:00:11,080
know nobody working in AI is actually

5
00:00:09,320 –> 00:00:12,240
trying to solve all the problems they’re

6
00:00:11,080 –> 00:00:13,759
you know they’re all just trying to do

7
00:00:12,240 –> 00:00:15,360
Machine Vision or something and that’s

8
00:00:13,759 –> 00:00:16,960
boring and we want to get back to more

9
00:00:15,360 –> 00:00:18,880
human you know really understand human

10
00:00:16,960 –> 00:00:21,279
intelligence and yeah know that was

11
00:00:18,880 –> 00:00:21,279
completely

12
00:00:26,039 –> 00:00:30,039
false there were very good reasons a lot

13
00:00:28,439 –> 00:00:32,279
of people were trying to solve the whole

14
00:00:30,039 –> 00:00:35,480
problem of intelligence and and and it

15
00:00:32,279 –> 00:00:37,040
was it was just this false narrative um

16
00:00:35,480 –> 00:00:40,039
and then people started talking about oh

17
00:00:37,040 –> 00:00:41,840
AGI there’s like this um there’s going

18
00:00:40,039 –> 00:00:43,600
to be this intelligence explosion when

19
00:00:41,840 –> 00:00:45,440
you have systems that can learn how to

20
00:00:43,600 –> 00:00:47,559
learn and then there’s going to be this

21
00:00:45,440 –> 00:00:50,079
exponential and exponential growth and

22
00:00:47,559 –> 00:00:54,359
we will be mere cockroaches after that

23
00:00:50,079 –> 00:00:56,160
well I think that is at least logical um

24
00:00:54,359 –> 00:00:58,320
but again it doesn’t make sense to talk

25
00:00:56,160 –> 00:01:01,160
about the system itself and if you

26
00:00:58,320 –> 00:01:02,879
actually look at the number of people on

27
00:01:01,160 –> 00:01:05,040
the planet that’s what’s happened since

28
00:01:02,879 –> 00:01:06,680
we’ve had writing so since we’ve been

29
00:01:05,040 –> 00:01:09,920
able to improve our own intelligence

30
00:01:06,680 –> 00:01:11,560
through our technology um we have been

31
00:01:09,920 –> 00:01:14,479
that explosion and that’s why we have a

32
00:01:11,560 –> 00:01:14,479
climate crisis right

33
00:01:15,320 –> 00:01:18,720
now are you

34
00:01:27,720 –> 00:01:32,560
human welcome to are you here human

35
00:01:30,280 –> 00:01:35,200
podcast on this episode you’ll hear a

36
00:01:32,560 –> 00:01:37,720
conversation I have with Joanna Bryson

37
00:01:35,200 –> 00:01:40,479
Joanna is a professor at the hery school

38
00:01:37,720 –> 00:01:43,040
in Berlin where she works on artificial

39
00:01:40,479 –> 00:01:46,920
intelligence ethics and collaborative

40
00:01:43,040 –> 00:01:50,240
cognition Joanna earned her PhD from MIT

41
00:01:46,920 –> 00:01:54,000
in 2001 and has since held positions in

42
00:01:50,240 –> 00:01:56,799
University of bath nothingham and Oxford

43
00:01:54,000 –> 00:02:00,320
University uh a thought leader in AI

44
00:01:56,799 –> 00:02:02,759
ethics of human AI relationships J

45
00:02:00,320 –> 00:02:06,920
published the influential influential

46
00:02:02,759 –> 00:02:10,679
essay robots should be slaved slaves and

47
00:02:06,920 –> 00:02:12,640
helped Define principles of Robotics her

48
00:02:10,679 –> 00:02:15,120
research has appeared in Science

49
00:02:12,640 –> 00:02:18,680
magazine and she consulted organizations

50
00:02:15,120 –> 00:02:21,480
like redcross uh on autonomous

51
00:02:18,680 –> 00:02:23,400
autonomous weapons Joanna’s

52
00:02:21,480 –> 00:02:25,640
contributions to the field have been

53
00:02:23,400 –> 00:02:28,640
recognized with an outstanding

54
00:02:25,640 –> 00:02:30,879
Achievement Award from cognition X in

55
00:02:28,640 –> 00:02:34,480
2017

56
00:02:30,879 –> 00:02:36,879
I’m excited to have her joining us today

57
00:02:34,480 –> 00:02:39,519
uh to share her expertise on this very

58
00:02:36,879 –> 00:02:43,800
crucial topic

59
00:02:39,519 –> 00:02:45,560
enjoy Joanna so nice to finally meet you

60
00:02:43,800 –> 00:02:47,720
and thank you thank you so much for

61
00:02:45,560 –> 00:02:49,080
finding time to do this nice to meet you

62
00:02:47,720 –> 00:02:52,959
too

63
00:02:49,080 –> 00:02:55,280
camela where do I even begin there is so

64
00:02:52,959 –> 00:02:57,840
much so much things you you’ve written

65
00:02:55,280 –> 00:03:01,319
about so many speeches I had to go

66
00:02:57,840 –> 00:03:04,000
through I wanted to um uh so definitely

67
00:03:01,319 –> 00:03:06,680
we will just cover a little bit of of

68
00:03:04,000 –> 00:03:09,560
what you’ve been working on and and

69
00:03:06,680 –> 00:03:13,040
doing but I guess the easiest would be

70
00:03:09,560 –> 00:03:15,480
to start from the beginning so your

71
00:03:13,040 –> 00:03:18,920
childhood uh on one of the

72
00:03:15,480 –> 00:03:24,280
interviews on one of the interviews you

73
00:03:18,920 –> 00:03:27,280
said that you became interested um in

74
00:03:24,280 –> 00:03:30,760
animal intelligence that’s what led you

75
00:03:27,280 –> 00:03:33,760
later to to studying behavioral um

76
00:03:30,760 –> 00:03:38,080
like Sciences U psychology and and

77
00:03:33,760 –> 00:03:41,000
eventually uh working on AI but what was

78
00:03:38,080 –> 00:03:42,439
the the spark what was the uh the

79
00:03:41,000 –> 00:03:45,040
trigger

80
00:03:42,439 –> 00:03:48,760
to like what what what made you

81
00:03:45,040 –> 00:03:51,439
interested in this you know I I it’s

82
00:03:48,760 –> 00:03:53,920
hard to say that I mean what I just find

83
00:03:51,439 –> 00:03:55,840
uh well life interesting but Behavior

84
00:03:53,920 –> 00:03:58,599
interesting I think differen is

85
00:03:55,840 –> 00:04:01,760
interesting and we I was fortunate to

86
00:03:58,599 –> 00:04:03,360
live my P my family backed up uh moved

87
00:04:01,760 –> 00:04:05,239
into a place that backed up onto a

88
00:04:03,360 –> 00:04:08,040
little Pond I mean it was that we called

89
00:04:05,239 –> 00:04:10,319
it the Lagoon but it was really it it

90
00:04:08,040 –> 00:04:12,079
was storm it was uh water from the

91
00:04:10,319 –> 00:04:13,959
storms from the street you know that

92
00:04:12,079 –> 00:04:16,079
went into this little but there was

93
00:04:13,959 –> 00:04:19,239
croads and you know we had snakes and

94
00:04:16,079 –> 00:04:20,600
frogs and whatever and um yeah I was

95
00:04:19,239 –> 00:04:21,919
just fascinated by all these different

96
00:04:20,600 –> 00:04:23,520
things and I actually think it’s a real

97
00:04:21,919 –> 00:04:25,520
shame now that people don’t have the

98
00:04:23,520 –> 00:04:26,720
experience of other living intelligent

99
00:04:25,520 –> 00:04:28,120
beings you know it would have been

100
00:04:26,720 –> 00:04:29,720
totally different to grow up in a time

101
00:04:28,120 –> 00:04:31,960
with horses and dog I mean we do have do

102
00:04:29,720 –> 00:04:35,000
dogs kind of but they’re really obedient

103
00:04:31,960 –> 00:04:37,440
so I think I think we have

104
00:04:35,000 –> 00:04:40,440
uh uh we don’t see that many animals in

105
00:04:37,440 –> 00:04:42,440
their natural uh ecosystem anymore but I

106
00:04:40,440 –> 00:04:44,560
was also interested in people but I

107
00:04:42,440 –> 00:04:45,800
figured out that when I told my parents

108
00:04:44,560 –> 00:04:47,080
you know things that I thought about

109
00:04:45,800 –> 00:04:48,479
animals that they thought that was

110
00:04:47,080 –> 00:04:49,800
interesting and if I told them things I

111
00:04:48,479 –> 00:04:53,120
thought about people they said I was

112
00:04:49,800 –> 00:04:55,479
wrong and so I figured out people were

113
00:04:53,120 –> 00:04:57,840
defensive one thing that’s funny and

114
00:04:55,479 –> 00:04:59,400
relevant to AI that as L were talking

115
00:04:57,840 –> 00:05:01,560
about my childhood which most people

116
00:04:59,400 –> 00:05:04,120
don’t ask about I I was just thinking

117
00:05:01,560 –> 00:05:07,080
the other day and I don’t know why about

118
00:05:04,120 –> 00:05:08,800
um about how I used to worry about

119
00:05:07,080 –> 00:05:10,880
whether if my stuffed animals were alive

120
00:05:08,800 –> 00:05:13,280
they would like me or not because people

121
00:05:10,880 –> 00:05:15,080
real living people did not like me and

122
00:05:13,280 –> 00:05:17,400
and I just assumed all my animals would

123
00:05:15,080 –> 00:05:19,680
like me but I didn’t know if it was

124
00:05:17,400 –> 00:05:22,960
true yeah actually I I played very

125
00:05:19,680 –> 00:05:25,360
similar game where I was um letting my

126
00:05:22,960 –> 00:05:27,680
cousin who was more more or less my age

127
00:05:25,360 –> 00:05:30,080
uh believe that when she closes the her

128
00:05:27,680 –> 00:05:33,240
eyes uh all the stuffed animals actually

129
00:05:30,080 –> 00:05:36,680
alive and that she has to behave uh for

130
00:05:33,240 –> 00:05:38,280
them to to to like her yes I don’t know

131
00:05:36,680 –> 00:05:39,880
if she still believes

132
00:05:38,280 –> 00:05:43,600
it

133
00:05:39,880 –> 00:05:47,000
maybe um okay and so so then what led

134
00:05:43,600 –> 00:05:48,479
you to you know to study this um

135
00:05:47,000 –> 00:05:50,199
professionally I mean the first time I

136
00:05:48,479 –> 00:05:52,160
really got to do it I I would love to

137
00:05:50,199 –> 00:05:53,440
have done it at school but most you know

138
00:05:52,160 –> 00:05:56,240
High School grade school that you don’t

139
00:05:53,440 –> 00:05:59,639
get to do that except at science fair so

140
00:05:56,240 –> 00:06:02,560
I did study uh uh dog communication at

141
00:05:59,639 –> 00:06:04,960
for sign and I got to uh I got a I got a

142
00:06:02,560 –> 00:06:08,199
first at the Illinois state level but

143
00:06:04,960 –> 00:06:11,599
didn’t get to go to Nationals um and uh

144
00:06:08,199 –> 00:06:13,680
I also studied um snail intelligence oh

145
00:06:11,599 –> 00:06:16,199
wow what’s what’s what’s interesting

146
00:06:13,680 –> 00:06:17,720
about that well I they just I read

147
00:06:16,199 –> 00:06:19,759
somewhere that worms could solve a te-

148
00:06:17,720 –> 00:06:21,400
maze and I wondered if snails could so

149
00:06:19,759 –> 00:06:23,199
we we found out that they tended to

150
00:06:21,400 –> 00:06:26,880
crawl out of the maze it wasn’t very

151
00:06:23,199 –> 00:06:29,960
exciting anyway okay so then I went to a

152
00:06:26,880 –> 00:06:31,599
University of Chicago and uh and I

153
00:06:29,960 –> 00:06:33,240
actually my best teacher at high school

154
00:06:31,599 –> 00:06:35,919
had been physics so I actually started

155
00:06:33,240 –> 00:06:38,039
out in physics but I wasn’t as thrilled

156
00:06:35,919 –> 00:06:39,560
to do because I managed to have three

157
00:06:38,039 –> 00:06:41,960
years of physics at high school two

158
00:06:39,560 –> 00:06:44,039
years of kempis and one year physics um

159
00:06:41,960 –> 00:06:45,800
so I wasn’t that thrilled to do sort of

160
00:06:44,039 –> 00:06:48,479
all the same physics over again only

161
00:06:45,800 –> 00:06:49,919
with Calculus I found that uh tedious

162
00:06:48,479 –> 00:06:52,360
and then I remembered that I’d always

163
00:06:49,919 –> 00:06:54,000
wanted to do um behavior and and there

164
00:06:52,360 –> 00:06:56,800
was something called Behavioral Science

165
00:06:54,000 –> 00:06:58,599
there um which at that time it’s really

166
00:06:56,800 –> 00:07:00,759
annoying they’ve changed what it means

167
00:06:58,599 –> 00:07:02,599
but at that time it the non-clinical

168
00:07:00,759 –> 00:07:05,479
psychology and and of all its

169
00:07:02,599 –> 00:07:07,680
disciplines so I I particularly liked um

170
00:07:05,479 –> 00:07:09,720
you know behavioral ecology um I don’t

171
00:07:07,680 –> 00:07:10,840
think it was called that I anyway there

172
00:07:09,720 –> 00:07:13,759
was something about the evolution of

173
00:07:10,840 –> 00:07:15,160
behavior and there was also um a course

174
00:07:13,759 –> 00:07:18,039
uh called developmental neuros

175
00:07:15,160 –> 00:07:20,120
psychology so really how how children

176
00:07:18,039 –> 00:07:22,840
start how how it works how children

177
00:07:20,120 –> 00:07:25,919
start working and you know various other

178
00:07:22,840 –> 00:07:28,160
uh neuroscience and and um ecological

179
00:07:25,919 –> 00:07:29,800
perspectives as well as you know one

180
00:07:28,160 –> 00:07:33,680
some social psychology some Clin iCal

181
00:07:29,800 –> 00:07:35,680
psychology MH and you because I remember

182
00:07:33,680 –> 00:07:38,319
you mentioned in some of the interviews

183
00:07:35,680 –> 00:07:42,080
I saw um that you were also considering

184
00:07:38,319 –> 00:07:44,120
neuro uh science right um I never

185
00:07:42,080 –> 00:07:45,840
seriously took consider Neuroscience

186
00:07:44,120 –> 00:07:47,800
because it’s unfortunate but in the in

187
00:07:45,840 –> 00:07:51,520
the United States if you have anything

188
00:07:47,800 –> 00:07:55,840
like that um you wind up fighting with

189
00:07:51,520 –> 00:07:57,879
uh with preds and for grades and and it

190
00:07:55,840 –> 00:08:00,199
just so it just puts people off from

191
00:07:57,879 –> 00:08:02,080
studying even botney you you just can’t

192
00:08:00,199 –> 00:08:04,240
get near the preds without having a

193
00:08:02,080 –> 00:08:08,240
miserable life that they put on

194
00:08:04,240 –> 00:08:10,960
themselves um so so I don’t that I I did

195
00:08:08,240 –> 00:08:12,800
think about it when I was at MIT uh

196
00:08:10,960 –> 00:08:15,680
because I I attended all the

197
00:08:12,800 –> 00:08:17,800
Neuroscience talks um and I thought

198
00:08:15,680 –> 00:08:19,759
about trying to switch over to that but

199
00:08:17,800 –> 00:08:21,120
in the end of the day I I was barely

200
00:08:19,759 –> 00:08:22,520
making it in computer science which I

201
00:08:21,120 –> 00:08:25,080
knew something about so I thought I

202
00:08:22,520 –> 00:08:28,319
should better stay there um but I did I

203
00:08:25,080 –> 00:08:30,800
I have been in um uh technically I’ve

204
00:08:28,319 –> 00:08:33,320
been in supposedly beh Neuroscience Labs

205
00:08:30,800 –> 00:08:35,399
twice both both with non-human primates

206
00:08:33,320 –> 00:08:36,680
and both with nothing um invasive

207
00:08:35,399 –> 00:08:39,080
happening so I don’t know why they were

208
00:08:36,680 –> 00:08:42,519
called that um but both at Harvard and

209
00:08:39,080 –> 00:08:44,399
and uh sh and at MIT I went to a lot of

210
00:08:42,519 –> 00:08:47,720
Neuroscience talks if I could Chicago I

211
00:08:44,399 –> 00:08:49,640
took the classes I could in it um and uh

212
00:08:47,720 –> 00:08:52,480
yeah they didn’t have anything like that

213
00:08:49,640 –> 00:08:54,480
at at uh bath at all and of course har

214
00:08:52,480 –> 00:08:57,000
school doesn’t have any neural science

215
00:08:54,480 –> 00:08:58,839
um but um but you know whenever I I get

216
00:08:57,000 –> 00:09:00,440
an opportunity I I love going to that

217
00:08:58,839 –> 00:09:04,000
it’s the same with like microbiology

218
00:09:00,440 –> 00:09:06,519
that it’s the evolution of behavior at

219
00:09:04,000 –> 00:09:07,880
any level isconnected yeah if you’re

220
00:09:06,519 –> 00:09:09,360
trying to understand machine learning

221
00:09:07,880 –> 00:09:11,560
and you don’t think it’s relevant that

222
00:09:09,360 –> 00:09:13,760
bacteria will sacrifice themselves for

223
00:09:11,560 –> 00:09:17,160
each other then you’re missing something

224
00:09:13,760 –> 00:09:18,920
you know uh you know so I I really loved

225
00:09:17,160 –> 00:09:21,240
uh all the opportunities I had that was

226
00:09:18,920 –> 00:09:23,240
something be was it worked well because

227
00:09:21,240 –> 00:09:25,760
it’s relevant to antibiotic resistance

228
00:09:23,240 –> 00:09:27,480
and things so a lot of good work on

229
00:09:25,760 –> 00:09:30,399
altruism and Cooperative behavior and

230
00:09:27,480 –> 00:09:31,200
things uh not only in uh but also in

231
00:09:30,399 –> 00:09:34,760
fruit

232
00:09:31,200 –> 00:09:36,279
flies but at bath but yeah people in in

233
00:09:34,760 –> 00:09:38,560
the UK at that time were it was very

234
00:09:36,279 –> 00:09:40,920
hard for them to study anything much

235
00:09:38,560 –> 00:09:44,959
bigger than fruit fly and and why did

236
00:09:40,920 –> 00:09:47,760
you um choose the academic uh Academia

237
00:09:44,959 –> 00:09:49,480
path rather than you know I always

238
00:09:47,760 –> 00:09:51,519
wanted that I mean I originally wanted

239
00:09:49,480 –> 00:09:54,560
to be a paleontologist but always always

240
00:09:51,519 –> 00:09:56,000
wanted to be a scientist um yeah so

241
00:09:54,560 –> 00:09:57,720
actually the the more interesting thing

242
00:09:56,000 –> 00:09:58,959
was that how I wound up in industry for

243
00:09:57,720 –> 00:10:00,640
a few years and that was because I

244
00:09:58,959 –> 00:10:02,920
couldn’t choose which discipline to go

245
00:10:00,640 –> 00:10:05,000
into so I thought I should pay off my

246
00:10:02,920 –> 00:10:07,360
debts and and see if I could make it in

247
00:10:05,000 –> 00:10:08,920
rock and roll and figure out which

248
00:10:07,360 –> 00:10:10,760
science I missed the most and then I

249
00:10:08,920 –> 00:10:12,760
still couldn’t figure out so I took AI

250
00:10:10,760 –> 00:10:14,079
is sort of halfway in between what I

251
00:10:12,760 –> 00:10:16,399
really cared about I still cared about

252
00:10:14,079 –> 00:10:17,600
in Behavior but what I was good at I

253
00:10:16,399 –> 00:10:19,160
knew I was an exceptionally good

254
00:10:17,600 –> 00:10:20,880
programmer and and I figured it would

255
00:10:19,160 –> 00:10:22,160
help me get into a good degree if I did

256
00:10:20,880 –> 00:10:23,839
what I was exceptionally good at and

257
00:10:22,160 –> 00:10:25,560
indeed I got into MIT so I couldn’t have

258
00:10:23,839 –> 00:10:29,079
been happier about

259
00:10:25,560 –> 00:10:31,640
that but you yeah but you end but you

260
00:10:29,079 –> 00:10:35,120
end ended up in in the right uh the

261
00:10:31,640 –> 00:10:37,920
prime time right now uh for AI so I’ve

262
00:10:35,120 –> 00:10:41,079
been in AI since uh well arguably since

263
00:10:37,920 –> 00:10:43,639
1986 I I tutored the first class was

264
00:10:41,079 –> 00:10:46,160
taught by uh what was his name Chris hen

265
00:10:43,639 –> 00:10:48,600
or something like that that they that

266
00:10:46,160 –> 00:10:50,160
the University of Chicago offered an AI

267
00:10:48,600 –> 00:10:51,519
um so I’ve been in it for a very long

268
00:10:50,160 –> 00:10:52,959
time and people used to tell me I should

269
00:10:51,519 –> 00:10:55,920
take it off my CV because it made me

270
00:10:52,959 –> 00:10:57,760
look dumb you know so now now the time

271
00:10:55,920 –> 00:10:59,800
has come to me but it’s not like now you

272
00:10:57,760 –> 00:11:01,959
made it involved

273
00:10:59,800 –> 00:11:03,760
I’m like unlike some people who who are

274
00:11:01,959 –> 00:11:06,519
are suddenly AI

275
00:11:03,760 –> 00:11:09,600
experts yeah it happens with all the you

276
00:11:06,519 –> 00:11:11,120
know big um topics right like we had

277
00:11:09,600 –> 00:11:12,600
interesting you know I was the kind like

278
00:11:11,120 –> 00:11:14,680
when I played soccer I was the kind of

279
00:11:12,600 –> 00:11:16,600
kid that that tried to go find a big

280
00:11:14,680 –> 00:11:18,240
open space in the field so that if the

281
00:11:16,600 –> 00:11:20,480
ball went anywhere I might get to do it

282
00:11:18,240 –> 00:11:22,560
I didn’t fight other kids for the ball

283
00:11:20,480 –> 00:11:26,200
and I was like that in in Academia for a

284
00:11:22,560 –> 00:11:28,600
long time too and then um s suddenly

285
00:11:26,200 –> 00:11:30,160
other stuff started coming at me and and

286
00:11:28,600 –> 00:11:31,880
towards me and it’s interesting because

287
00:11:30,160 –> 00:11:34,079
it’s not just the other academics it’s

288
00:11:31,880 –> 00:11:36,440
also like sudden there’s like almost

289
00:11:34,079 –> 00:11:38,959
like flies swarming around you the uh

290
00:11:36,440 –> 00:11:40,720
the conses and things so all these start

291
00:11:38,959 –> 00:11:42,079
these weird things start happening and

292
00:11:40,720 –> 00:11:43,760
trying to figure out what matters and

293
00:11:42,079 –> 00:11:45,760
what doesn’t but at some point I just

294
00:11:43,760 –> 00:11:47,880
realized it was it was time to start you

295
00:11:45,760 –> 00:11:50,040
know stop avoiding it that I I I had

296
00:11:47,880 –> 00:11:52,760
enough uh confidence and seniority that

297
00:11:50,040 –> 00:11:54,720
I could stick in something that was

298
00:11:52,760 –> 00:11:56,519
popular sometimes feel like maybe I’m

299
00:11:54,720 –> 00:11:58,040
not putting enough effort into you know

300
00:11:56,519 –> 00:12:00,560
like I see people self-promoting right

301
00:11:58,040 –> 00:12:02,560
over the top of my work and I think you

302
00:12:00,560 –> 00:12:03,959
know what should I be doing that but I’m

303
00:12:02,560 –> 00:12:06,200
just too interested I’ve always been

304
00:12:03,959 –> 00:12:09,399
interested and so I i’ I chased the

305
00:12:06,200 –> 00:12:12,839
science more no no it’s amazing because

306
00:12:09,399 –> 00:12:15,199
you right most of the scientists

307
00:12:12,839 –> 00:12:17,600
unfortunately you know turn to the other

308
00:12:15,199 –> 00:12:20,639
side so they they become advisors or

309
00:12:17,600 –> 00:12:22,440
they they just you know go on AAL that’s

310
00:12:20,639 –> 00:12:26,839
well maybe I mean the thing is there’s

311
00:12:22,440 –> 00:12:28,920
such a um that who you hear from is

312
00:12:26,839 –> 00:12:30,880
different it is certainly true that I do

313
00:12:28,920 –> 00:12:32,800
not um bring in funding the way an

314
00:12:30,880 –> 00:12:35,199
academic is supposed to and I was

315
00:12:32,800 –> 00:12:37,079
fortunate to find a school that didn’t

316
00:12:35,199 –> 00:12:38,519
care actually bath had sort of decided

317
00:12:37,079 –> 00:12:41,160
they didn’t care anymore either it

318
00:12:38,519 –> 00:12:43,079
became evident that I was you know again

319
00:12:41,160 –> 00:12:47,000
I was like the only AI person there for

320
00:12:43,079 –> 00:12:48,800
for like I don’t know 15 years and um

321
00:12:47,000 –> 00:12:50,480
and when they started to realize how

322
00:12:48,800 –> 00:12:52,760
important I was because other people

323
00:12:50,480 –> 00:12:54,360
started making my field important then I

324
00:12:52,760 –> 00:12:57,399
mean I’d always been well connected in

325
00:12:54,360 –> 00:12:59,399
the field um I got you know but but the

326
00:12:57,399 –> 00:13:03,000
um sorry I digressed too much might want

327
00:12:59,399 –> 00:13:04,959
to cut no no it’s fine it’s anyway so so

328
00:13:03,000 –> 00:13:06,480
yeah I’d I’d always been reasonably big

329
00:13:04,959 –> 00:13:08,279
in the field but bath didn’t have anyone

330
00:13:06,480 –> 00:13:10,000
there to recognized that and then when

331
00:13:08,279 –> 00:13:12,240
they started oh when I was on news night

332
00:13:10,000 –> 00:13:13,920
that was the main thing so anyway they

333
00:13:12,240 –> 00:13:15,399
they once they started realizing that

334
00:13:13,920 –> 00:13:17,240
the field mattered and stuff then they

335
00:13:15,399 –> 00:13:19,399
were pretty tolerant of the fact that I

336
00:13:17,240 –> 00:13:21,160
wasn’t bringing in the money myself and

337
00:13:19,399 –> 00:13:22,839
both bath and heris sko have managed to

338
00:13:21,160 –> 00:13:24,399
bring in a lot of money doing exactly

339
00:13:22,839 –> 00:13:26,720
what I do but it wasn’t me that brought

340
00:13:24,399 –> 00:13:28,279
it in and I keep trying to write grants

341
00:13:26,720 –> 00:13:29,920
actually more about the Behavioral

342
00:13:28,279 –> 00:13:31,920
Science and less about about the AI and

343
00:13:29,920 –> 00:13:34,000
I’m not as successful but but I’m happy

344
00:13:31,920 –> 00:13:36,680
to help my institutions right now since

345
00:13:34,000 –> 00:13:39,920
you are based in in Berlin do you feel

346
00:13:36,680 –> 00:13:43,199
there is more um attention and more

347
00:13:39,920 –> 00:13:45,440
resources put into um learning uh

348
00:13:43,199 –> 00:13:50,800
academic learning uh around

349
00:13:45,440 –> 00:13:52,360
AI um well not not uh originally no the

350
00:13:50,800 –> 00:13:55,240
the thing part of the reason I’m here is

351
00:13:52,360 –> 00:13:56,959
because of brexit because uh you know I

352
00:13:55,240 –> 00:13:58,759
I had not originally thought about you

353
00:13:56,959 –> 00:14:01,040
know governing AI it was not one of my

354
00:13:58,759 –> 00:14:04,320
conc concerns but I got sucked into it

355
00:14:01,040 –> 00:14:05,880
by people who who uh did notice and

356
00:14:04,320 –> 00:14:07,480
worry about those things and I’m glad

357
00:14:05,880 –> 00:14:10,440
they did you know people who were policy

358
00:14:07,480 –> 00:14:12,320
aware and so the Britain was a fantastic

359
00:14:10,440 –> 00:14:15,600
place to be involved in this I was very

360
00:14:12,320 –> 00:14:18,639
much uh mentored uh you know and and and

361
00:14:15,600 –> 00:14:20,839
and included and that was great um but

362
00:14:18,639 –> 00:14:22,880
uh the whole brexit situation is just

363
00:14:20,839 –> 00:14:24,600
insane and not just brexit itself but

364
00:14:22,880 –> 00:14:27,920
also what the conservative government is

365
00:14:24,600 –> 00:14:30,399
doing to Academia um and and and indeed

366
00:14:27,920 –> 00:14:32,600
to to digital Reg digital governance

367
00:14:30,399 –> 00:14:34,680
there was immense competence in the UK

368
00:14:32,600 –> 00:14:36,120
I’m just having some people I’m debating

369
00:14:34,680 –> 00:14:37,440
with some people in a think tank right

370
00:14:36,120 –> 00:14:39,480
now because they’re saying oh nobody in

371
00:14:37,440 –> 00:14:41,480
AI knows anything and you know there’s

372
00:14:39,480 –> 00:14:43,360
no connection and like no what are you

373
00:14:41,480 –> 00:14:44,720
talking about that you know the Royal

374
00:14:43,360 –> 00:14:46,360
Society has done a very good job of

375
00:14:44,720 –> 00:14:48,560
appointing really well-connected people

376
00:14:46,360 –> 00:14:51,639
in there’s been people doing policy for

377
00:14:48,560 –> 00:14:54,360
decades you know there was you know

378
00:14:51,639 –> 00:14:56,320
there was the the lead of the of the the

379
00:14:54,360 –> 00:14:58,519
appropriate Ministry uh I think it was

380
00:14:56,320 –> 00:15:01,000
called dcms at the time you know Tom

381
00:14:58,519 –> 00:15:04,480
Robin very very you know everybody knew

382
00:15:01,000 –> 00:15:06,519
what was going on and yet um but it just

383
00:15:04,480 –> 00:15:08,399
it’s active destruction if they’re if

384
00:15:06,519 –> 00:15:11,440
they’re taking if you know sunak is

385
00:15:08,399 –> 00:15:13,199
taking advice from Elon Musk instead of

386
00:15:11,440 –> 00:15:15,000
you know these leading academics that’s

387
00:15:13,199 –> 00:15:16,920
his choice and if he’s disrupting there

388
00:15:15,000 –> 00:15:19,079
were like in fact there was maybe too

389
00:15:16,920 –> 00:15:21,040
many there was like four or five uh

390
00:15:19,079 –> 00:15:23,399
centers of excellence and that were

391
00:15:21,040 –> 00:15:25,920
working on the on the AI and digital

392
00:15:23,399 –> 00:15:27,680
governance within the UK government and

393
00:15:25,920 –> 00:15:29,040
and if they don’t have competent if they

394
00:15:27,680 –> 00:15:31,279
don’t have links to people who know

395
00:15:29,040 –> 00:15:32,959
what’s going on there then it’s because

396
00:15:31,279 –> 00:15:35,519
well honestly I think it’s corruption

397
00:15:32,959 –> 00:15:38,000
because because the the big Tech really

398
00:15:35,519 –> 00:15:40,279
does not want to be regulated even

399
00:15:38,000 –> 00:15:41,680
though the gdpr massively improved their

400
00:15:40,279 –> 00:15:43,079
business case I mean what are

401
00:15:41,680 –> 00:15:45,560
governments doing when they regulate

402
00:15:43,079 –> 00:15:49,040
they’re trying to strengthen economies

403
00:15:45,560 –> 00:15:51,600
right and the and the US uh West Coast

404
00:15:49,040 –> 00:15:53,959
uh tech companies are just assuming that

405
00:15:51,600 –> 00:15:56,399
that this is adversarial that that it’s

406
00:15:53,959 –> 00:15:59,519
that it’s uh zero sum for reasons that

407
00:15:56,399 –> 00:16:01,519
are not at all clear um but I think that

408
00:15:59,519 –> 00:16:04,399
actually the digital regulation that

409
00:16:01,519 –> 00:16:07,440
they are now subjecting themselves to in

410
00:16:04,399 –> 00:16:10,399
the EU will actually help them running

411
00:16:07,440 –> 00:16:12,720
their own businesses and and uh well

412
00:16:10,399 –> 00:16:15,160
besides expanding their access to the EU

413
00:16:12,720 –> 00:16:18,880
and growing our digital economy as did

414
00:16:15,160 –> 00:16:20,680
the G so I think um I think it’s insane

415
00:16:18,880 –> 00:16:24,800
that they fought so hard against us and

416
00:16:20,680 –> 00:16:27,759
I think that if the UK is uh backing the

417
00:16:24,800 –> 00:16:29,560
wrong horses on this it it’s not just

418
00:16:27,759 –> 00:16:31,720
you know like a tragedy from the British

419
00:16:29,560 –> 00:16:33,920
perspective it’s also a huge security

420
00:16:31,720 –> 00:16:37,160
Hazard I mean it’s that you know they’ve

421
00:16:33,920 –> 00:16:38,920
taken uh half the H literally like half

422
00:16:37,160 –> 00:16:41,399
the military out of the EU right it’s

423
00:16:38,920 –> 00:16:43,319
basically France and ger and and Britain

424
00:16:41,399 –> 00:16:44,880
were the only ones that really had the

425
00:16:43,319 –> 00:16:47,079
active you know militaries I mean

426
00:16:44,880 –> 00:16:48,720
Germany has some capacities but yeah and

427
00:16:47,079 –> 00:16:50,639
obviously Finland and every there’s a

428
00:16:48,720 –> 00:16:53,720
bunch of like you know things but the

429
00:16:50,639 –> 00:16:56,639
biggest things were Britain and uh

430
00:16:53,720 –> 00:16:58,440
France and and now and also Britain was

431
00:16:56,639 –> 00:17:00,360
really leading in a lot of areas of

432
00:16:58,440 –> 00:17:04,480
governance as well and the including the

433
00:17:00,360 –> 00:17:05,839
digital sphere so it’s it’s um yeah it’s

434
00:17:04,480 –> 00:17:08,679
just insane if people are saying they

435
00:17:05,839 –> 00:17:10,919
can’t find confidence in you came anyway

436
00:17:08,679 –> 00:17:14,000
no no no you’re completely right even

437
00:17:10,919 –> 00:17:16,480
even affects earlier stage meaning like

438
00:17:14,000 –> 00:17:20,160
erasmos for example right it was so easy

439
00:17:16,480 –> 00:17:23,799
to to to trans transfer knowledge and

440
00:17:20,160 –> 00:17:27,480
and exchange and exchange talent and uh

441
00:17:23,799 –> 00:17:29,840
right now I’m I’m with you on this and

442
00:17:27,480 –> 00:17:33,480
and even really the I don’t mean just

443
00:17:29,840 –> 00:17:35,880
that Berlin is not a s it is also it is

444
00:17:33,480 –> 00:17:37,280
also actively really interesting so I

445
00:17:35,880 –> 00:17:40,320
think it’s really interesting why they

446
00:17:37,280 –> 00:17:44,120
had so little digital competence before

447
00:17:40,320 –> 00:17:46,679
um and I I I actually have a suspicion

448
00:17:44,120 –> 00:17:49,840
this is the conjecture let’s say that

449
00:17:46,679 –> 00:17:51,559
both Russia and Germany um they felt

450
00:17:49,840 –> 00:17:53,600
more secure about having an economy

451
00:17:51,559 –> 00:17:56,000
that’s grounded in the land so that

452
00:17:53,600 –> 00:17:57,440
people just can’t leave right so Putin

453
00:17:56,000 –> 00:17:59,159
you know if anything seems to be happy

454
00:17:57,440 –> 00:18:01,840
to have all the digital people chased

455
00:17:59,159 –> 00:18:04,240
out of out of uh Russia because they

456
00:18:01,840 –> 00:18:07,039
tend to be liberal and and I think

457
00:18:04,240 –> 00:18:09,280
Germany was either you know suspicious

458
00:18:07,039 –> 00:18:11,280
of or the government had a lot of you

459
00:18:09,280 –> 00:18:12,600
know hearing of you know the digital

460
00:18:11,280 –> 00:18:14,799
because again it’s very highly portable

461
00:18:12,600 –> 00:18:16,520
it’s new and they they had such a strong

462
00:18:14,799 –> 00:18:19,360
you know car and that kind of

463
00:18:16,520 –> 00:18:23,320
manufacturing petrochemical Industries

464
00:18:19,360 –> 00:18:27,080
so I think that um but but I think

465
00:18:23,320 –> 00:18:29,280
Germany realizes the value of of the

466
00:18:27,080 –> 00:18:32,000
digital economy but more than that

467
00:18:29,280 –> 00:18:34,120
sees the incredible importance of having

468
00:18:32,000 –> 00:18:35,360
a stable Europe and to have a stable

469
00:18:34,120 –> 00:18:38,400
Europe you have to have a well-governed

470
00:18:35,360 –> 00:18:40,960
digital sphere um and so they get that

471
00:18:38,400 –> 00:18:43,440
and so um so it’s been really just

472
00:18:40,960 –> 00:18:44,960
amazing and to come here obviously it’s

473
00:18:43,440 –> 00:18:47,440
a bigger city than bath you know I have

474
00:18:44,960 –> 00:18:49,400
a little more to do uh but but but

475
00:18:47,440 –> 00:18:53,600
mostly it’s just incredible to be

476
00:18:49,400 –> 00:18:56,360
talking to um so many Ministries and you

477
00:18:53,600 –> 00:18:58,280
know so many people so many embassies

478
00:18:56,360 –> 00:19:00,240
including the US Embassy I’m originally

479
00:18:58,280 –> 00:19:03,440
American I still hold both British and

480
00:19:00,240 –> 00:19:05,600
American passports and uh but I’ve never

481
00:19:03,440 –> 00:19:07,360
been had America ask me so many

482
00:19:05,600 –> 00:19:10,039
questions and you know bring me into so

483
00:19:07,360 –> 00:19:11,880
many events so they they also are paying

484
00:19:10,039 –> 00:19:13,320
more attention because I’m in Berlin I

485
00:19:11,880 –> 00:19:14,559
don’t know if I’d been in London if I

486
00:19:13,320 –> 00:19:15,919
would have been invited to more stuff

487
00:19:14,559 –> 00:19:17,760
you know because it’s just but you know

488
00:19:15,919 –> 00:19:20,280
baath is only an hour and a half out of

489
00:19:17,760 –> 00:19:22,080
London so I do think it’s more you know

490
00:19:20,280 –> 00:19:23,679
partly AI becoming a bigger deal but

491
00:19:22,080 –> 00:19:25,400
more that like oh you know if you’re in

492
00:19:23,679 –> 00:19:27,440
the UK you’re not doing too much damage

493
00:19:25,400 –> 00:19:30,320
wait why are you in

494
00:19:27,440 –> 00:19:31,919
Germany yes and then you know France is

495
00:19:30,320 –> 00:19:33,840
trying to lead as well they’re they’re

496
00:19:31,919 –> 00:19:36,799
are quite having quite a big success

497
00:19:33,840 –> 00:19:38,240
successes I I actually really quite like

498
00:19:36,799 –> 00:19:40,000
oh it’s so interesting actually I really

499
00:19:38,240 –> 00:19:41,679
like the French policy although it’s

500
00:19:40,000 –> 00:19:43,280
quite Divergent and they’ve made some

501
00:19:41,679 –> 00:19:46,080
big mistakes too I mean everybody makes

502
00:19:43,280 –> 00:19:47,320
some mistakes um but in general it’s so

503
00:19:46,080 –> 00:19:48,520
interesting to get a better

504
00:19:47,320 –> 00:19:50,400
understanding of Germany and who it’s

505
00:19:48,520 –> 00:19:52,400
been in historically I mean we all know

506
00:19:50,400 –> 00:19:54,280
like two or three stories of the history

507
00:19:52,400 –> 00:19:55,600
but a lot of people forget that like

508
00:19:54,280 –> 00:19:58,120
actually the Germans invented the

509
00:19:55,600 –> 00:20:01,320
welfare state you know and and the fact

510
00:19:58,120 –> 00:20:03,679
that they have the um their labor unions

511
00:20:01,320 –> 00:20:05,720
at the sea Suite is a consequence of

512
00:20:03,679 –> 00:20:07,880
1918 that they were worried about

513
00:20:05,720 –> 00:20:10,480
Revolution here and apparently the

514
00:20:07,880 –> 00:20:12,080
Marshall fund is partly because both

515
00:20:10,480 –> 00:20:14,080
Russia and America were afraid of

516
00:20:12,080 –> 00:20:15,720
another Revolution here and they really

517
00:20:14,080 –> 00:20:18,480
really so they just threw money at

518
00:20:15,720 –> 00:20:20,400
trying to there were a lot of people

519
00:20:18,480 –> 00:20:23,120
that were more democratically oriented

520
00:20:20,400 –> 00:20:25,280
than than the than the fascists um and

521
00:20:23,120 –> 00:20:28,919
so it wasn’t hard to find people to to

522
00:20:25,280 –> 00:20:29,919
back up you know um but anyway no it’s

523
00:20:28,919 –> 00:20:31,520
been really interesting because the

524
00:20:29,919 –> 00:20:33,520
French really do have a different

525
00:20:31,520 –> 00:20:35,400
perspective and I I do think it’s really

526
00:20:33,520 –> 00:20:36,559
important to have that diversity and so

527
00:20:35,400 –> 00:20:38,559
this is one of the things I’ve been

528
00:20:36,559 –> 00:20:40,600
telling America um when when they talk

529
00:20:38,559 –> 00:20:42,520
to me because they they they really want

530
00:20:40,600 –> 00:20:44,440
to see that they say hey look we are the

531
00:20:42,520 –> 00:20:46,240
most AI in the world and it’s true you

532
00:20:44,440 –> 00:20:50,120
know at the end of World War II the

533
00:20:46,240 –> 00:20:52,280
world uh the the US had um more than

534
00:20:50,120 –> 00:20:53,799
half the world’s GDP and that’s no

535
00:20:52,280 –> 00:20:59,240
longer true it’s more like

536
00:20:53,799 –> 00:21:01,640
24% but if you look at um Payton’s uh

537
00:20:59,240 –> 00:21:04,760
uh and and also the market cap of the

538
00:21:01,640 –> 00:21:08,000
companies are holding those patents um

539
00:21:04,760 –> 00:21:09,480
they’re they’re at about 50% now um of

540
00:21:08,000 –> 00:21:10,880
the of the world so there’s they’re

541
00:21:09,480 –> 00:21:12,120
dominating like they used to dominate

542
00:21:10,880 –> 00:21:14,159
the whole world that’s how they dominate

543
00:21:12,120 –> 00:21:16,080
AI so I can see why they’re saying oh

544
00:21:14,159 –> 00:21:17,440
look it’s working out well for us but at

545
00:21:16,080 –> 00:21:19,559
the same time those two measures are

546
00:21:17,440 –> 00:21:21,039
also kind of biased by the fact that

547
00:21:19,559 –> 00:21:23,720
that you know there’s like a there’s a

548
00:21:21,039 –> 00:21:25,120
securist thing going there um and it’s

549
00:21:23,720 –> 00:21:26,480
not entirely working out you know look

550
00:21:25,120 –> 00:21:29,320
at what’s happening with the democracy

551
00:21:26,480 –> 00:21:31,960
and everything so so there are

552
00:21:29,320 –> 00:21:33,760
but it’s a global Trend in a way right

553
00:21:31,960 –> 00:21:35,039
you can see that and I don’t think it’d

554
00:21:33,760 –> 00:21:38,400
be a good idea to have the whole world

555
00:21:35,039 –> 00:21:42,400
just have one side of laws no of course

556
00:21:38,400 –> 00:21:45,600
extreme extremes never good idea um okay

557
00:21:42,400 –> 00:21:50,240
but let’s go back to to the subject um

558
00:21:45,600 –> 00:21:54,559
of of AI per se so um if you can yes and

559
00:21:50,240 –> 00:21:58,000
uh so I in all of your talks you are

560
00:21:54,559 –> 00:21:59,120
very vocal about the risks of um it’s

561
00:21:58,000 –> 00:22:01,480
it’s a difficult

562
00:21:59,120 –> 00:22:04,919
pronoun to pronounce from me said word

563
00:22:01,480 –> 00:22:08,080
antho anthropomorphism exactly that one

564
00:22:04,919 –> 00:22:11,360
making making AI feel or look like human

565
00:22:08,080 –> 00:22:15,760
um and so so yes attributing them

566
00:22:11,360 –> 00:22:19,320
humanlike um traits uh why do you think

567
00:22:15,760 –> 00:22:21,720
human are prone to doing that oh well

568
00:22:19,320 –> 00:22:24,000
it’s not it’s not it’s not just we’re

569
00:22:21,720 –> 00:22:25,360
prone to doing that um so I can talk

570
00:22:24,000 –> 00:22:26,919
about this but first of all I have to

571
00:22:25,360 –> 00:22:28,120
say that not all of my talks are about

572
00:22:26,919 –> 00:22:29,799
that a lot of my talks are actually

573
00:22:28,120 –> 00:22:32,919
about things like economics and public

574
00:22:29,799 –> 00:22:34,679
goods investment and cooperation and and

575
00:22:32,919 –> 00:22:36,840
Rec recently a lot about the market

576
00:22:34,679 –> 00:22:39,240
concentration and geopolitics and power

577
00:22:36,840 –> 00:22:42,320
also related to AI

578
00:22:39,240 –> 00:22:43,679
yes why so again this this matters with

579
00:22:42,320 –> 00:22:45,000
respect to what I was just talking about

580
00:22:43,679 –> 00:22:47,080
one of the big things that we worry

581
00:22:45,000 –> 00:22:48,360
about we’re a highly social species and

582
00:22:47,080 –> 00:22:49,840
so we really want to know who our

583
00:22:48,360 –> 00:22:53,039
friends are you know we want to know

584
00:22:49,840 –> 00:22:54,600
who’s in group who’s outgroup and um so

585
00:22:53,039 –> 00:22:56,279
artificial intelligence you know it’s it

586
00:22:54,600 –> 00:22:59,240
doesn’t just happen to be there it’s an

587
00:22:56,279 –> 00:23:01,159
artifact so so companies design it in

588
00:22:59,240 –> 00:23:04,039
order to make it feel like you really

589
00:23:01,159 –> 00:23:06,200
are important you know because so that

590
00:23:04,039 –> 00:23:08,760
you will feel good about Bing and and

591
00:23:06,200 –> 00:23:10,960
using their their their products right

592
00:23:08,760 –> 00:23:12,720
so it’s not like a coincidence I I

593
00:23:10,960 –> 00:23:14,640
wanted man is super smart I mean I

594
00:23:12,720 –> 00:23:16,279
already knew him I I was at a conference

595
00:23:14,640 –> 00:23:18,240
some super super smart developmental

596
00:23:16,279 –> 00:23:22,080
psychologist he’s like you know I want

597
00:23:18,240 –> 00:23:23,840
to do a study why why uh why we love AI

598
00:23:22,080 –> 00:23:25,559
because you know you you you know in

599
00:23:23,840 –> 00:23:28,559
every movie you always wind up loving

600
00:23:25,559 –> 00:23:30,440
the robot and I’m just like that you

601
00:23:28,559 –> 00:23:32,200
people spend tens of Millions on writing

602
00:23:30,440 –> 00:23:34,200
the narratives for you to they choose

603
00:23:32,200 –> 00:23:36,400
who you’re going to

604
00:23:34,200 –> 00:23:40,880
love the same with social media right

605
00:23:36,400 –> 00:23:43,240
like and and um designing the systems um

606
00:23:40,880 –> 00:23:45,799
around how psychology works and H how

607
00:23:43,240 –> 00:23:47,520
addictions work yeah I don’t think the

608
00:23:45,799 –> 00:23:50,200
the systems were deliberately designed

609
00:23:47,520 –> 00:23:51,120
around the addiction um and and

610
00:23:50,200 –> 00:23:53,279
sometimes I think some of that’s

611
00:23:51,120 –> 00:23:54,840
exaggerated too but it’s certainly true

612
00:23:53,279 –> 00:23:57,159
that the successful ones were the ones

613
00:23:54,840 –> 00:23:58,320
that you know that that to some extent

614
00:23:57,159 –> 00:24:00,400
that that they were feeding into that a

615
00:23:58,320 –> 00:24:01,679
bit so it’s there’s there’s a certain

616
00:24:00,400 –> 00:24:03,240
level where it’s designed it’s a certain

617
00:24:01,679 –> 00:24:07,600
level where there’s selection into what

618
00:24:03,240 –> 00:24:09,720
works you know and um but anyway yeah so

619
00:24:07,600 –> 00:24:11,000
I have the feeling that we we we’ve gone

620
00:24:09,720 –> 00:24:13,919
slightly off the piss we were going to

621
00:24:11,000 –> 00:24:13,919
talk about

622
00:24:14,039 –> 00:24:18,919
um um about making oh about

623
00:24:17,240 –> 00:24:21,520
anthropomorphism yeah so the so the

624
00:24:18,919 –> 00:24:23,799
point is the part of it is that that the

625
00:24:21,520 –> 00:24:26,720
companies do what what keeps us

626
00:24:23,799 –> 00:24:28,520
believing you know engaged and a lot of

627
00:24:26,720 –> 00:24:31,559
people are engaged by people not every

628
00:24:28,520 –> 00:24:34,080
one and and uh so that’s that’s part of

629
00:24:31,559 –> 00:24:35,679
it um but I think it is also there’s a

630
00:24:34,080 –> 00:24:37,440
lot of projection people don’t want to

631
00:24:35,679 –> 00:24:39,159
die and they really want to believe that

632
00:24:37,440 –> 00:24:41,279
if they upload themselves in AI that

633
00:24:39,159 –> 00:24:43,440
they’ll live forever or other people

634
00:24:41,279 –> 00:24:45,840
feel more um maternal or fraternal they

635
00:24:43,440 –> 00:24:47,880
feel like they are going to die but they

636
00:24:45,840 –> 00:24:50,240
want to have eternal powerful you know

637
00:24:47,880 –> 00:24:52,480
friends or or children or something or

638
00:24:50,240 –> 00:24:54,080
lovers the number of people that that

639
00:24:52,480 –> 00:24:56,679
you know seriously think that they’ve

640
00:24:54,080 –> 00:24:58,640
married an avatar or whatever it you

641
00:24:56,679 –> 00:25:00,600
know it’s just a little weird and and

642
00:24:58,640 –> 00:25:02,120
you know again there’s a feminist

643
00:25:00,600 –> 00:25:03,440
perspective of this that you could think

644
00:25:02,120 –> 00:25:05,520
that something you could own that you

645
00:25:03,440 –> 00:25:07,559
could turn off and on um that you could

646
00:25:05,520 –> 00:25:09,919
clone that you could alter that that’s

647
00:25:07,559 –> 00:25:11,799
your best friend you know that that’s

648
00:25:09,919 –> 00:25:13,039
that’s you know well in a way it’s it’s

649
00:25:11,799 –> 00:25:17,320
better than having a woman that you’re

650
00:25:13,039 –> 00:25:21,480
trying to do that to you know but but uh

651
00:25:17,320 –> 00:25:23,559
yes yes yeah like um the the systems

652
00:25:21,480 –> 00:25:26,919
such as you I guess you know the replica

653
00:25:23,559 –> 00:25:28,960
and what you’ve mentioned um around

654
00:25:26,919 –> 00:25:31,799
talking or

655
00:25:28,960 –> 00:25:34,600
like uploading and feeding the the the

656
00:25:31,799 –> 00:25:38,640
algorithms the your the the Persona of

657
00:25:34,600 –> 00:25:43,039
your diseased or past uh people from

658
00:25:38,640 –> 00:25:47,279
from your past it’s it I find it a bit

659
00:25:43,039 –> 00:25:50,000
scary and worrying that people prefer to

660
00:25:47,279 –> 00:25:53,159
many people prefer to get stuck in a

661
00:25:50,000 –> 00:25:55,919
loop of of their past instead of um you

662
00:25:53,159 –> 00:25:57,960
know creating new memories and I don’t

663
00:25:55,919 –> 00:26:02,200
know if it’s I don’t know if you see it

664
00:25:57,960 –> 00:26:05,520
in in a similar way but the it was one

665
00:26:02,200 –> 00:26:10,440
in one of the um Black Mirror episode

666
00:26:05,520 –> 00:26:13,520
and the whole recording of of your

667
00:26:10,440 –> 00:26:17,240
surroundings with you know rayan like uh

668
00:26:13,520 –> 00:26:20,640
Google’s where you store like you create

669
00:26:17,240 –> 00:26:23,760
those um recordings you create data of

670
00:26:20,640 –> 00:26:27,480
your um presentence and then just you

671
00:26:23,760 –> 00:26:29,880
keep relieving it like and like I said

672
00:26:27,480 –> 00:26:32,440
without uh instead of you know just

673
00:26:29,880 –> 00:26:35,679
moving on and I I don’t know I just find

674
00:26:32,440 –> 00:26:37,440
it people have always asked about uh who

675
00:26:35,679 –> 00:26:39,159
who’s caught up in the past you know who

676
00:26:37,440 –> 00:26:41,480
tries to just keep things all the same

677
00:26:39,159 –> 00:26:44,039
who who uh reads their same books over

678
00:26:41,480 –> 00:26:46,720
and over and over um you know so I don’t

679
00:26:44,039 –> 00:26:48,159
I don’t think that’s changed that much

680
00:26:46,720 –> 00:26:50,399
um I do think when you’re going through

681
00:26:48,159 –> 00:26:51,960
a grieving process there may be a role

682
00:26:50,399 –> 00:26:54,240
as long as you have a good understanding

683
00:26:51,960 –> 00:26:56,000
that the AI is actually just like it’s

684
00:26:54,240 –> 00:26:57,600
another kind of recording you know

685
00:26:56,000 –> 00:26:58,840
people don’t have a great understanding

686
00:26:57,600 –> 00:27:00,440
that of right now you know they think

687
00:26:58,840 –> 00:27:02,320
that chat GPT is like some kind of

688
00:27:00,440 –> 00:27:04,039
entity rather than just like a weird

689
00:27:02,320 –> 00:27:06,080
Library interface which it is it’s a

690
00:27:04,039 –> 00:27:08,399
weird Library interface it’s it’s sort

691
00:27:06,080 –> 00:27:11,200
of predicts uh it predicts what you

692
00:27:08,399 –> 00:27:14,399
might say next based on some some

693
00:27:11,200 –> 00:27:16,880
culture has been trained on right so the

694
00:27:14,399 –> 00:27:18,559
uh so you can do that with you know with

695
00:27:16,880 –> 00:27:20,799
with people you miss and you can maybe

696
00:27:18,559 –> 00:27:22,760
even learn stuff I mean I was part of

697
00:27:20,799 –> 00:27:25,880
one of these experiments that Anna ster

698
00:27:22,760 –> 00:27:28,720
here in Berlin also actually she ran a

699
00:27:25,880 –> 00:27:30,919
um experiment where um they took a whole

700
00:27:28,720 –> 00:27:33,279
lot of uh Dan dennett’s writings and

701
00:27:30,919 –> 00:27:35,320
they trained up a version of dennet and

702
00:27:33,279 –> 00:27:37,480
then they asked then they asked uh

703
00:27:35,320 –> 00:27:39,840
questions and they got like four answers

704
00:27:37,480 –> 00:27:41,919
from from this you know Chat dennet

705
00:27:39,840 –> 00:27:46,399
thing and then they asked dennet himself

706
00:27:41,919 –> 00:27:48,960
he was still alive and uh and and people

707
00:27:46,399 –> 00:27:51,600
were absolutely at chance they they

708
00:27:48,960 –> 00:27:54,159
couldn’t um tell which one and I think

709
00:27:51,600 –> 00:27:55,480
that’s partly because he actually I I

710
00:27:54,159 –> 00:27:57,640
thought I was in an interview with him

711
00:27:55,480 –> 00:28:00,039
with a journalist about this and it was

712
00:27:57,640 –> 00:28:01,919
really interesting he he basically uh

713
00:28:00,039 –> 00:28:04,360
was accommodating to old age and he said

714
00:28:01,919 –> 00:28:06,600
look you you become famous for the ideas

715
00:28:04,360 –> 00:28:09,039
you had as a young man and not all your

716
00:28:06,600 –> 00:28:10,760
ideas just your best ideas and so then

717
00:28:09,039 –> 00:28:13,760
everyone who’s like read you and even

718
00:28:10,760 –> 00:28:15,720
your friends who are generous remember

719
00:28:13,760 –> 00:28:17,320
the best ideas you had and if they talk

720
00:28:15,720 –> 00:28:20,120
to the real you they’re gonna hear the

721
00:28:17,320 –> 00:28:21,720
random selection of bad ideas you have

722
00:28:20,120 –> 00:28:23,279
and that’s that was basically that he

723
00:28:21,720 –> 00:28:25,519
was flattered that people thought he was

724
00:28:23,279 –> 00:28:27,440
smarter than he really was yeah you know

725
00:28:25,519 –> 00:28:29,240
so he was as smart as his but Greatest

726
00:28:27,440 –> 00:28:32,600
Hits rather than as far as his normal

727
00:28:29,240 –> 00:28:37,960
self exactly yeah but it’s I guess it’s

728
00:28:32,600 –> 00:28:40,679
it’s with any any sort of living or um

729
00:28:37,960 –> 00:28:42,519
past uh artists right like you they have

730
00:28:40,679 –> 00:28:45,279
to create so much and only the

731
00:28:42,519 –> 00:28:47,720
percentage of their work is is it

732
00:28:45,279 –> 00:28:49,159
becomes famous and well I I was trying

733
00:28:47,720 –> 00:28:51,279
to

734
00:28:49,159 –> 00:28:53,799
um I would I would want you know I was

735
00:28:51,279 –> 00:28:55,559
one of the experimental subjects and and

736
00:28:53,799 –> 00:28:57,360
the main thing I was looking at was I’m

737
00:28:55,559 –> 00:28:58,600
going I don’t think he still believes

738
00:28:57,360 –> 00:28:59,840
that that must be something on his

739
00:28:58,600 –> 00:29:02,399
little text you know there’ be things

740
00:28:59,840 –> 00:29:03,679
where just the time didn’t make sense

741
00:29:02,399 –> 00:29:04,919
because I because i’ had been involved

742
00:29:03,679 –> 00:29:06,919
with some of the projects and I knew

743
00:29:04,919 –> 00:29:09,640
that he was you know but I still I was

744
00:29:06,919 –> 00:29:09,640
still at chance

745
00:29:10,279 –> 00:29:20,200
so but about that and few years ago you

746
00:29:15,559 –> 00:29:22,559
um wrote this uh article on for wir one

747
00:29:20,200 –> 00:29:27,320
day AI will seem as human as everyone

748
00:29:22,559 –> 00:29:30,039
anyone H was then it was 2020 I think um

749
00:29:27,320 –> 00:29:31,600
no no it was uh I think it was 22

750
00:29:30,039 –> 00:29:33,360
because it was the year it was right

751
00:29:31,600 –> 00:29:35,919
after the lamba thing happened so it

752
00:29:33,360 –> 00:29:38,519
time goes fast but it’s only two years

753
00:29:35,919 –> 00:29:41,799
it’s a it’s a feels like a long a long

754
00:29:38,519 –> 00:29:44,240
time to you know judging on what how how

755
00:29:41,799 –> 00:29:47,480
much of a progress uh we can see at at

756
00:29:44,240 –> 00:29:50,440
least in a commercial uh space in AI do

757
00:29:47,480 –> 00:29:53,760
you feel or like have you Revisited any

758
00:29:50,440 –> 00:29:56,640
of the beliefs you included at that time

759
00:29:53,760 –> 00:29:59,679
um around about you know the progress um

760
00:29:56,640 –> 00:30:02,120
or the way AI is is going to be

761
00:29:59,679 –> 00:30:05,919
constructed and used and you know where

762
00:30:02,120 –> 00:30:07,679
is it going or I I honestly haven’t R it

763
00:30:05,919 –> 00:30:09,360
uh in the last couple years so I can’t

764
00:30:07,679 –> 00:30:10,799
but I can’t think of anything that was

765
00:30:09,360 –> 00:30:12,480
wrong I haven’t seen something thought

766
00:30:10,799 –> 00:30:14,080
out that I was wrong in fact more often

767
00:30:12,480 –> 00:30:16,760
I think this is exactly what I said was

768
00:30:14,080 –> 00:30:19,240
going to happen and you know what what

769
00:30:16,760 –> 00:30:22,159
drives I mean I particularly was uh

770
00:30:19,240 –> 00:30:24,880
worried about the fact that people

771
00:30:22,159 –> 00:30:27,000
thought that um oh yeah I may have a

772
00:30:24,880 –> 00:30:28,159
regret I’ll come back to that okay so I

773
00:30:27,000 –> 00:30:29,320
was particularly worried at the time

774
00:30:28,159 –> 00:30:31,120
that people are saying oh don’t worry

775
00:30:29,320 –> 00:30:34,720
because humans will be more creative and

776
00:30:31,120 –> 00:30:36,279
whatever um and you know I and I think

777
00:30:34,720 –> 00:30:37,840
they were really undermining themselves

778
00:30:36,279 –> 00:30:39,480
they had kind of a religious faith that

779
00:30:37,840 –> 00:30:41,320
humans were more creative and yet any

780
00:30:39,480 –> 00:30:42,960
[ __ ] could look at the output of this

781
00:30:41,320 –> 00:30:44,919
stuff and see that was kind of creative

782
00:30:42,960 –> 00:30:46,480
right and maybe it’s not your favorite

783
00:30:44,919 –> 00:30:47,720
stuff for some people it is and you

784
00:30:46,480 –> 00:30:49,240
don’t want to undermine them and say oh

785
00:30:47,720 –> 00:30:50,919
you have no taste you think that the B

786
00:30:49,240 –> 00:30:55,519
is better right you know I just told you

787
00:30:50,919 –> 00:30:57,639
that story about D it so so uh so I you

788
00:30:55,519 –> 00:30:59,120
know so I wanted to say look there’s

789
00:30:57,639 –> 00:31:00,240
something more fundamental and this is

790
00:30:59,120 –> 00:31:02,000
something I’ve been working on for a

791
00:31:00,240 –> 00:31:04,240
long time so I kind of had something to

792
00:31:02,000 –> 00:31:06,000
write about it but I just did not like

793
00:31:04,240 –> 00:31:08,159
the way that the convers the public

794
00:31:06,000 –> 00:31:11,159
conversation was going so I I released

795
00:31:08,159 –> 00:31:12,919
basically some of my book um to into

796
00:31:11,159 –> 00:31:14,960
wired and I say released I send it to

797
00:31:12,919 –> 00:31:17,720
them they’re like nah I was like what no

798
00:31:14,960 –> 00:31:19,440
this is important you know and so there

799
00:31:17,720 –> 00:31:20,600
it got heavily edited and I think it’s

800
00:31:19,440 –> 00:31:22,279
pretty much okay there’s like one

801
00:31:20,600 –> 00:31:24,480
section I think isn’t coherent oh yeah

802
00:31:22,279 –> 00:31:28,519
the one thing that I I thought may have

803
00:31:24,480 –> 00:31:30,639
been wrong is that the I was taking the

804
00:31:28,519 –> 00:31:32,880
limit case that even in the case where

805
00:31:30,639 –> 00:31:35,679
you cannot tell it apart at all in any

806
00:31:32,880 –> 00:31:37,080
context you know touring touring test is

807
00:31:35,679 –> 00:31:39,760
completely

808
00:31:37,080 –> 00:31:42,039
past what does that mean to humans right

809
00:31:39,760 –> 00:31:43,559
what and how is our how and basically I

810
00:31:42,039 –> 00:31:44,919
predicted that that our life most

811
00:31:43,559 –> 00:31:48,519
people’s lives are not going to change

812
00:31:44,919 –> 00:31:50,919
that much um and because everything that

813
00:31:48,519 –> 00:31:53,720
we value is based on the basically the

814
00:31:50,919 –> 00:31:56,600
social problems of Apes right and and

815
00:31:53,720 –> 00:31:58,360
even like if you look the at the graphs

816
00:31:56,600 –> 00:31:59,880
that come out like they people say oh

817
00:31:58,360 –> 00:32:01,080
look at how much faster machine learning

818
00:31:59,880 –> 00:32:03,279
is going it’s like looks like it’s

819
00:32:01,080 –> 00:32:05,120
really zooming up but the point is that

820
00:32:03,279 –> 00:32:08,039
once it gets a little bit above human

821
00:32:05,120 –> 00:32:09,760
level it just plateaus why because any

822
00:32:08,039 –> 00:32:13,399
task we care about is something that we

823
00:32:09,760 –> 00:32:15,240
do right we it’s like you know nobody

824
00:32:13,399 –> 00:32:17,240
cares about the tasks that at which we

825
00:32:15,240 –> 00:32:18,799
were already superhuman like you know

826
00:32:17,240 –> 00:32:20,559
like this building can contain more

827
00:32:18,799 –> 00:32:22,320
people than any person can contain right

828
00:32:20,559 –> 00:32:23,880
or or the you know calculator can do

829
00:32:22,320 –> 00:32:26,200
arithmetic fast than any person could do

830
00:32:23,880 –> 00:32:28,039
it and everybody stopped caring so all

831
00:32:26,200 –> 00:32:29,880
the stuff we think is intelligence

832
00:32:28,039 –> 00:32:31,279
actually sort of humanlike and then

833
00:32:29,880 –> 00:32:34,039
there’s no way to be more human-like

834
00:32:31,279 –> 00:32:36,000
than a human so you you basically run

835
00:32:34,039 –> 00:32:40,279
out of you you run out of stuff you can

836
00:32:36,000 –> 00:32:42,720
do you know we have uh you know all all

837
00:32:40,279 –> 00:32:47,440
all all chess programs can play better

838
00:32:42,720 –> 00:32:49,200
than 9 99 99 you know n NS percent of

839
00:32:47,440 –> 00:32:52,919
people but people still play chess with

840
00:32:49,200 –> 00:32:55,240
each other and you know and everybody um

841
00:32:52,919 –> 00:32:56,240
you know people still have pup quizzes

842
00:32:55,240 –> 00:32:57,840
they don’t they don’t sit there and

843
00:32:56,240 –> 00:32:59,440
Google onto the table like why would

844
00:32:57,840 –> 00:33:00,799
would they do that you know and you know

845
00:32:59,440 –> 00:33:02,639
maybe you know some teenagers did it

846
00:33:00,799 –> 00:33:05,000
once in a while or something but at the

847
00:33:02,639 –> 00:33:06,799
end of the day all these things are

848
00:33:05,000 –> 00:33:09,279
human relationships and jobs are human

849
00:33:06,799 –> 00:33:11,320
relationships and you know the idea that

850
00:33:09,279 –> 00:33:13,960
you could just replace everyone I mean

851
00:33:11,320 –> 00:33:15,799
there you can replace too many people in

852
00:33:13,960 –> 00:33:17,880
you know so if you aren’t thinking

853
00:33:15,799 –> 00:33:19,880
correctly about this then you can make

854
00:33:17,880 –> 00:33:21,880
big mistakes and and this is I think the

855
00:33:19,880 –> 00:33:23,240
biggest issue in AI ethics right now I

856
00:33:21,880 –> 00:33:24,519
mean a lot of people are disagreeing

857
00:33:23,240 –> 00:33:27,960
about this but I just said something

858
00:33:24,519 –> 00:33:30,120
else um on on social media but what of

859
00:33:27,960 –> 00:33:32,279
the biggest things that happened was uh

860
00:33:30,120 –> 00:33:35,960
the in October 7th of

861
00:33:32,279 –> 00:33:39,039
2023 that there there was um there

862
00:33:35,960 –> 00:33:42,600
weren’t enough humans watching the

863
00:33:39,039 –> 00:33:44,279
perimeter of the Gaza Israel divide and

864
00:33:42,600 –> 00:33:47,960
because they thought that having a few

865
00:33:44,279 –> 00:33:50,159
cameras was enough and I’m not sure who

866
00:33:47,960 –> 00:33:52,799
thought they having a few cameras was

867
00:33:50,159 –> 00:33:54,360
enough and I don’t even know you know

868
00:33:52,799 –> 00:33:56,320
what what might have happened around

869
00:33:54,360 –> 00:33:58,480
those cameras but I had identified and I

870
00:33:56,320 –> 00:34:01,159
managed to find the the video

871
00:33:58,480 –> 00:34:03,480
of me saying in 2018 to Eric Schmidt

872
00:34:01,159 –> 00:34:06,039
this is a bad idea you do not want to

873
00:34:03,480 –> 00:34:08,040
have no people and a few cameras because

874
00:34:06,039 –> 00:34:09,839
first of all you know that’s like one

875
00:34:08,040 –> 00:34:11,639
point of failure one algorithm that you

876
00:34:09,839 –> 00:34:13,839
could work around and and you can break

877
00:34:11,639 –> 00:34:16,359
security and so don’t you know don’t

878
00:34:13,839 –> 00:34:19,240
don’t give people such a simple Target

879
00:34:16,359 –> 00:34:21,879
but secondly um the corruption potential

880
00:34:19,240 –> 00:34:24,320
for corruption once you do that then if

881
00:34:21,879 –> 00:34:26,000
somebody does want something to happen

882
00:34:24,320 –> 00:34:28,320
then it’s relatively easy for them to

883
00:34:26,000 –> 00:34:30,599
Mastermind that you want blowers you

884
00:34:28,320 –> 00:34:32,200
want soldiers you you you know it’s not

885
00:34:30,599 –> 00:34:33,520
that we’re going to eliminate my lies

886
00:34:32,200 –> 00:34:36,359
it’s that we are going to eliminate

887
00:34:33,520 –> 00:34:37,560
people stopping and Reporting my lies

888
00:34:36,359 –> 00:34:39,520
and that was the other thing I said

889
00:34:37,560 –> 00:34:43,280
earlier was that what we’re doing to

890
00:34:39,520 –> 00:34:45,800
journalism um uh by by just taking all

891
00:34:43,280 –> 00:34:47,079
the money away from newspapers right I

892
00:34:45,800 –> 00:34:49,720
don’t know again this is like I don’t

893
00:34:47,079 –> 00:34:50,839
think that was the original plan like to

894
00:34:49,720 –> 00:34:52,480
eliminate journalism I think the

895
00:34:50,839 –> 00:34:55,320
original plan was like gez how do we pay

896
00:34:52,480 –> 00:34:56,639
for our magic search machine right but

897
00:34:55,320 –> 00:34:58,160
but basically all that money now is

898
00:34:56,639 –> 00:35:00,000
going to Google and Facebook

899
00:34:58,160 –> 00:35:03,000
and and it’s very very hard to keep the

900
00:35:00,000 –> 00:35:05,800
journalist employed and um and now

901
00:35:03,000 –> 00:35:07,359
people are stealing their their text and

902
00:35:05,800 –> 00:35:09,440
and making fake people and pretending

903
00:35:07,359 –> 00:35:11,280
that they wrote the text and this is the

904
00:35:09,440 –> 00:35:12,560
kind of problems I think it’s solvable I

905
00:35:11,280 –> 00:35:14,839
don’t think it’s like oh my God it’s the

906
00:35:12,560 –> 00:35:17,040
ends of the world but we have to realize

907
00:35:14,839 –> 00:35:18,400
it’s important and choose to solve it

908
00:35:17,040 –> 00:35:21,079
right you know just like you know

909
00:35:18,400 –> 00:35:23,839
smoking causing cancer and and nuclear

910
00:35:21,079 –> 00:35:26,000
bi biological weapons you know we this

911
00:35:23,839 –> 00:35:28,079
is something that requires investment

912
00:35:26,000 –> 00:35:29,760
from the government um and from

913
00:35:28,079 –> 00:35:31,520
governments which means investment from

914
00:35:29,760 –> 00:35:34,560
taxpayers we all have to say this is

915
00:35:31,520 –> 00:35:39,280
something that matters like pay the

916
00:35:34,560 –> 00:35:41,640
those um payment um pay walls solve this

917
00:35:39,280 –> 00:35:44,000
issue only very in a very very small

918
00:35:41,640 –> 00:35:48,520
percentage usually people they got used

919
00:35:44,000 –> 00:35:51,240
to not paying uh and and being okay with

920
00:35:48,520 –> 00:35:54,240
being advertised um even though most of

921
00:35:51,240 –> 00:35:58,760
the things were you know

922
00:35:54,240 –> 00:36:01,200
um misleading to to to put it lightly um

923
00:35:58,760 –> 00:36:05,040
so what do you see as a potential

924
00:36:01,200 –> 00:36:08,640
solution you know for for journalism to

925
00:36:05,040 –> 00:36:10,280
survive right now I well either I I

926
00:36:08,640 –> 00:36:13,400
actually think journalism is a lot like

927
00:36:10,280 –> 00:36:15,560
higher education um they’re both and you

928
00:36:13,400 –> 00:36:16,920
know uh autocratic governments find them

929
00:36:15,560 –> 00:36:19,440
threatening because they’re basically

930
00:36:16,920 –> 00:36:21,200
information seeking and disseminating um

931
00:36:19,440 –> 00:36:23,119
and they’re they’re absolutely essential

932
00:36:21,200 –> 00:36:26,640
to a democracy in fact not just

933
00:36:23,119 –> 00:36:28,319
democracy if the the cultural revolution

934
00:36:26,640 –> 00:36:30,280
was stopped by China because they

935
00:36:28,319 –> 00:36:32,640
realize they’re falling behind Russia

936
00:36:30,280 –> 00:36:34,839
technologically you know so it it isn’t

937
00:36:32,640 –> 00:36:36,920
only you know so that you can vote it’s

938
00:36:34,839 –> 00:36:38,839
also so you can like you know have a

939
00:36:36,920 –> 00:36:43,119
working uh economy and all kinds of

940
00:36:38,839 –> 00:36:46,640
things you know but um so I think that

941
00:36:43,119 –> 00:36:48,400
uh one solution would be to um think a

942
00:36:46,640 –> 00:36:50,720
little harder about how to fund both and

943
00:36:48,400 –> 00:36:53,040
to bring them together um and maybe

944
00:36:50,720 –> 00:36:55,560
house them together um but there I do

945
00:36:53,040 –> 00:36:56,880
think that there’s going it doesn’t make

946
00:36:55,560 –> 00:36:58,319
sense so if you think the British went

947
00:36:56,880 –> 00:37:00,240
through this situation while I was there

948
00:36:58,319 –> 00:37:01,520
where they they sort of decided that

949
00:37:00,240 –> 00:37:02,760
maybe half of people should go to

950
00:37:01,520 –> 00:37:04,480
undergraduate now I’m not sure that

951
00:37:02,760 –> 00:37:05,839
number is true some people said that you

952
00:37:04,480 –> 00:37:07,480
know it got it was through a

953
00:37:05,839 –> 00:37:09,319
miscommunication with Tony Blair or

954
00:37:07,480 –> 00:37:10,599
something but once the government had

955
00:37:09,319 –> 00:37:13,760
announced the target they couldn’t go

956
00:37:10,599 –> 00:37:15,359
back so they they uh but anyway so so

957
00:37:13,760 –> 00:37:17,640
let’s just take the simple case half of

958
00:37:15,359 –> 00:37:20,119
people are getting higher education who

959
00:37:17,640 –> 00:37:22,319
pays for it so historically in Britain

960
00:37:20,119 –> 00:37:25,400
the taxpayers have paid for 5% of people

961
00:37:22,319 –> 00:37:28,599
to go to university um it doesn’t really

962
00:37:25,400 –> 00:37:31,640
make sense for everyone

963
00:37:28,599 –> 00:37:33,680
to uh pay the same amount and half the

964
00:37:31,640 –> 00:37:34,880
people to get more benefit out of it now

965
00:37:33,680 –> 00:37:37,160
if you can say that doing an

966
00:37:34,880 –> 00:37:39,040
undergraduate degree is work and it is

967
00:37:37,160 –> 00:37:41,480
you know you’re taking uh three or four

968
00:37:39,040 –> 00:37:44,160
years out of out of when you could have

969
00:37:41,480 –> 00:37:47,400
been getting paid you know it’s it’s a

970
00:37:44,160 –> 00:37:49,359
bunch of effort um that you know that

971
00:37:47,400 –> 00:37:51,920
then then you could say well you know

972
00:37:49,359 –> 00:37:54,839
maybe it equals equalizes out but in

973
00:37:51,920 –> 00:37:56,440
general um the reason people do it all

974
00:37:54,839 –> 00:37:57,839
that effort is because they expect over

975
00:37:56,440 –> 00:38:00,319
the period of their lifetime to get a

976
00:37:57,839 –> 00:38:01,960
higher return and so I think you have to

977
00:38:00,319 –> 00:38:03,880
do some kind of balance between like

978
00:38:01,960 –> 00:38:06,400
this is a public good it does help us

979
00:38:03,880 –> 00:38:07,960
all to have these educated people um

980
00:38:06,400 –> 00:38:09,599
again to some level and and we should

981
00:38:07,960 –> 00:38:12,680
come up with a better metric of exactly

982
00:38:09,599 –> 00:38:16,800
how many people need what education

983
00:38:12,680 –> 00:38:18,400
um but then you also need um then you

984
00:38:16,800 –> 00:38:21,240
also need to figure out how much do they

985
00:38:18,400 –> 00:38:23,000
contribute so I think actually the the

986
00:38:21,240 –> 00:38:24,560
solution that liberal Democrats came up

987
00:38:23,000 –> 00:38:26,920
with when they were sort of semi in

988
00:38:24,560 –> 00:38:29,680
power what was sensible that they said

989
00:38:26,920 –> 00:38:32,280
oh well you you you pay some proportion

990
00:38:29,680 –> 00:38:34,160
of your salary over some threshold for

991
00:38:32,280 –> 00:38:36,400
some number of years like I think it was

992
00:38:34,160 –> 00:38:38,119
30 or something and then it’s all so if

993
00:38:36,400 –> 00:38:40,480
somebody does decide to go into you know

994
00:38:38,119 –> 00:38:42,200
the Arts or or volunteerism or something

995
00:38:40,480 –> 00:38:44,640
then they never pay back and it’s not so

996
00:38:42,200 –> 00:38:46,920
it’s not like a regular debt um so you

997
00:38:44,640 –> 00:38:50,280
only get the extra benefit that that you

998
00:38:46,920 –> 00:38:51,920
got uh from if you basically only you

999
00:38:50,280 –> 00:38:53,960
only you have to make more you had to

1000
00:38:51,920 –> 00:38:56,319
make more than average to have your

1001
00:38:53,960 –> 00:38:57,599
overall tuition costs be higher than

1002
00:38:56,319 –> 00:38:59,720
they would have been before the this

1003
00:38:57,599 –> 00:39:01,079
whole thing started and and if you’re

1004
00:38:59,720 –> 00:39:03,440
making more than average well then

1005
00:39:01,079 –> 00:39:04,920
presumably your education’s helping you

1006
00:39:03,440 –> 00:39:07,000
um I mean which is of course a big

1007
00:39:04,920 –> 00:39:08,400
presumption but it’s some kind of it’s

1008
00:39:07,000 –> 00:39:09,720
none of this stuff is ever going to be

1009
00:39:08,400 –> 00:39:11,839
perfect it’s all about choosing

1010
00:39:09,720 –> 00:39:14,359
something that makes incentives right so

1011
00:39:11,839 –> 00:39:16,520
Society can improve itself uh as much as

1012
00:39:14,359 –> 00:39:17,960
possible that’s that’s the point so

1013
00:39:16,520 –> 00:39:20,119
anyway I thought that was pretty good

1014
00:39:17,960 –> 00:39:21,760
but it was pretty complicated even the

1015
00:39:20,119 –> 00:39:24,160
liberal Democrats didn’t try to explain

1016
00:39:21,760 –> 00:39:27,880
at the next election of course they got

1017
00:39:24,160 –> 00:39:30,520
slau I remember when I so I studied I

1018
00:39:27,880 –> 00:39:32,720
started my like you know as I’m in I’m

1019
00:39:30,520 –> 00:39:35,800
polish I I started University in Poland

1020
00:39:32,720 –> 00:39:38,280
and I moved to to UK to London to study

1021
00:39:35,800 –> 00:39:42,880
to continue studies and I remember you

1022
00:39:38,280 –> 00:39:47,319
know Poland is um has a lower income per

1023
00:39:42,880 –> 00:39:51,359
capita um as than UK and I remember it

1024
00:39:47,319 –> 00:39:53,400
was so much easier for for us to to get

1025
00:39:51,359 –> 00:39:57,760
uh I think it was grants but also some

1026
00:39:53,400 –> 00:40:00,200
some kind of credits uh to study

1027
00:39:57,760 –> 00:40:03,880
and since brexit I think that that was

1028
00:40:00,200 –> 00:40:05,720
taken away as well wow yeah no I think

1029
00:40:03,880 –> 00:40:07,839
uh yeah no I do think the British did a

1030
00:40:05,720 –> 00:40:08,880
lot of things to try to make it work um

1031
00:40:07,839 –> 00:40:10,160
they but they were just trying to figure

1032
00:40:08,880 –> 00:40:12,200
out how to fund and they had a pretty

1033
00:40:10,160 –> 00:40:14,720
good University system and I think they

1034
00:40:12,200 –> 00:40:16,760
still have one um by Legacy but but they

1035
00:40:14,720 –> 00:40:18,800
are losing you they’re showing a lot of

1036
00:40:16,760 –> 00:40:20,560
departments and things it’s it’s it’s

1037
00:40:18,800 –> 00:40:23,800
very

1038
00:40:20,560 –> 00:40:27,480
sad yes I I I agree with that you

1039
00:40:23,800 –> 00:40:30,760
mentioned the transparency um issue and

1040
00:40:27,480 –> 00:40:35,880
the accountability uh of those of

1041
00:40:30,760 –> 00:40:40,800
Designing uh systems which can bend

1042
00:40:35,880 –> 00:40:42,839
um the public uh opinion and and

1043
00:40:40,800 –> 00:40:46,400
eventually how they vote and how how

1044
00:40:42,839 –> 00:40:49,160
they um make choices important choices

1045
00:40:46,400 –> 00:40:52,480
so how do you think or who do you think

1046
00:40:49,160 –> 00:40:54,960
should be respon held responsible for

1047
00:40:52,480 –> 00:40:58,440
for Designing those systems for misuse

1048
00:40:54,960 –> 00:41:03,000
and um

1049
00:40:58,440 –> 00:41:05,440
like in creating those those models um

1050
00:41:03,000 –> 00:41:08,280
that they are not so biased as they are

1051
00:41:05,440 –> 00:41:10,800
now okay all right let’s leave bias out

1052
00:41:08,280 –> 00:41:12,599
of this for a second and go just back to

1053
00:41:10,800 –> 00:41:16,400
the accountability question so the so

1054
00:41:12,599 –> 00:41:18,400
the the basic point is that AI is not no

1055
00:41:16,400 –> 00:41:19,880
different from any other Advanced

1056
00:41:18,400 –> 00:41:23,520
complicated technology you have to deal

1057
00:41:19,880 –> 00:41:26,040
with so just like a car or or you know

1058
00:41:23,520 –> 00:41:27,800
just like a car that um if there’s

1059
00:41:26,040 –> 00:41:29,839
something that the design

1060
00:41:27,800 –> 00:41:31,240
or developer has done wrong so the

1061
00:41:29,839 –> 00:41:32,560
brakes are failing or something then

1062
00:41:31,240 –> 00:41:34,599
they have then they’re liable and they

1063
00:41:32,560 –> 00:41:37,040
have to do a recall if it’s the person

1064
00:41:34,599 –> 00:41:38,920
driving it that got drunk or whatever or

1065
00:41:37,040 –> 00:41:40,200
or just was not bothering to pay

1066
00:41:38,920 –> 00:41:42,599
attention or something playing with

1067
00:41:40,200 –> 00:41:44,280
their phone then it’s their fault and so

1068
00:41:42,599 –> 00:41:46,720
those are the only two kinds of people

1069
00:41:44,280 –> 00:41:48,119
that can be at fault basically the other

1070
00:41:46,720 –> 00:41:49,680
thing with any digital system whether

1071
00:41:48,119 –> 00:41:52,800
it’s AI or not of course is that you

1072
00:41:49,680 –> 00:41:55,359
could also have hackers um but then

1073
00:41:52,800 –> 00:41:57,480
again if you have hackers then that’s a

1074
00:41:55,359 –> 00:41:59,440
problem either of the Developers who

1075
00:41:57,480 –> 00:42:01,040
didn’t provide you enough cyber security

1076
00:41:59,440 –> 00:42:05,359
or the users that didn’t keep their

1077
00:42:01,040 –> 00:42:07,400
cyber security upgraded updated so um so

1078
00:42:05,359 –> 00:42:10,079
either way basically there’s only the

1079
00:42:07,400 –> 00:42:12,400
the and the the owners operators and the

1080
00:42:10,079 –> 00:42:16,000
people that that sold them the system

1081
00:42:12,400 –> 00:42:18,359
now because AI has very you so funny AI

1082
00:42:16,000 –> 00:42:20,359
people are like oh but you know but we

1083
00:42:18,359 –> 00:42:22,119
have all these software libraries and

1084
00:42:20,359 –> 00:42:23,880
you know data and everything it’s so

1085
00:42:22,119 –> 00:42:25,559
complicated you know it’s exactly the

1086
00:42:23,880 –> 00:42:26,880
same for building a fighter jet or a

1087
00:42:25,559 –> 00:42:29,079
hospital or something you know there’s

1088
00:42:26,880 –> 00:42:30,839
you have to Source all the products you

1089
00:42:29,079 –> 00:42:32,720
have to worry about your your supply

1090
00:42:30,839 –> 00:42:34,960
chain you have to you have to vet all

1091
00:42:32,720 –> 00:42:37,319
those people it’s normal it’s just a

1092
00:42:34,960 –> 00:42:41,040
software is becoming normal only well I

1093
00:42:37,319 –> 00:42:43,640
mean it’s normal for for um for

1094
00:42:41,040 –> 00:42:45,319
dangerous stuff you know like building a

1095
00:42:43,640 –> 00:42:47,319
big structure that can fall on people

1096
00:42:45,319 –> 00:42:49,720
then you have to get licensed you know

1097
00:42:47,319 –> 00:42:52,240
it it used to be that you you that you

1098
00:42:49,720 –> 00:42:53,800
had to get licenses to be an architect

1099
00:42:52,240 –> 00:42:56,200
or to build your own house or whatever

1100
00:42:53,800 –> 00:42:57,880
but now you have to all the Architects

1101
00:42:56,200 –> 00:42:59,480
get licens at school

1102
00:42:57,880 –> 00:43:00,960
and then um the building has to get

1103
00:42:59,480 –> 00:43:03,200
planning permission and the building has

1104
00:43:00,960 –> 00:43:04,880
to get inspected you know you don’t just

1105
00:43:03,200 –> 00:43:07,280
build it used to be any rich person

1106
00:43:04,880 –> 00:43:09,440
could build things anywhere but then as

1107
00:43:07,280 –> 00:43:12,200
ordinary people’s lives literally became

1108
00:43:09,440 –> 00:43:14,359
more valuable they were able to sue if

1109
00:43:12,200 –> 00:43:16,400
if a bunch of them got crushed right and

1110
00:43:14,359 –> 00:43:17,920
you see still to this day around

1111
00:43:16,400 –> 00:43:19,280
different parts of the world that

1112
00:43:17,920 –> 00:43:21,680
different people worry about this to

1113
00:43:19,280 –> 00:43:23,640
different extents you know it was really

1114
00:43:21,680 –> 00:43:25,839
tragic that in Turkey you know that you

1115
00:43:23,640 –> 00:43:28,200
saw that only The Institute of

1116
00:43:25,839 –> 00:43:30,920
engineering had followed had complied

1117
00:43:28,200 –> 00:43:32,400
with the with the laws the regulations

1118
00:43:30,920 –> 00:43:33,960
about how to build and their building

1119
00:43:32,400 –> 00:43:36,040
was perfectly still standing when

1120
00:43:33,960 –> 00:43:39,720
everything around it was collapsed you

1121
00:43:36,040 –> 00:43:41,280
know it was awful um but but it showed

1122
00:43:39,720 –> 00:43:43,200
that that it was corruption and not

1123
00:43:41,280 –> 00:43:45,280
following the law and not enforcing the

1124
00:43:43,200 –> 00:43:48,559
law that had led to all those hundreds

1125
00:43:45,280 –> 00:43:50,280
of thousands of deaths right so uh I

1126
00:43:48,559 –> 00:43:51,640
it’s the same thing and and we have to

1127
00:43:50,280 –> 00:43:53,319
deal with it the fact that we make

1128
00:43:51,640 –> 00:43:54,839
systems that they’re so essential to

1129
00:43:53,319 –> 00:43:56,720
people’s lives now we’re not just

1130
00:43:54,839 –> 00:43:58,280
building toys anymore we’re not even

1131
00:43:56,720 –> 00:43:59,760
just building spreadsheets and stuff

1132
00:43:58,280 –> 00:44:02,359
that helps with an individual business

1133
00:43:59,760 –> 00:44:04,880
now we build things the entire complexes

1134
00:44:02,359 –> 00:44:07,680
depend on and so we have to expect that

1135
00:44:04,880 –> 00:44:11,079
we have to do a better job so basically

1136
00:44:07,680 –> 00:44:12,720
um you you are responsible if you bought

1137
00:44:11,079 –> 00:44:15,720
it and you’ve used it wrong you’re

1138
00:44:12,720 –> 00:44:17,559
responsible um and you know even that if

1139
00:44:15,720 –> 00:44:20,000
if if it was too easy to use wrong you

1140
00:44:17,559 –> 00:44:21,359
might consider the company responsible

1141
00:44:20,000 –> 00:44:23,160
if the company finds out that it was

1142
00:44:21,359 –> 00:44:25,119
somewhere in the further back like so

1143
00:44:23,160 –> 00:44:27,319
there was some company that you know

1144
00:44:25,119 –> 00:44:29,079
sourced something from somewhere else

1145
00:44:27,319 –> 00:44:30,800
that’s their problem and so they they

1146
00:44:29,079 –> 00:44:32,160
can worry that about that with respect

1147
00:44:30,800 –> 00:44:34,160
to the contracts that they have with

1148
00:44:32,160 –> 00:44:35,960
their suppliers and vendors but you as

1149
00:44:34,160 –> 00:44:37,359
the user don’t need to worry about that

1150
00:44:35,960 –> 00:44:38,839
just like if you went into a bank you

1151
00:44:37,359 –> 00:44:40,319
don’t care which employee was the one

1152
00:44:38,839 –> 00:44:41,880
who stole your money you just need your

1153
00:44:40,319 –> 00:44:43,240
money back and and that’s what the

1154
00:44:41,880 –> 00:44:45,000
government is going to force the bank to

1155
00:44:43,240 –> 00:44:48,880
do for you hopefully you’ve got that

1156
00:44:45,000 –> 00:44:50,400
right lawyers and things um so uh so

1157
00:44:48,880 –> 00:44:53,240
yeah that’s the situation that we are

1158
00:44:50,400 –> 00:44:55,319
now in that’s what the AI act says and

1159
00:44:53,240 –> 00:44:57,119
it was just such a big deal I mean I

1160
00:44:55,319 –> 00:44:59,800
just can’t believe how much trouble

1161
00:44:57,119 –> 00:45:02,800
there was around such a simple idea you

1162
00:44:59,800 –> 00:45:04,400
know that basically um software is is

1163
00:45:02,800 –> 00:45:06,280
actually a product it’s not not a

1164
00:45:04,400 –> 00:45:08,680
service I mean it’s sort of it can

1165
00:45:06,280 –> 00:45:10,720
provide a service but the Ser the

1166
00:45:08,680 –> 00:45:12,720
software itself is a product and you you

1167
00:45:10,720 –> 00:45:15,160
have all the obligations including being

1168
00:45:12,720 –> 00:45:17,920
able to prove that it was your user who

1169
00:45:15,160 –> 00:45:20,559
misused it so that is on the developer

1170
00:45:17,920 –> 00:45:22,119
if they can’t show that if if they can’t

1171
00:45:20,559 –> 00:45:24,480
show that the user is the one who messed

1172
00:45:22,119 –> 00:45:26,160
up then it’s on them they’re the ones

1173
00:45:24,480 –> 00:45:30,240
reliable to society if there’s damage

1174
00:45:26,160 –> 00:45:32,720
cost mhm is there any aspect you don’t

1175
00:45:30,240 –> 00:45:36,599
agree or you think it’s missing from the

1176
00:45:32,720 –> 00:45:38,400
ACT um I think to be honest I read it a

1177
00:45:36,599 –> 00:45:39,800
bunch of times before the final one and

1178
00:45:38,400 –> 00:45:42,720
now that it’s over I kind of haven’t

1179
00:45:39,800 –> 00:45:44,520
read the final version um but I

1180
00:45:42,720 –> 00:45:46,440
understand that there’s that looked like

1181
00:45:44,520 –> 00:45:48,800
a lot of stuff got shoehorned into it

1182
00:45:46,440 –> 00:45:52,160
towards the end and that there’s two

1183
00:45:48,800 –> 00:45:54,760
wide of holes open to interpretation and

1184
00:45:52,160 –> 00:45:56,359
interpretation means massive lobbying

1185
00:45:54,760 –> 00:45:58,720
from people with more money than they

1186
00:45:56,359 –> 00:46:00,880
should be allowed have you know the that

1187
00:45:58,720 –> 00:46:02,680
it’s literally that that the US has

1188
00:46:00,880 –> 00:46:04,599
stopped enforcing antitrust is a big

1189
00:46:02,680 –> 00:46:06,920
part of the problem inequality is

1190
00:46:04,599 –> 00:46:10,359
spiraling and then you have you have

1191
00:46:06,920 –> 00:46:12,839
these just insane processes so but in by

1192
00:46:10,359 –> 00:46:16,599
by and large no I think it’s great um I

1193
00:46:12,839 –> 00:46:19,119
guess the entire thing about um about uh

1194
00:46:16,599 –> 00:46:20,720
generative AI that or Foundation models

1195
00:46:19,119 –> 00:46:22,880
however you want to call it I don’t

1196
00:46:20,720 –> 00:46:25,240
think we needed it at all I think that

1197
00:46:22,880 –> 00:46:27,280
the framework we had was adequately

1198
00:46:25,240 –> 00:46:28,880
totally adequate but unfortunately Ely a

1199
00:46:27,280 –> 00:46:30,800
whole lot of legislators were were

1200
00:46:28,880 –> 00:46:32,960
fooled into believing that oh this is

1201
00:46:30,800 –> 00:46:34,319
beyond what we thought of and and the

1202
00:46:32,960 –> 00:46:35,880
only good thing that came out of that

1203
00:46:34,319 –> 00:46:38,359
was that it helped them think more about

1204
00:46:35,880 –> 00:46:41,240
IP and the IP behind the data so I think

1205
00:46:38,359 –> 00:46:43,520
there was a little bit of of uh stuff

1206
00:46:41,240 –> 00:46:46,480
there but anyway as far as I understand

1207
00:46:43,520 –> 00:46:49,280
the final law version of the law um what

1208
00:46:46,480 –> 00:46:51,559
was going on was that uh that a couple

1209
00:46:49,280 –> 00:46:53,839
of big tech companies had put you know a

1210
00:46:51,559 –> 00:46:56,960
couple like four or five have put

1211
00:46:53,839 –> 00:46:58,200
billions I mean billions and billions I

1212
00:46:56,960 –> 00:47:00,960
car San thing but this year we’re

1213
00:46:58,200 –> 00:47:04,160
talking about dollars not Stars sadly um

1214
00:47:00,960 –> 00:47:05,839
so billions and billions ofll same into

1215
00:47:04,160 –> 00:47:08,000
these products that it’s not clear to

1216
00:47:05,839 –> 00:47:09,839
have any real use I mean they they do

1217
00:47:08,000 –> 00:47:12,480
make some people more productive in some

1218
00:47:09,839 –> 00:47:15,000
contexts but it’s not clear whether they

1219
00:47:12,480 –> 00:47:17,400
make them enough more productive to pay

1220
00:47:15,000 –> 00:47:21,040
for the costs the overhead the Energy

1221
00:47:17,400 –> 00:47:22,280
overhead um and so while those people

1222
00:47:21,040 –> 00:47:23,319
are trying to sort that out the same

1223
00:47:22,280 –> 00:47:24,839
thing was true with Google at the

1224
00:47:23,319 –> 00:47:26,040
beginning incidentally this is going

1225
00:47:24,839 –> 00:47:28,359
back to when they finally figured out

1226
00:47:26,040 –> 00:47:29,800
the advertising model right but so at

1227
00:47:28,359 –> 00:47:31,160
the beginning Google just said forget

1228
00:47:29,800 –> 00:47:32,559
that we’re just we’re just creating a

1229
00:47:31,160 –> 00:47:35,720
great product and we’ll worry about you

1230
00:47:32,559 –> 00:47:37,440
know how to pay for it later um but what

1231
00:47:35,720 –> 00:47:39,079
they’re worried about is that in fact

1232
00:47:37,440 –> 00:47:40,720
you don’t need that much data this is

1233
00:47:39,079 –> 00:47:42,520
like when I was talking about Chess

1234
00:47:40,720 –> 00:47:44,359
right every chess program even once

1235
00:47:42,520 –> 00:47:46,160
people like hack together in a garage

1236
00:47:44,359 –> 00:47:48,400
and sell to kids on like little plastic

1237
00:47:46,160 –> 00:47:50,520
toys it’s playing better than humans

1238
00:47:48,400 –> 00:47:52,640
right you don’t need to be you know deep

1239
00:47:50,520 –> 00:47:55,119
blue anymore to you know you probably

1240
00:47:52,640 –> 00:47:57,160
still need to be that to be CASRO but

1241
00:47:55,119 –> 00:48:01,280
you don’t you don’t need that much power

1242
00:47:57,160 –> 00:48:03,720
to to beat 99% of people and so so that

1243
00:48:01,280 –> 00:48:05,319
was what uh there there was right before

1244
00:48:03,720 –> 00:48:07,440
all this stuff happened there was this

1245
00:48:05,319 –> 00:48:09,119
release a leak from Google saying we

1246
00:48:07,440 –> 00:48:11,520
don’t have a moat and neither does open

1247
00:48:09,119 –> 00:48:13,359
AI we’re putting on this investment in

1248
00:48:11,520 –> 00:48:15,599
and and you can find these cheap

1249
00:48:13,359 –> 00:48:18,000
products that um you don’t have to pay

1250
00:48:15,599 –> 00:48:20,079
much for that that do basically the same

1251
00:48:18,000 –> 00:48:21,559
thing and why you know why is anyone

1252
00:48:20,079 –> 00:48:23,599
going to pay for the amount the amount

1253
00:48:21,559 –> 00:48:25,319
of money that we’re we’re spending and

1254
00:48:23,599 –> 00:48:27,680
so basically what they really really

1255
00:48:25,319 –> 00:48:29,400
wanted was that you know ban anybody

1256
00:48:27,680 –> 00:48:31,960
else working on that but them make it

1257
00:48:29,400 –> 00:48:34,400
just prohibitively expensive and instead

1258
00:48:31,960 –> 00:48:36,160
what the AI act actually does it says

1259
00:48:34,400 –> 00:48:37,839
this only regulates people over a

1260
00:48:36,160 –> 00:48:39,440
certain size which as far as I

1261
00:48:37,839 –> 00:48:41,720
understand there’s only two models big

1262
00:48:39,440 –> 00:48:43,680
enough to be so the only people that

1263
00:48:41,720 –> 00:48:45,200
have extra stuff in the AI act are

1264
00:48:43,680 –> 00:48:48,000
exactly the people that caused all this

1265
00:48:45,200 –> 00:48:50,319
trouble and cost the EU so much effort

1266
00:48:48,000 –> 00:48:52,440
um and time and and and loss of sleep

1267
00:48:50,319 –> 00:48:54,079
and everything else so I think that’s

1268
00:48:52,440 –> 00:48:58,040
kind if it’s true that’s kind of wicked

1269
00:48:54,079 –> 00:49:01,680
I love it it would other it would

1270
00:48:58,040 –> 00:49:03,160
be like catastrophic to to small

1271
00:49:01,680 –> 00:49:06,240
startups and and companies who are

1272
00:49:03,160 –> 00:49:08,799
trying to you know um

1273
00:49:06,240 –> 00:49:10,760
challenge a big word but but yeah know

1274
00:49:08,799 –> 00:49:12,359
that the the but I mean that was the

1275
00:49:10,760 –> 00:49:14,160
thing Sam Altman was going out to the

1276
00:49:12,359 –> 00:49:16,160
Congress saying oh please please I’m a

1277
00:49:14,160 –> 00:49:17,480
good guy you know bind my hands and

1278
00:49:16,160 –> 00:49:19,920
anyone else that looks like me because

1279
00:49:17,480 –> 00:49:21,559
hey I’m leading and then but don’t but

1280
00:49:19,920 –> 00:49:23,640
don’t do any other kind of AI because

1281
00:49:21,559 –> 00:49:26,319
that would hurt the and it’s like Oh you

1282
00:49:23,640 –> 00:49:28,960
mean the SNES like say Microsoft who is

1283
00:49:26,319 –> 00:49:32,480
funding actually paying your billions

1284
00:49:28,960 –> 00:49:34,480
your 14 billion electricity bills and

1285
00:49:32,480 –> 00:49:36,400
and uh you know basically all the stuff

1286
00:49:34,480 –> 00:49:39,280
he said that other kind of AI is like

1287
00:49:36,400 –> 00:49:43,119
you know search is AI right that like

1288
00:49:39,280 –> 00:49:44,599
that was like that’s like AI 101 in 1993

1289
00:49:43,119 –> 00:49:46,960
that you know AI is basically just

1290
00:49:44,599 –> 00:49:48,559
search right search is ai ai is search

1291
00:49:46,960 –> 00:49:51,440
you know so you’re always trying to find

1292
00:49:48,559 –> 00:49:53,559
the next thing to do right so it’s it’s

1293
00:49:51,440 –> 00:49:54,839
integral like all the all these software

1294
00:49:53,559 –> 00:49:56,359
products that these companies are

1295
00:49:54,839 –> 00:49:59,200
producing they don’t want those to be

1296
00:49:56,359 –> 00:50:00,839
right ated which is nuts you know it

1297
00:49:59,200 –> 00:50:03,200
it’s just it’s just nuts because like I

1298
00:50:00,839 –> 00:50:07,040
said it’s not that honorous um they just

1299
00:50:03,200 –> 00:50:09,240
need to you know to grow up and and deal

1300
00:50:07,040 –> 00:50:11,760
with you know work together as a sector

1301
00:50:09,240 –> 00:50:13,280
to protect the people on the planet so

1302
00:50:11,760 –> 00:50:15,240
that they can still have a company in

1303
00:50:13,280 –> 00:50:16,640
another decade right it doesn’t make

1304
00:50:15,240 –> 00:50:20,160
sense that they’re that they’re that

1305
00:50:16,640 –> 00:50:22,160
they’re intervening so much um but I do

1306
00:50:20,160 –> 00:50:24,520
think I do think some of them sincerely

1307
00:50:22,160 –> 00:50:25,680
believe that I mean but but what they’re

1308
00:50:24,520 –> 00:50:27,720
sincerely believing is just

1309
00:50:25,680 –> 00:50:30,240
undereducated under informed and they

1310
00:50:27,720 –> 00:50:31,960
they really ought to get you know bring

1311
00:50:30,240 –> 00:50:33,640
in people that actually understand you

1312
00:50:31,960 –> 00:50:35,720
know political science that actually

1313
00:50:33,640 –> 00:50:37,839
understand uh you know various other

1314
00:50:35,720 –> 00:50:41,839
things you know that product product

1315
00:50:37,839 –> 00:50:43,839
liability and uh you yeah they do but

1316
00:50:41,839 –> 00:50:47,520
you know judging from what happened

1317
00:50:43,839 –> 00:50:51,520
recently with with the safety board

1318
00:50:47,520 –> 00:50:53,960
safety um Board of uh open AI you know

1319
00:50:51,520 –> 00:50:56,079
removing them and creating new

1320
00:50:53,960 –> 00:50:57,319
one that there so there’s so many

1321
00:50:56,079 –> 00:51:00,359
problems like a said this is there’s

1322
00:50:57,319 –> 00:51:01,640
just so much money around and um there’s

1323
00:51:00,359 –> 00:51:04,200
some real questions about what’s going

1324
00:51:01,640 –> 00:51:07,680
on with open AI I mean on the one hand

1325
00:51:04,200 –> 00:51:11,040
you know safety that is like long

1326
00:51:07,680 –> 00:51:12,640
termism is not really safety um and so

1327
00:51:11,040 –> 00:51:14,240
but on the other hand yeah there was

1328
00:51:12,640 –> 00:51:16,040
this whole hope that they actually did

1329
00:51:14,240 –> 00:51:18,839
have some kind of board independency you

1330
00:51:16,040 –> 00:51:20,319
want that and so again I it’s it’s just

1331
00:51:18,839 –> 00:51:23,400
like a big soap oper and I don’t have

1332
00:51:20,319 –> 00:51:24,960
time for so poers you know you know if

1333
00:51:23,400 –> 00:51:26,640
somebody wants to pay me I’ll go study

1334
00:51:24,960 –> 00:51:28,000
it but right now I’m more interested in

1335
00:51:26,640 –> 00:51:29,359
some other problems I’m I’m mostly

1336
00:51:28,000 –> 00:51:31,319
working on political polarization

1337
00:51:29,359 –> 00:51:35,760
actually right yeah yeah yeah yeah I can

1338
00:51:31,319 –> 00:51:39,799
see okay and the bias aspect which you

1339
00:51:35,760 –> 00:51:42,680
oh right okay so yeah so so too much

1340
00:51:39,799 –> 00:51:44,240
bias so this is the whole thing it’s not

1341
00:51:42,680 –> 00:51:46,079
only the AI act there’s also something

1342
00:51:44,240 –> 00:51:47,480
called the Digital Services Act and the

1343
00:51:46,079 –> 00:51:49,799
digital markets act but we won’t talk

1344
00:51:47,480 –> 00:51:51,240
about that right now but these are three

1345
00:51:49,799 –> 00:51:52,960
really important pieces of of

1346
00:51:51,240 –> 00:51:54,040
legislation and the general data

1347
00:51:52,960 –> 00:51:55,880
protection regulation which is

1348
00:51:54,040 –> 00:51:57,880
underlying at all that’s where the human

1349
00:51:55,880 –> 00:51:59,799
rights comes in incidentally all the

1350
00:51:57,880 –> 00:52:02,240
human rights is like fundamental to EU

1351
00:51:59,799 –> 00:52:05,119
law and gdpr is basically the one that

1352
00:52:02,240 –> 00:52:06,640
protects our human rights right okay but

1353
00:52:05,119 –> 00:52:09,119
then one up from that the Digital

1354
00:52:06,640 –> 00:52:11,520
Services Act and the AI act both say we

1355
00:52:09,119 –> 00:52:13,440
need to know whether you built your

1356
00:52:11,520 –> 00:52:16,079
software well right what are you doing

1357
00:52:13,440 –> 00:52:19,559
with it we need enough transparency so

1358
00:52:16,079 –> 00:52:21,040
the thing is about bias is that you know

1359
00:52:19,559 –> 00:52:23,520
we used to have to teach people again

1360
00:52:21,040 –> 00:52:26,280
back to AI 101 or machine learning 101

1361
00:52:23,520 –> 00:52:28,480
in this case bias is just regularities

1362
00:52:26,280 –> 00:52:29,960
right you expect a model to find out

1363
00:52:28,480 –> 00:52:32,559
that there’s certain things that happen

1364
00:52:29,960 –> 00:52:35,440
all the time so if you read the paper I

1365
00:52:32,559 –> 00:52:37,520
wrote on on on bias and in 2017 that

1366
00:52:35,440 –> 00:52:40,079
showed you know it was the big paper

1367
00:52:37,520 –> 00:52:42,200
that showed that um that the implicit

1368
00:52:40,079 –> 00:52:43,720
biases that people have are also

1369
00:52:42,200 –> 00:52:47,240
reflected in the software that’s trained

1370
00:52:43,720 –> 00:52:49,119
on the outs of those people right but it

1371
00:52:47,240 –> 00:52:51,240
also at the same time showed that those

1372
00:52:49,119 –> 00:52:53,559
implicit biases were reflecting lived

1373
00:52:51,240 –> 00:52:55,559
reality so like if you know you have

1374
00:52:53,559 –> 00:52:58,359
this sexist belief that you know nurses

1375
00:52:55,559 –> 00:53:00,119
are female and and programmers are male

1376
00:52:58,359 –> 00:53:01,799
but then if you looked at the US La

1377
00:53:00,119 –> 00:53:03,760
Labor Statistics that’s what was

1378
00:53:01,799 –> 00:53:05,359
happening and so there and it wasn’t

1379
00:53:03,760 –> 00:53:08,319
just like you know in general it was

1380
00:53:05,359 –> 00:53:11,559
like it was actually a pretty uh a very

1381
00:53:08,319 –> 00:53:13,400
very strong correlation um so in a way

1382
00:53:11,559 –> 00:53:15,240
our implicit biases and we could we

1383
00:53:13,400 –> 00:53:16,720
didn’t know this before like you know in

1384
00:53:15,240 –> 00:53:19,799
fact there was big arguments about

1385
00:53:16,720 –> 00:53:21,480
whether um that you know that these that

1386
00:53:19,799 –> 00:53:23,160
that you know the the conservatives were

1387
00:53:21,480 –> 00:53:25,920
saying you guys are trying to change

1388
00:53:23,160 –> 00:53:28,440
reality and the and the um and the

1389
00:53:25,920 –> 00:53:30,280
Liberals were saying oh no you know like

1390
00:53:28,440 –> 00:53:32,280
the the world is actually Fair it was

1391
00:53:30,280 –> 00:53:34,160
created to be fair and and you guys are

1392
00:53:32,280 –> 00:53:35,520
the ones who are just messing it up well

1393
00:53:34,160 –> 00:53:38,079
it turns out the conservatives were

1394
00:53:35,520 –> 00:53:40,520
right you know the world isn’t fair and

1395
00:53:38,079 –> 00:53:43,160
but I I still think that’s okay that

1396
00:53:40,520 –> 00:53:45,640
that the that that we choose to make the

1397
00:53:43,160 –> 00:53:48,160
world more fair than it was before right

1398
00:53:45,640 –> 00:53:50,079
I’m I’m I’m totally behind that but but

1399
00:53:48,160 –> 00:53:53,400
the but the path is different between

1400
00:53:50,079 –> 00:53:56,480
those two so anyway my point is

1401
00:53:53,400 –> 00:53:57,760
that we can’t use uh no amount of

1402
00:53:56,480 –> 00:54:00,000
transparency is going to make the world

1403
00:53:57,760 –> 00:54:02,559
more fair than it is right that that

1404
00:54:00,000 –> 00:54:05,599
takes policy decisions and real effort

1405
00:54:02,559 –> 00:54:07,240
you know and and and some things there

1406
00:54:05,599 –> 00:54:09,960
there is no Chang in the fact that you

1407
00:54:07,240 –> 00:54:11,960
know the women when they’re pregnant

1408
00:54:09,960 –> 00:54:14,079
that there’s nine months you know like

1409
00:54:11,960 –> 00:54:16,040
that’s women and it’s not meant right

1410
00:54:14,079 –> 00:54:18,160
that maybe some biologist will change

1411
00:54:16,040 –> 00:54:20,079
that but basically that’s that’s that’s

1412
00:54:18,160 –> 00:54:22,240
the way so some things are fundamentally

1413
00:54:20,079 –> 00:54:24,480
unfair and it’s not that’s not the only

1414
00:54:22,240 –> 00:54:27,319
thing there’s also um like there’s more

1415
00:54:24,480 –> 00:54:29,280
genetic risks taken with you know like

1416
00:54:27,319 –> 00:54:30,680
like this is not this is not a human

1417
00:54:29,280 –> 00:54:32,240
decision this is true like in you know

1418
00:54:30,680 –> 00:54:34,000
fruit flies and things too right you

1419
00:54:32,240 –> 00:54:36,200
know it’s just that the mixing the

1420
00:54:34,000 –> 00:54:37,760
mixing of the genomes level you know so

1421
00:54:36,200 –> 00:54:39,760
so there’s certain things that are just

1422
00:54:37,760 –> 00:54:41,400
not the same at least I don’t know how

1423
00:54:39,760 –> 00:54:43,920
you want to Define but they’re certainly

1424
00:54:41,400 –> 00:54:45,440
not the same but we have chosen what

1425
00:54:43,920 –> 00:54:47,720
what we try to do with our system of

1426
00:54:45,440 –> 00:54:49,200
justice is say that shouldn’t when

1427
00:54:47,720 –> 00:54:51,480
you’re born that every child should have

1428
00:54:49,200 –> 00:54:54,040
at least a certain amount of of of

1429
00:54:51,480 –> 00:54:55,559
chance of doing well no matter who their

1430
00:54:54,040 –> 00:54:57,359
parents were no matter what genes they

1431
00:54:55,559 –> 00:54:59,040
happen to have and again there’s some

1432
00:54:57,359 –> 00:55:00,839
things we can’t overcome you know like

1433
00:54:59,040 –> 00:55:02,160
if you’re if you’re born with no brain

1434
00:55:00,839 –> 00:55:04,400
that happens right and there’s nothing

1435
00:55:02,160 –> 00:55:06,880
we can do about that the best of our

1436
00:55:04,400 –> 00:55:08,839
abilities we try to help everyone uh be

1437
00:55:06,880 –> 00:55:11,680
a productive member of society and have

1438
00:55:08,839 –> 00:55:13,520
POS possibilities and beliefs right so

1439
00:55:11,680 –> 00:55:16,640
um that’s not about you know that’s not

1440
00:55:13,520 –> 00:55:19,119
something you can fix by by better data

1441
00:55:16,640 –> 00:55:21,480
right but you but but nevertheless

1442
00:55:19,119 –> 00:55:24,319
having said all that absolutely it is

1443
00:55:21,480 –> 00:55:27,160
possible to build an unfair AI system

1444
00:55:24,319 –> 00:55:28,799
you can deliberately set it up to hurt

1445
00:55:27,160 –> 00:55:30,039
people that are of A different race or a

1446
00:55:28,799 –> 00:55:32,359
different ethnicity or a different

1447
00:55:30,039 –> 00:55:35,000
religion or whatever or gender or

1448
00:55:32,359 –> 00:55:37,319
whatever you know you you can easily go

1449
00:55:35,000 –> 00:55:38,880
into the system and say it’s you’re

1450
00:55:37,319 –> 00:55:40,720
going to be less likely to get into

1451
00:55:38,880 –> 00:55:43,839
college in this context or right

1452
00:55:40,720 –> 00:55:45,400
whatever you’re not g to get Alone um

1453
00:55:43,839 –> 00:55:48,039
you know in China right now you’re not

1454
00:55:45,400 –> 00:55:49,960
going to be able to fly if you if you

1455
00:55:48,039 –> 00:55:52,280
used to study wegers even if you let

1456
00:55:49,960 –> 00:55:53,440
alone if you are Wier right you know and

1457
00:55:52,280 –> 00:55:55,400
they’re going to say it’s a social

1458
00:55:53,440 –> 00:55:56,920
credit system but it’s actually you know

1459
00:55:55,400 –> 00:55:59,039
it’s not about like bad things you’ve

1460
00:55:56,920 –> 00:56:01,319
done in terms of like smoking on the

1461
00:55:59,039 –> 00:56:03,640
train it’s bad things that you’ve done

1462
00:56:01,319 –> 00:56:06,079
as in pursuing former policy which has

1463
00:56:03,640 –> 00:56:08,440
now been changed and updated right and

1464
00:56:06,079 –> 00:56:10,559
so that is why we need transparency we

1465
00:56:08,440 –> 00:56:13,359
need to be able to see whether there’s

1466
00:56:10,559 –> 00:56:15,280
been extra bias put in and the other

1467
00:56:13,359 –> 00:56:17,039
thing is that when people do figure out

1468
00:56:15,280 –> 00:56:18,559
ways to make the system more fair and

1469
00:56:17,039 –> 00:56:21,200
you’ll never make it perfectly fair but

1470
00:56:18,559 –> 00:56:23,480
we have found ways to make it more fair

1471
00:56:21,200 –> 00:56:24,960
then that becomes best practice and

1472
00:56:23,480 –> 00:56:26,720
particularly large companies have an

1473
00:56:24,960 –> 00:56:28,039
obligation to follow the best practice

1474
00:56:26,720 –> 00:56:30,119
they have to show they’ve done due

1475
00:56:28,039 –> 00:56:32,200
diligence and so that’s what we try to

1476
00:56:30,119 –> 00:56:33,799
do and and so that’s why that’s sitting

1477
00:56:32,200 –> 00:56:35,920
on top of what I was talking about

1478
00:56:33,799 –> 00:56:38,119
before bias is just one thing though

1479
00:56:35,920 –> 00:56:40,319
there’s so many ways you can hurt people

1480
00:56:38,119 –> 00:56:42,039
and you can and if you want to you can

1481
00:56:40,319 –> 00:56:43,839
pretend that it was you know the AI

1482
00:56:42,039 –> 00:56:46,839
signed off on it or something and it’s

1483
00:56:43,839 –> 00:56:49,000
just rubbish you know and and it’s a way

1484
00:56:46,839 –> 00:56:50,799
to get out of trying to get out of

1485
00:56:49,000 –> 00:56:53,880
responsibility but we’re we’re hopefully

1486
00:56:50,799 –> 00:56:55,960
educating enough judges that that that

1487
00:56:53,880 –> 00:56:57,079
um decision doesn’t wash that excuse

1488
00:56:55,960 –> 00:57:01,160
does not not

1489
00:56:57,079 –> 00:57:04,960
was how can we incentivize and support

1490
00:57:01,160 –> 00:57:06,240
the AI for good um efforts well I mean

1491
00:57:04,960 –> 00:57:08,400
there’s people that put tons of money

1492
00:57:06,240 –> 00:57:09,559
into AI for good and again some of those

1493
00:57:08,400 –> 00:57:11,280
people are the people who don’t want to

1494
00:57:09,559 –> 00:57:13,799
be regulated so I’m a little worried

1495
00:57:11,280 –> 00:57:15,960
about such things I I think that what we

1496
00:57:13,799 –> 00:57:19,680
have done with the European regulation

1497
00:57:15,960 –> 00:57:22,440
is fantastic and it’s just making m to

1498
00:57:19,680 –> 00:57:24,960
the best to the best of our abilities

1499
00:57:22,440 –> 00:57:27,520
making it a fair market where where

1500
00:57:24,960 –> 00:57:28,960
people can pursue uh complaints you know

1501
00:57:27,520 –> 00:57:30,960
where we can check and see whether or

1502
00:57:28,960 –> 00:57:32,240
not people were doing the right thing

1503
00:57:30,960 –> 00:57:34,839
that’s going to solve a lot of the

1504
00:57:32,240 –> 00:57:36,280
problems there are other things that are

1505
00:57:34,839 –> 00:57:37,839
that that didn’t I said it wasn’t going

1506
00:57:36,280 –> 00:57:40,440
to mention digital Market side but there

1507
00:57:37,839 –> 00:57:42,480
are other problems which are that some

1508
00:57:40,440 –> 00:57:44,520
things are just by Nature very large and

1509
00:57:42,480 –> 00:57:45,960
then the market can’t help take care of

1510
00:57:44,520 –> 00:57:48,319
it and so that means even more

1511
00:57:45,960 –> 00:57:50,160
governance but again what does that mean

1512
00:57:48,319 –> 00:57:51,640
it’s like so like electricity bills and

1513
00:57:50,160 –> 00:57:53,720
things like that you get citizens to

1514
00:57:51,640 –> 00:57:55,640
work together and they decide what would

1515
00:57:53,720 –> 00:57:56,920
be fair or what you know is practicable

1516
00:57:55,640 –> 00:57:58,400
and things and then they work with the

1517
00:57:56,920 –> 00:57:59,920
government and they enforce that on the

1518
00:57:58,400 –> 00:58:02,680
people who provide electricity or

1519
00:57:59,920 –> 00:58:07,280
provide trains or whatever provide post

1520
00:58:02,680 –> 00:58:10,640
offices um so so I think there will be

1521
00:58:07,280 –> 00:58:13,400
some it’s not AI itself but you know AI

1522
00:58:10,640 –> 00:58:14,960
is just a suite of software uh uh

1523
00:58:13,400 –> 00:58:16,520
techniques which can be applied to all

1524
00:58:14,960 –> 00:58:18,480
kinds of things but stuff like search

1525
00:58:16,520 –> 00:58:21,000
stuff like Social Media stuff like

1526
00:58:18,480 –> 00:58:23,200
generative AI that might very well turn

1527
00:58:21,000 –> 00:58:25,680
out to be more like electricity and

1528
00:58:23,200 –> 00:58:27,160
plumbing and less like you know I don’t

1529
00:58:25,680 –> 00:58:32,720
know car

1530
00:58:27,160 –> 00:58:34,920
so so in that case then um then then we

1531
00:58:32,720 –> 00:58:37,520
will need to have even more regulation

1532
00:58:34,920 –> 00:58:39,200
but still the basic principles of real

1533
00:58:37,520 –> 00:58:40,839
transparency to How It’s Made and making

1534
00:58:39,200 –> 00:58:42,200
sure people are following good practice

1535
00:58:40,839 –> 00:58:45,400
and they’re doing things for the right

1536
00:58:42,200 –> 00:58:47,640
reasons uh that that’s the key how do

1537
00:58:45,400 –> 00:58:51,319
you see this like where do you see it

1538
00:58:47,640 –> 00:58:54,359
going in terms of human and AI um

1539
00:58:51,319 –> 00:58:56,520
collaboration in terms of you know maybe

1540
00:58:54,359 –> 00:58:59,000
arriving to

1541
00:58:56,520 –> 00:59:00,720
AGI so first of all my most recent

1542
00:58:59,000 –> 00:59:03,119
philosophy paper in fact the only one

1543
00:59:00,720 –> 00:59:05,000
I’ve written with actual philosophers is

1544
00:59:03,119 –> 00:59:07,079
called do we collaborate with what we

1545
00:59:05,000 –> 00:59:09,280
design and the answer is no that’s the

1546
00:59:07,079 –> 00:59:10,559
wrong metaphor the only way that we can

1547
00:59:09,280 –> 00:59:12,680
do the kinds of improvement we’re

1548
00:59:10,559 –> 00:59:14,880
talking about is to hold the people

1549
00:59:12,680 –> 00:59:17,240
responsible for the systems they develop

1550
00:59:14,880 –> 00:59:18,799
and they own and that they operate they

1551
00:59:17,240 –> 00:59:21,119
are the ones we have to hold to account

1552
00:59:18,799 –> 00:59:23,400
there is no way to hold a design system

1553
00:59:21,119 –> 00:59:26,480
to account it doesn’t make sense so it

1554
00:59:23,400 –> 00:59:29,000
is not helpful to think about um about

1555
00:59:26,480 –> 00:59:30,119
AI itself as Pier or whatever it’s it’s

1556
00:59:29,000 –> 00:59:32,799
it’s like what we were talking about

1557
00:59:30,119 –> 00:59:35,319
before with replica like you’re just you

1558
00:59:32,799 –> 00:59:36,960
you you’re pretending that something is

1559
00:59:35,319 –> 00:59:39,440
held to account but actually you’re just

1560
00:59:36,960 –> 00:59:42,200
letting the people who misused some

1561
00:59:39,440 –> 00:59:44,920
technology off the hook all right so so

1562
00:59:42,200 –> 00:59:47,359
collaborating out the window now okay

1563
00:59:44,920 –> 00:59:49,760
AGI it depends what you define how you

1564
00:59:47,359 –> 00:59:51,359
mean it um when I first heard the term

1565
00:59:49,760 –> 00:59:53,920
it was people that were saying oh you

1566
00:59:51,359 –> 00:59:55,640
know nobody working in AI is actually

1567
00:59:53,920 –> 00:59:56,839
trying to solve all the problems they

1568
00:59:55,640 –> 00:59:58,359
you know they’re all just trying to do

1569
00:59:56,839 –> 01:00:00,000
Machine Vision or something and that’s

1570
00:59:58,359 –> 01:00:01,559
boring and we want to get back to more

1571
01:00:00,000 –> 01:00:03,440
human you know really understand human

1572
01:00:01,559 –> 01:00:05,240
intelligence and you know that was

1573
01:00:03,440 –> 01:00:06,880
completely false there were very good

1574
01:00:05,240 –> 01:00:08,799
reasons a lot of people were trying to

1575
01:00:06,880 –> 01:00:10,720
solve the whole problem of intelligence

1576
01:00:08,799 –> 01:00:13,640
and and and it was it was just this

1577
01:00:10,720 –> 01:00:15,799
false narrative um and then people

1578
01:00:13,640 –> 01:00:18,119
started talking about oh AGI there’s

1579
01:00:15,799 –> 01:00:19,799
like this um there’s going to be this

1580
01:00:18,119 –> 01:00:22,119
intelligence explosion when you have

1581
01:00:19,799 –> 01:00:23,039
systems that can learn how to learn and

1582
01:00:22,119 –> 01:00:25,119
then there’s going to be this

1583
01:00:23,039 –> 01:00:29,480
exponential and exponential growth and

1584
01:00:25,119 –> 01:00:32,799
we will be m is after that well I think

1585
01:00:29,480 –> 01:00:34,000
that is at least logical um but again it

1586
01:00:32,799 –> 01:00:36,559
doesn’t make sense to talk about the

1587
01:00:34,000 –> 01:00:39,480
system itself and if you actually look

1588
01:00:36,559 –> 01:00:40,760
at the number of people on the planet

1589
01:00:39,480 –> 01:00:42,960
that’s what’s happened since we’ve had

1590
01:00:40,760 –> 01:00:45,079
writing so since we’ve been able to

1591
01:00:42,960 –> 01:00:47,680
improve our own intelligence through our

1592
01:00:45,079 –> 01:00:49,079
technology um we have been that

1593
01:00:47,680 –> 01:00:52,280
explosion and that’s why we have a

1594
01:00:49,079 –> 01:00:54,240
climate crisis right now right so so if

1595
01:00:52,280 –> 01:00:56,520
that’s what you mean by AGI then you’re

1596
01:00:54,240 –> 01:00:58,280
you know it it happened 10,000 years ago

1597
01:00:56,520 –> 01:01:00,720
and we need to start dealing with it a

1598
01:00:58,280 –> 01:01:03,119
little better than we have been um and

1599
01:01:00,720 –> 01:01:05,480
then finally uh there’s like this thing

1600
01:01:03,119 –> 01:01:06,720
about well you know AI is going to be

1601
01:01:05,480 –> 01:01:09,039
controlling everything and we’re just

1602
01:01:06,720 –> 01:01:11,640
gting these magic magic machines that

1603
01:01:09,039 –> 01:01:13,280
that take over the world and um then

1604
01:01:11,640 –> 01:01:15,200
what we do to you know ensure that

1605
01:01:13,280 –> 01:01:17,559
there’s employment or whatever and these

1606
01:01:15,200 –> 01:01:19,440
guys basically wind up you know they

1607
01:01:17,559 –> 01:01:20,799
they’ve stuck in this extra piece which

1608
01:01:19,440 –> 01:01:22,760
as I said doesn’t make sense it’s just

1609
01:01:20,799 –> 01:01:25,359
an extension an extension of human

1610
01:01:22,760 –> 01:01:26,880
agency but they’ve hypothesized this

1611
01:01:25,359 –> 01:01:28,799
extra piece and then they start trying

1612
01:01:26,880 –> 01:01:30,920
to solve as I said the basic problems of

1613
01:01:28,799 –> 01:01:32,920
political science you know and economics

1614
01:01:30,920 –> 01:01:34,520
and whatever and and they just they just

1615
01:01:32,920 –> 01:01:36,640
come back to exactly where we already

1616
01:01:34,520 –> 01:01:38,920
are and the best thing those guys can do

1617
01:01:36,640 –> 01:01:40,799
is take their enormous brains and and

1618
01:01:38,920 –> 01:01:42,319
and get another degree and help us work

1619
01:01:40,799 –> 01:01:44,640
on those problems because there are

1620
01:01:42,319 –> 01:01:46,880
serious problems governance is hard the

1621
01:01:44,640 –> 01:01:49,280
economy has to change security has to

1622
01:01:46,880 –> 01:01:51,079
change because we’re at a situation

1623
01:01:49,280 –> 01:01:53,480
where we can no longer have exponential

1624
01:01:51,079 –> 01:01:55,319
un exponential growth of of the number

1625
01:01:53,480 –> 01:01:58,200
of people so we need to think of a

1626
01:01:55,319 –> 01:02:00,359
better way to stabilize our our our

1627
01:01:58,200 –> 01:02:03,039
biosphere um and we need to do it in a

1628
01:02:00,359 –> 01:02:05,000
way that that presumably that causes as

1629
01:02:03,039 –> 01:02:08,400
little suffering as possible as well as

1630
01:02:05,000 –> 01:02:12,359
as as little further few uh further

1631
01:02:08,400 –> 01:02:13,480
extinctions so um yeah that that’s why I

1632
01:02:12,359 –> 01:02:18,319
think about

1633
01:02:13,480 –> 01:02:22,039
AI okay just to close it how so do you

1634
01:02:18,319 –> 01:02:26,279
see um more benefits or

1635
01:02:22,039 –> 01:02:31,200
more like positives uh helping us to

1636
01:02:26,279 –> 01:02:34,160
combat such um existential like crisises

1637
01:02:31,200 –> 01:02:36,559
oh yeah absolutely I don’t I mean like

1638
01:02:34,160 –> 01:02:38,319
it’s almost it’s Inseparable from who it

1639
01:02:36,559 –> 01:02:41,160
is and what we are to be humans that we

1640
01:02:38,319 –> 01:02:43,119
have we are these these Apes that have

1641
01:02:41,160 –> 01:02:46,319
wielded technology to be in this

1642
01:02:43,119 –> 01:02:48,039
situation but what I see now although um

1643
01:02:46,319 –> 01:02:49,720
we’re using a lot of this capacity to

1644
01:02:48,039 –> 01:02:52,640
kind of slow down change especially the

1645
01:02:49,720 –> 01:02:55,440
powerful when we really have to change

1646
01:02:52,640 –> 01:02:57,160
um we can do it very quickly and so we

1647
01:02:55,440 –> 01:02:59,520
have seen some really you know

1648
01:02:57,160 –> 01:03:01,920
incredible you know if you look at

1649
01:02:59,520 –> 01:03:04,400
compare Co to Spanish Flu compare you

1650
01:03:01,920 –> 01:03:08,319
know 2008 to to

1651
01:03:04,400 –> 01:03:11,400
1929 um you know if you look at Germany

1652
01:03:08,319 –> 01:03:14,400
that we we changed our entire source of

1653
01:03:11,400 –> 01:03:17,160
fuel from basically Russian oil and gas

1654
01:03:14,400 –> 01:03:19,520
to liquid natural gas in six months six

1655
01:03:17,160 –> 01:03:21,880
months you know every business then I

1656
01:03:19,520 –> 01:03:25,359
told you this like most of the industry

1657
01:03:21,880 –> 01:03:27,240
um we reduced uh Demand by 20% and

1658
01:03:25,359 –> 01:03:28,960
that’s after Decades of being total you

1659
01:03:27,240 –> 01:03:31,760
know save the save the polar bears and

1660
01:03:28,960 –> 01:03:34,760
everything you know so so we C we are

1661
01:03:31,760 –> 01:03:36,520
capable of great change we are already

1662
01:03:34,760 –> 01:03:37,960
too slow at it there are already people

1663
01:03:36,520 –> 01:03:40,559
that are drowning and dying of heat

1664
01:03:37,960 –> 01:03:41,920
strokes and whatever um but but we

1665
01:03:40,559 –> 01:03:43,559
aren’t going to be so slow that there’s

1666
01:03:41,920 –> 01:03:46,559
really going to be the end of humanity I

1667
01:03:43,559 –> 01:03:48,119
don’t think um it it seems very unlikely

1668
01:03:46,559 –> 01:03:49,680
scientists always say that say oh she

1669
01:03:48,119 –> 01:03:51,520
wasn’t sure no no no scientists always

1670
01:03:49,680 –> 01:03:55,559
say that so very very very unlikely that

1671
01:03:51,520 –> 01:03:56,960
we would do that it you know the the um

1672
01:03:55,559 –> 01:03:58,720
but we could kill a lot of people if we

1673
01:03:56,960 –> 01:04:01,480
have a nuclear war a lot of people would

1674
01:03:58,720 –> 01:04:03,279
die um and and and if we don’t figure

1675
01:04:01,480 –> 01:04:05,079
out how to help the people in India get

1676
01:04:03,279 –> 01:04:07,160
to parts of the world that AR a little

1677
01:04:05,079 –> 01:04:08,640
less hot than where they are right now

1678
01:04:07,160 –> 01:04:10,279
they’re going to die right so there’s

1679
01:04:08,640 –> 01:04:12,799
like a lot of there’s a lot of

1680
01:04:10,279 –> 01:04:15,640
situations I’m sorry Pakistan

1681
01:04:12,799 –> 01:04:17,400
too so uh you know there’s a lot of

1682
01:04:15,640 –> 01:04:19,000
situations right now we have to figure

1683
01:04:17,400 –> 01:04:20,880
out how to stop uh people from

1684
01:04:19,000 –> 01:04:24,039
flattening cities but it’s not clear to

1685
01:04:20,880 –> 01:04:26,240
how to do that um either you know it’s

1686
01:04:24,039 –> 01:04:27,839
not it’s not easy but there are there’s

1687
01:04:26,240 –> 01:04:29,119
only four or five countries that would

1688
01:04:27,839 –> 01:04:31,839
that are doing that right now but we

1689
01:04:29,119 –> 01:04:33,359
need to stop those guys too so I think

1690
01:04:31,839 –> 01:04:35,400
that we have a lot of real immediate

1691
01:04:33,359 –> 01:04:38,000
problems but we also are coming up with

1692
01:04:35,400 –> 01:04:39,760
Solutions at an incredible rate and I

1693
01:04:38,000 –> 01:04:42,520
think this is

1694
01:04:39,760 –> 01:04:44,760
life it’s constant Innovation and and

1695
01:04:42,520 –> 01:04:46,920
constant attempt to basically keep

1696
01:04:44,760 –> 01:04:49,160
projecting yourself into the future

1697
01:04:46,920 –> 01:04:51,599
Joanna it’s it’s been pleasure thank you

1698
01:04:49,160 –> 01:04:54,079
thank you so much for for sharing all

1699
01:04:51,599 –> 01:04:56,559
this uh I will have to rewatch it

1700
01:04:54,079 –> 01:04:59,240
because so many aspects you’ve um

1701
01:04:56,559 –> 01:05:02,359
mentioned I haven’t uh you know heard or

1702
01:04:59,240 –> 01:05:04,680
read about um but they are very very

1703
01:05:02,359 –> 01:05:07,279
crucial to understand uh to

1704
01:05:04,680 –> 01:05:10,599
understanding the the threats and and

1705
01:05:07,279 –> 01:05:12,160
opportunities of of what we are creating

1706
01:05:10,599 –> 01:05:14,799
well you’re very welcome it was nice to

1707
01:05:12,160 –> 01:05:17,559
meet you nice to meet you J thank you

1708
01:05:14,799 –> 01:05:17,559
bye bye