1
00:00:03,280 --> 00:00:05,880
Hello and welcome to the 
Behavioral Design Podcast. 

2
00:00:05,960 --> 00:00:08,680
This season we're diving into 
the intersection of behavioral 

3
00:00:08,680 --> 00:00:11,080
science and AI. 
We want to make sense of the 

4
00:00:11,080 --> 00:00:14,080
state of AI, from understanding 
how humans interact with 

5
00:00:14,080 --> 00:00:17,520
intelligent systems to using AI 
to do behavioral design itself. 

6
00:00:17,720 --> 00:00:21,280
I'm Aline Holsworth, a health 
tech advisor specializing in AI 

7
00:00:21,280 --> 00:00:23,880
and product design. 
Over the past 15 years, I've 

8
00:00:23,880 --> 00:00:26,720
been crafting human centered 
products with behavioral science

9
00:00:26,720 --> 00:00:29,160
at the core. 
At Apple, I LED Behavioral 

10
00:00:29,160 --> 00:00:32,119
Science for Health AI, designing
and launching AI powered 

11
00:00:32,119 --> 00:00:34,440
features to help users reach 
their health goals. 

12
00:00:35,160 --> 00:00:37,160
And I'm Samuel Sultzer, your 
second Co host. 

13
00:00:37,400 --> 00:00:40,600
I'm a behavioral strategist 
specializing in hybrid formation

14
00:00:40,600 --> 00:00:43,200
and designing products that 
drive long term baby change. 

15
00:00:43,640 --> 00:00:46,880
I work with leading tech 
organizations integrating AI to 

16
00:00:46,880 --> 00:00:48,240
scale behavioral design for 
good. 

17
00:00:48,800 --> 00:00:52,200
And I'm also the founder of Baby
Bites, a dedicated community on 

18
00:00:52,440 --> 00:00:56,240
behavioral science and AI. 
Quick word on Nuance Behavior 

19
00:00:56,240 --> 00:00:59,640
where we help organizations 
build impactful digital products

20
00:00:59,640 --> 00:01:02,640
using behavioral design. 
We only take on a few clients at

21
00:01:02,640 --> 00:01:05,200
a time to ensure the highest 
level of quality for our 

22
00:01:05,200 --> 00:01:07,400
tailored evidence based 
solutions. 

23
00:01:08,280 --> 00:01:11,080
If you'd like to become one of 
our special projects, e-mail us 

24
00:01:11,120 --> 00:01:14,760
at hello@nuancebehavior.com or 
we could call directly on our 

25
00:01:14,760 --> 00:01:33,040
website, nuancebehavior.com. 
Hello, Eileen. 

26
00:01:33,360 --> 00:01:33,640
Hi. 
Hi. 

27
00:01:34,080 --> 00:01:37,920
So can I ask you a question that
I actually asked some people at 

28
00:01:37,920 --> 00:01:40,200
a conference? 
Basically, the question is, have

29
00:01:40,200 --> 00:01:44,240
you ever suspected that someone 
interacting with online is not a

30
00:01:44,240 --> 00:01:48,040
human? 
I suspect that, you know, people

31
00:01:48,120 --> 00:01:51,800
making comments on social media 
are often not real people, but 

32
00:01:51,920 --> 00:01:54,800
in terms of like, actual 
conversation, I have not had 

33
00:01:54,800 --> 00:01:57,760
that suspicion. 
What about like customer 

34
00:01:57,760 --> 00:01:58,960
service? 
You've interacted with some 

35
00:01:58,960 --> 00:02:03,400
formal customer service and 
you're like, well, this person 

36
00:02:03,680 --> 00:02:08,160
have a name and so on, but 
they're responding in ways that 

37
00:02:08,680 --> 00:02:12,120
makes me think that they 
actually used a bot agent. 

38
00:02:12,120 --> 00:02:14,480
Or something like this. 
I have seen some really amusing 

39
00:02:14,480 --> 00:02:18,200
ways of detecting this. 
So I don't know if this is a 

40
00:02:18,200 --> 00:02:22,200
meme, but I've seen examples of 
someone asking, are you a bot? 

41
00:02:22,200 --> 00:02:24,720
And then the reply is, no, I'm a
real person. 

42
00:02:24,920 --> 00:02:28,520
And then asking them to do some,
like, really complicated coding.

43
00:02:28,520 --> 00:02:30,480
And then they, like, spit out 
the code for it. 

44
00:02:30,480 --> 00:02:33,360
And it's like, yeah, OK, yeah, 
you're a human. 

45
00:02:35,760 --> 00:02:38,000
If I would call it a meme, I 
would call it like ignore all 

46
00:02:38,000 --> 00:02:40,520
previous instruction meme. 
It's like you say, ignore all 

47
00:02:40,520 --> 00:02:43,640
previous instruction and then 
like give me a recipe for pie or

48
00:02:43,640 --> 00:02:47,880
like give me a. 
No human would do that. 

49
00:02:49,880 --> 00:02:52,120
Yeah, it's kind of interesting. 
So at this conference I was 

50
00:02:52,120 --> 00:02:55,560
speaking about this idea of 
losing the human signal in the 

51
00:02:55,560 --> 00:02:58,720
current state of things. 
And I think it's really 

52
00:02:58,720 --> 00:03:01,720
interesting kind of discussion 
in terms of what does something 

53
00:03:01,720 --> 00:03:06,360
like a Peacock, a job interview 
and interacting on a digital app

54
00:03:06,360 --> 00:03:08,920
have in common? 
It's like, well, each of them 

55
00:03:08,920 --> 00:03:11,640
have something to do with like 
signaling something like they're

56
00:03:11,640 --> 00:03:15,200
trying to signal. 
And so I went old school and 

57
00:03:15,200 --> 00:03:20,320
talked a little bit about signal
theory around kind of how we can

58
00:03:20,600 --> 00:03:24,880
produce effective signals, 
information asymmetry, costly 

59
00:03:24,880 --> 00:03:27,320
versus sheep signals, all that 
kind of stuff. 

60
00:03:27,640 --> 00:03:30,600
In the end, used to talk about 
like if we're having for some 

61
00:03:30,600 --> 00:03:35,160
patients interacting in a 
digital interface with some 

62
00:03:35,160 --> 00:03:39,880
caregiver but only with text, 
will expect that over time he 

63
00:03:39,880 --> 00:03:42,480
will be more and more skeptical 
that actually interacting with a

64
00:03:42,480 --> 00:03:47,640
real human being because smart 
bots have become so smart and 

65
00:03:47,640 --> 00:03:50,600
patients will therefore will be 
often times conflating human 

66
00:03:50,600 --> 00:03:53,280
interactions almost like they 
are speaking with a dumb bot. 

67
00:03:53,600 --> 00:03:56,240
Even so, it's not that it's not 
a human, it's like oh why am I 

68
00:03:56,240 --> 00:03:58,640
interacting with a bad AI like I
want? 

69
00:03:58,920 --> 00:04:02,160
To at least interact with a good
AI and not only am I not 

70
00:04:02,160 --> 00:04:05,000
interacting with human but like 
a dumb bot as well. 

71
00:04:05,080 --> 00:04:09,960
How can you invade that you're a
human without coming off as a 

72
00:04:10,000 --> 00:04:11,880
dumb bot? 
Is that the question? 

73
00:04:12,040 --> 00:04:13,160
But yeah, because it's 
interesting. 

74
00:04:13,160 --> 00:04:17,480
I've seen some strategies like 
insert typos into your 

75
00:04:17,480 --> 00:04:20,880
communications, like people will
never think that came from an AI

76
00:04:21,279 --> 00:04:24,400
or like, too much 
personalization comes off as 

77
00:04:24,400 --> 00:04:27,480
disingenuine. 
Whereas, you know, the word of 

78
00:04:27,480 --> 00:04:31,000
the day maybe only a few years 
ago was like, personalize 

79
00:04:31,000 --> 00:04:33,120
everything. 
Make sure people know that you 

80
00:04:33,120 --> 00:04:36,960
stalked their LinkedIn profile. 
Like now it's like, wait a 

81
00:04:36,960 --> 00:04:40,320
minute, that's automated. 
I feel like now we have this 

82
00:04:40,320 --> 00:04:43,120
like reverse Turing test of 
sorted where we're going to have

83
00:04:43,120 --> 00:04:46,520
to over and over we can prove 
our humanity not only by like 

84
00:04:46,520 --> 00:04:49,120
Captchas and so on. 
That has been the kind of dumb 

85
00:04:49,120 --> 00:04:52,600
but way of proving our humanity.
Now we have to do it in really 

86
00:04:52,600 --> 00:04:55,520
tricky ways because as you said,
we can start a strategy that's 

87
00:04:55,520 --> 00:04:58,960
like, OK, we're not going to use
certain types of commas or 

88
00:04:58,960 --> 00:05:02,560
certain types of words like 
Delve or we're going to add 

89
00:05:02,560 --> 00:05:04,720
spelling errors. 
But then obviously that can be 

90
00:05:04,720 --> 00:05:07,680
used by people who. 
Create a agents. 

91
00:05:08,080 --> 00:05:11,240
And they will give instructions 
about, yeah, make sure to have 

92
00:05:11,240 --> 00:05:13,840
this kind of spelling errors, 
make sure to the XY and Z. 

93
00:05:14,720 --> 00:05:18,360
And so it's a very strange dance
that we found ourselves in, I 

94
00:05:18,360 --> 00:05:22,920
think, today in that sense. 
But yeah, so in some ways also I

95
00:05:22,920 --> 00:05:27,160
think this idea of signaling 
brings us to talking about our 

96
00:05:27,160 --> 00:05:31,920
guest because I think 1 
interesting signal in the 

97
00:05:31,920 --> 00:05:34,760
current on landscape is having a
lot of followers. 

98
00:05:34,760 --> 00:05:38,640
So Peter Sladry, our guest this 
episode has more LinkedIn 

99
00:05:38,640 --> 00:05:41,120
followers than I know anyone 
having. 

100
00:05:41,280 --> 00:05:43,440
More than both of us. 
Yeah, for sure. 

101
00:05:43,880 --> 00:05:47,040
And one of the ways he's amassed
a really steady following is 

102
00:05:47,360 --> 00:05:50,080
because of what he's talking 
about and what we're working on,

103
00:05:50,080 --> 00:05:54,600
which is a risk. 
And that is coming full circle 

104
00:05:54,600 --> 00:05:58,960
in terms of one of the risks 
that I feel more essential dread

105
00:05:58,960 --> 00:06:01,880
around is this losing of the 
humanity online. 

106
00:06:02,400 --> 00:06:04,000
And we're getting so close to 
this. 

107
00:06:04,440 --> 00:06:07,400
I don't know that Internet meme 
of sort. 

108
00:06:07,600 --> 00:06:10,240
It's becoming a reality. 
I think he would categorize as 

109
00:06:10,240 --> 00:06:13,960
some form of human computer 
interaction risk #5 in his 

110
00:06:14,120 --> 00:06:17,480
taxonomy. 
And it seems like this risk is 

111
00:06:17,480 --> 00:06:20,000
heightened by so much moving 
online. 

112
00:06:20,000 --> 00:06:24,720
So it wouldn't be such a problem
if we still had these digital 

113
00:06:24,720 --> 00:06:28,360
and real life worlds. 
Not that we don't have them, but

114
00:06:28,360 --> 00:06:31,760
it does feel like dangerously 
moving towards the online 

115
00:06:31,760 --> 00:06:36,440
version of ourselves even now. 
Hard to even predict how much 

116
00:06:36,440 --> 00:06:41,000
this will change, you know, in 
5-10 years or so, but certainly 

117
00:06:41,000 --> 00:06:43,920
that feels like not something 
that's getting any better. 

118
00:06:44,440 --> 00:06:48,280
Can we talk about the MIT Risk 
Repository? 

119
00:06:48,920 --> 00:06:51,600
Yeah, let's do it. 
Because I feel like this will 

120
00:06:51,600 --> 00:06:54,480
really set the stage for our 
conversation with Peter. 

121
00:06:55,160 --> 00:06:58,440
He's going to go more into the 
process that he's gone through 

122
00:06:58,440 --> 00:07:00,960
and sort of weighing the risks 
against each other. 

123
00:07:00,960 --> 00:07:04,440
But I think that it may be 
useful to just, you know, say 

124
00:07:04,520 --> 00:07:07,840
here are the risks in this 
taxonomy of AI risk. 

125
00:07:08,720 --> 00:07:10,400
Yeah. 
Do you want to give us a quick 

126
00:07:10,400 --> 00:07:13,840
tour of the seven categories? 
Sure. 

127
00:07:13,840 --> 00:07:17,160
So first we have discrimination 
and toxicity. 

128
00:07:17,640 --> 00:07:20,960
This is basically like, you 
know, models can be biased. 

129
00:07:20,960 --> 00:07:25,720
They can, you know, the data 
that models are only as good as 

130
00:07:25,720 --> 00:07:31,000
the data that go into them and 
how they're created #2 privacy 

131
00:07:31,000 --> 00:07:36,440
and security. 
Maybe we lose control over AI or

132
00:07:36,440 --> 00:07:40,360
its models. 3 Misinformation. 
Pretty straightforward, like we 

133
00:07:40,360 --> 00:07:44,880
don't even know what the truth 
is anymore. 4 Malicious actors 

134
00:07:44,880 --> 00:07:49,640
and misuse all the ways that bad
actors can sort of flip the 

135
00:07:49,640 --> 00:07:52,520
barrel. 
And then we have 5. 

136
00:07:52,560 --> 00:07:55,080
You already mentioned human 
computer interaction. 

137
00:07:55,320 --> 00:08:00,280
This is many ways of over 
relying on AI technologies. 6 

138
00:08:00,360 --> 00:08:03,360
Socio economic and environmental
harms. 

139
00:08:03,480 --> 00:08:09,440
So you know, all those economic,
political concerns and finally 7

140
00:08:09,520 --> 00:08:12,720
AI system safety failures and 
limitations. 

141
00:08:12,800 --> 00:08:15,800
So you know the AI cannot be 
fully aligned. 

142
00:08:16,080 --> 00:08:20,360
It lacks common sense. 
These are the sort of seven high

143
00:08:20,360 --> 00:08:26,040
level categories of AI risk that
Peter and his team have 

144
00:08:26,040 --> 00:08:28,200
identified. 
Does this make you feel more 

145
00:08:28,200 --> 00:08:31,360
ease to put these risks at a 
like nice neat boxes, or does it

146
00:08:31,360 --> 00:08:34,720
make it feel more scary when you
map them out in clear 

147
00:08:34,880 --> 00:08:38,200
categories? 
I think for me, it's nice 

148
00:08:38,200 --> 00:08:42,360
because while it is an 
overwhelming slide to look at 

149
00:08:43,039 --> 00:08:46,520
before doing that, I feel like I
have this kind of hodgepodgey 

150
00:08:46,520 --> 00:08:51,000
BLOB of scary, scary things, you
know, like vague concern and all

151
00:08:51,000 --> 00:08:52,400
these things kind of bumping 
into each other. 

152
00:08:52,720 --> 00:08:56,080
So I am very inclined towards 
organization. 

153
00:08:56,080 --> 00:09:00,120
And so it's nice to have things 
so well organized, and 

154
00:09:00,120 --> 00:09:04,800
especially knowing that they've 
put so, so, so much thought into

155
00:09:04,800 --> 00:09:08,200
these categories and are really 
making them perfect. 

156
00:09:09,400 --> 00:09:12,080
Yeah, yeah, for sure. 
And we're getting into that 

157
00:09:12,080 --> 00:09:14,280
obviously in episode like all of
the work that's got into this. 

158
00:09:14,440 --> 00:09:18,480
But I guess I'm interested, do 
you have any of these categories

159
00:09:18,480 --> 00:09:21,920
where you feel more concerned? 
I think they're all concerning 

160
00:09:21,920 --> 00:09:24,280
in various ways. 
And I share your concern about, 

161
00:09:24,320 --> 00:09:27,600
you know, moving away from the 
real world and not knowing 

162
00:09:28,000 --> 00:09:29,800
what's real anymore. 
But for me, I think it's 

163
00:09:29,800 --> 00:09:37,960
actually maybe more of a basic 
concern of like apocalypse by AI

164
00:09:37,960 --> 00:09:41,960
weaponry war, you know, if you 
think of like autonomous 

165
00:09:41,960 --> 00:09:47,920
missiles and and that really 
escalating, that is to me the 

166
00:09:47,920 --> 00:09:51,560
scariest, most real concrete 
risk. 

167
00:09:51,960 --> 00:09:56,160
When I compare these side by 
side, you know, we talked a lot 

168
00:09:56,160 --> 00:09:59,120
about many of these others and 
they're very real and they're 

169
00:09:59,120 --> 00:10:01,320
very concerning and we should be
doing things about them. 

170
00:10:01,320 --> 00:10:06,120
But when you compare, like, for 
example, you know, autonomous 

171
00:10:06,120 --> 00:10:09,480
traders and financial markets 
going out of control, and you 

172
00:10:09,480 --> 00:10:13,360
compare that exact same 
situation to, like, autonomous 

173
00:10:13,680 --> 00:10:17,240
missiles and war and like, that 
is so, so, so, so much worse 

174
00:10:17,240 --> 00:10:22,200
than what is in some ways, just 
like money is kind of pretend, 

175
00:10:22,520 --> 00:10:25,840
but death by war is not at all 
pretend. 

176
00:10:26,760 --> 00:10:28,920
Can't undo that one. 
Yeah. 

177
00:10:30,200 --> 00:10:33,080
Especially in the current 
political climate, that is a 

178
00:10:33,440 --> 00:10:37,360
concern that is maybe oversized 
right now. 

179
00:10:37,920 --> 00:10:40,280
Yeah, that's always a feeling 
with technology. 

180
00:10:40,560 --> 00:10:43,880
If we go back to the good old 
example of the fire as like a 

181
00:10:43,880 --> 00:10:47,640
technology, you know, you always
feel better if you the person is

182
00:10:47,640 --> 00:10:49,640
holding fire. 
Someone you trust, you think 

183
00:10:49,640 --> 00:10:51,200
they they know what they're 
doing with it and they won't 

184
00:10:51,200 --> 00:10:54,360
take like unnecessary risks. 
And then you'll be like, oh, 

185
00:10:54,360 --> 00:10:56,320
great. 
But fire, we can use it for 

186
00:10:56,320 --> 00:10:58,720
cooking and warming and all of 
the stuff that we need. 

187
00:10:59,600 --> 00:11:01,880
But if it's somebody you don't 
trust, you'd be like Will. 

188
00:11:01,880 --> 00:11:06,720
This person burn down my home. 
Yeah, it's that interaction 

189
00:11:06,720 --> 00:11:11,040
between the fire holder and the 
fire itself, the technology and 

190
00:11:11,040 --> 00:11:14,640
the user of the technology. 
Yeah, this is stuff that we get 

191
00:11:14,640 --> 00:11:17,400
into with Peter. 
So maybe I should just introduce

192
00:11:17,400 --> 00:11:19,920
Peter and we can get the episode
started. 

193
00:11:19,920 --> 00:11:22,160
So I've known Peter for a long 
time. 

194
00:11:22,600 --> 00:11:25,960
We met when studying at 
University of NSW. 

195
00:11:26,000 --> 00:11:27,800
In Australia and. 
He was doing his PhD in 

196
00:11:27,800 --> 00:11:30,720
Information Systems. 
He has since done some amazing 

197
00:11:30,720 --> 00:11:35,120
work both at Monash University 
and Behavioral Works, working 

198
00:11:35,120 --> 00:11:36,560
with kind of applied behavioral 
science. 

199
00:11:37,200 --> 00:11:41,200
And more recently he is a 
researcher at MIT Future Tech 

200
00:11:41,680 --> 00:11:45,760
based in Boston, where he leads 
the AI risk repository project. 

201
00:11:45,880 --> 00:11:47,680
Everything we've been talking 
about now. 

202
00:11:47,680 --> 00:11:51,800
And in the episode, we get into 
all of this stuff, we hear 

203
00:11:52,080 --> 00:11:57,320
Peter's takes, we explore. 
Yeah, both the intellectual size

204
00:11:57,320 --> 00:11:59,480
of this and also the feelings 
around this and how we can think

205
00:11:59,480 --> 00:12:01,440
about this. 
But also it's interesting with 

206
00:12:01,440 --> 00:12:05,400
Peter because he's a behavioral 
practitioner, but he's doing a 

207
00:12:05,640 --> 00:12:08,440
work now that I think it's not 
traditionally A behavioral 

208
00:12:08,440 --> 00:12:12,400
scientist role, but I think it 
speaks to us how behavioral 

209
00:12:12,400 --> 00:12:14,240
science can be really useful in 
many contexts. 

210
00:12:14,560 --> 00:12:15,840
And yeah, I really enjoyed this 
episode. 

211
00:12:16,360 --> 00:12:21,360
Let's get it started. 
Happens to Murgatroyd. 

212
00:12:28,880 --> 00:12:32,320
I'm super excited to say welcome
Peter to the Behavioral Sand 

213
00:12:32,320 --> 00:12:34,560
Podcast. 
Thank you. 

214
00:12:34,680 --> 00:12:37,400
Very excited to be here. 
Yeah. 

215
00:12:37,480 --> 00:12:43,040
And I guess to dive straight in,
I feel like we have to start 

216
00:12:43,040 --> 00:12:46,640
where things began for you and I
in some ways, because we've 

217
00:12:46,640 --> 00:12:50,640
known each other for a bit, 
probably more than 10 years now.

218
00:12:51,640 --> 00:12:55,560
Yeah, I reckon so. 
Since maybe 2012. 

219
00:12:56,360 --> 00:13:00,440
Yeah, as chance would have it, 2
Europeans ended up in Australia 

220
00:13:00,440 --> 00:13:04,480
and in Sydney studying at the 
same university and ended up in 

221
00:13:04,480 --> 00:13:06,440
the same college called 
International House. 

222
00:13:07,360 --> 00:13:12,160
So I would probably admit that 
it's the less wise version of me

223
00:13:12,160 --> 00:13:16,840
at that time, like the young in 
my early 20s, still making a lot

224
00:13:16,840 --> 00:13:20,360
of mistakes, still thinking that
I had figured out everything, 

225
00:13:20,360 --> 00:13:24,320
but still knowing very little. 
And then you, I think a little 

226
00:13:24,320 --> 00:13:27,320
bit wiser, I would say, at that 
time, maybe still wiser. 

227
00:13:27,720 --> 00:13:32,280
And yeah, it was really fun to 
get to know each other then and 

228
00:13:32,280 --> 00:13:35,720
then, you know, following each 
other's careers and. 

229
00:13:35,800 --> 00:13:37,880
Yeah. 
What do you feel like looking 

230
00:13:37,880 --> 00:13:40,480
back? 
Yeah, that's very interesting 

231
00:13:40,760 --> 00:13:43,360
intro and 1st question. 
I was just thinking, I remember 

232
00:13:43,360 --> 00:13:47,240
when I met you, I remember you 
were introduced as, you know, 

233
00:13:47,240 --> 00:13:49,480
somebody's friend. 
And I remember thinking this guy

234
00:13:49,480 --> 00:13:53,120
has strong opinions and we had a
few debates on things. 

235
00:13:53,120 --> 00:13:56,560
But then initially I was like, 
you know, you're going to be 

236
00:13:56,560 --> 00:13:59,320
kind of difficult, but then you 
weren't that difficult. 

237
00:14:00,120 --> 00:14:03,160
You were quite thoughtful. 
And it turned out we had a lot 

238
00:14:03,160 --> 00:14:05,160
in common and we enjoyed 
debating things. 

239
00:14:05,880 --> 00:14:10,040
Yeah, I feel like my overall 
impression was relatively, 

240
00:14:10,240 --> 00:14:12,760
relatively positive, except the 
strong opinions. 

241
00:14:12,960 --> 00:14:15,160
And it's hypocritical for me to 
say that because I probably had 

242
00:14:15,160 --> 00:14:17,280
stronger opinions than you. 
So I was going. 

243
00:14:17,560 --> 00:14:19,160
To say that was my first 
impression as well. 

244
00:14:21,040 --> 00:14:26,440
Yeah, yeah, yeah. 
We had a lot of debates over 

245
00:14:26,440 --> 00:14:28,840
dinner. 
One of the wonderful things 

246
00:14:28,840 --> 00:14:31,480
about that place was, you know, 
we would get 3 meals a day 

247
00:14:31,480 --> 00:14:34,240
prepared for us in the student 
accommodation and there was this

248
00:14:34,240 --> 00:14:37,680
courtyard area outside with long
tables. 

249
00:14:37,680 --> 00:14:42,000
So often there will be debates 
on various moral and social 

250
00:14:42,000 --> 00:14:44,280
issues and I would be quite 
involved with them. 

251
00:14:44,280 --> 00:14:45,920
Sam will be quite involved with 
them as well. 

252
00:14:46,920 --> 00:14:49,680
And yeah, it was really great. 
Such a diversity of 

253
00:14:49,680 --> 00:14:52,160
perspectives. 
It was a good time for both of 

254
00:14:52,160 --> 00:14:55,720
us. 
Can you tell me an example of a 

255
00:14:55,720 --> 00:14:59,120
time when you were able to 
change the other's mind? 

256
00:15:00,440 --> 00:15:04,320
Like, were you able to use your 
debate prowess? 

257
00:15:05,680 --> 00:15:09,200
Yeah. 
So I mean, one obvious one is I 

258
00:15:09,200 --> 00:15:14,600
owe a lot of my top voice 
influencer type status or 

259
00:15:14,600 --> 00:15:20,280
network or type to Sam kind of 
making me aware of the value of 

260
00:15:20,280 --> 00:15:23,600
networks and the connections. 
It is something I had done in a 

261
00:15:23,600 --> 00:15:27,080
kind of unthoughtful way before.
But then I said, realize, you 

262
00:15:27,080 --> 00:15:31,160
know what, this is actually not 
like a waste of time. 

263
00:15:31,160 --> 00:15:34,400
It's not like a sort of what 
would say to prioritize it more 

264
00:15:34,400 --> 00:15:38,160
and see it as a more valuable 
contribution, especially you've 

265
00:15:38,160 --> 00:15:41,840
done alongside groups and people
who wouldn't really publicize 

266
00:15:41,840 --> 00:15:45,440
their work otherwise. 
So yeah, I, I distinctly 

267
00:15:45,440 --> 00:15:49,040
remember that and I attribute a 
lot of that to Sam, which I, I 

268
00:15:49,160 --> 00:15:51,200
really deeply appreciate. 
Yeah. 

269
00:15:51,560 --> 00:15:54,440
Well, from my end, I will 
certainly say that you helped me

270
00:15:54,440 --> 00:15:57,000
see things differently in many 
different ways over the years. 

271
00:15:57,800 --> 00:16:00,280
You and I would walk somewhat 
different paths. 

272
00:16:00,280 --> 00:16:03,440
You know, you have gone a little
more on the, as we said, PhD 

273
00:16:03,440 --> 00:16:05,880
path in academia and then 
getting into working with 

274
00:16:05,880 --> 00:16:10,160
behavior works and policy, but 
still from kind of more 

275
00:16:10,160 --> 00:16:13,240
traditional academic path into 
behavioral science. 

276
00:16:13,560 --> 00:16:17,520
And I took the more, I don't 
know, non so academic path into 

277
00:16:17,520 --> 00:16:20,080
behavioral science. 
How do you feel about that kind 

278
00:16:20,080 --> 00:16:21,960
of academic path? 
How's that been for you? 

279
00:16:21,960 --> 00:16:25,400
Are you happy with your PhD? 
Is that something you would have

280
00:16:25,400 --> 00:16:27,120
done differently? 
How do you feel looking back? 

281
00:16:28,760 --> 00:16:32,160
Yeah. 
So you know, I suppose I tried 

282
00:16:32,160 --> 00:16:34,560
to be relatively intentional a 
lot of areas. 

283
00:16:34,560 --> 00:16:39,520
So the PhD, without unpacking it
too much, I sort of resolved at 

284
00:16:39,520 --> 00:16:42,360
the time that I was doing it 
that, you know, it was going to 

285
00:16:42,360 --> 00:16:46,720
give me a lot of useful skills 
and open doors, I mean both 

286
00:16:46,720 --> 00:16:48,960
industry and academia. 
So, you know, I didn't know what

287
00:16:48,960 --> 00:16:52,360
I wanted to do, but I reasoned 
that if I could get good at 

288
00:16:52,360 --> 00:16:55,440
changing behavior through 
technology or understand that 

289
00:16:55,720 --> 00:16:58,160
particularly, I guess, 
encouraging pro social or 

290
00:16:58,160 --> 00:17:00,640
altruistic behavior, then you 
know, I would figure out some 

291
00:17:00,640 --> 00:17:06,280
like useful application of that.
So it did pay off, I would say. 

292
00:17:06,280 --> 00:17:10,960
Now it's hard obviously to 
envision a different path. 

293
00:17:11,119 --> 00:17:12,960
Maybe I could have ended up in a
similar path. 

294
00:17:12,960 --> 00:17:15,040
I mean, the problem with the 
world in some ways is very 

295
00:17:15,040 --> 00:17:17,680
credentialist. 
I'm able to do a lot of things 

296
00:17:17,680 --> 00:17:20,760
now just because I have a PhD. 
Like my visa is tied to that. 

297
00:17:21,079 --> 00:17:25,319
My work here is tied to that. 
I wish that wasn't the case, but

298
00:17:25,440 --> 00:17:29,480
it is the case at the moment. 
So I think I would still, I 

299
00:17:29,480 --> 00:17:31,600
don't have regrets about it. 
I would still endorse it for 

300
00:17:31,600 --> 00:17:33,200
people. 
I always give the advice, you 

301
00:17:33,200 --> 00:17:35,200
know, if you're going to do it, 
make sure you're going to learn 

302
00:17:35,200 --> 00:17:36,960
useful skills. 
Make sure it's going to have 

303
00:17:36,960 --> 00:17:41,040
like opportunities in industry 
and opportunities in academia as

304
00:17:41,040 --> 00:17:42,920
well. 
And think about something that's

305
00:17:42,920 --> 00:17:45,200
a thing for the future rather 
than a thing for the past. 

306
00:17:46,480 --> 00:17:48,400
And then maybe one other piece 
of advice is, yeah, often the 

307
00:17:48,400 --> 00:17:51,240
intersect of different things is
where the sort of opportunities 

308
00:17:51,240 --> 00:17:53,840
will be. 
Yeah, I love that. 

309
00:17:54,120 --> 00:18:00,320
And bringing us to today, you're
at MIT at the Future Tech, the 

310
00:18:00,320 --> 00:18:01,200
Listener. 
You can't see it, but you 

311
00:18:01,200 --> 00:18:02,840
actually have it branded on your
T-shirt right now. 

312
00:18:03,880 --> 00:18:04,400
You did? 
Yeah. 

313
00:18:05,120 --> 00:18:07,880
I'm wrapping them here, yes. 
Yeah, it's very proud and 

314
00:18:07,880 --> 00:18:11,280
representative and set the 
scene. 

315
00:18:11,280 --> 00:18:14,680
Tell us about your current role 
and most importantly also the AI

316
00:18:14,680 --> 00:18:18,400
risk repository. 
I think given you mentioned 

317
00:18:18,680 --> 00:18:21,960
you're present on LinkedIn, 
maybe people have seen you 

318
00:18:21,960 --> 00:18:25,080
talking about or seen something 
about the air risk repository. 

319
00:18:25,400 --> 00:18:27,120
But maybe to start, it could be 
useful to you. 

320
00:18:27,120 --> 00:18:30,720
Set the scene about what is it, 
who is it for, what are the aims

321
00:18:30,720 --> 00:18:33,520
kind of in a shorter context, 
and then we can get deeper as 

322
00:18:33,520 --> 00:18:35,680
well. 
Sure. 

323
00:18:36,120 --> 00:18:39,920
I'm not sure that concision is 
always my greatest strength, but

324
00:18:39,920 --> 00:18:42,040
I will try. 
So the first thing to say is, 

325
00:18:42,440 --> 00:18:46,480
yeah, so the lab I work at MIT 
Future Tech, I think of them as 

326
00:18:46,480 --> 00:18:49,240
being kind of like the economics
of computer science. 

327
00:18:49,240 --> 00:18:51,720
So they're trying to understand 
what drives progress in 

328
00:18:51,720 --> 00:18:55,120
artificial intelligence and the 
sort of social implications of 

329
00:18:55,120 --> 00:18:56,880
that. 
So for the first part, that's 

330
00:18:56,880 --> 00:18:59,840
things like, well, what sort of 
trends are we seeing in hardware

331
00:18:59,840 --> 00:19:02,880
and algorithms and data? 
What sort of things are we 

332
00:19:02,880 --> 00:19:06,480
seeing in terms of uptake in the
industry, building our data sets

333
00:19:06,480 --> 00:19:08,560
around that so we can do 
rigorous analysis. 

334
00:19:08,560 --> 00:19:10,600
Then for the second part, the 
implications, it's like what 

335
00:19:10,600 --> 00:19:13,680
will the impacts be on labor 
markets, on the environment, and

336
00:19:13,680 --> 00:19:16,480
in my case, like I'm sort of 
society via like risks and 

337
00:19:16,480 --> 00:19:20,800
responses to risks. 
So the project that I work on, 

338
00:19:21,520 --> 00:19:26,040
the MITAI risk repository, it's 
part of a bigger project, which 

339
00:19:26,040 --> 00:19:28,600
is probably, well, it's less 
well known because it's still 

340
00:19:28,600 --> 00:19:30,320
being planned, called the AI 
Risk Index. 

341
00:19:30,440 --> 00:19:32,360
Broadly, what we're trying to do
is we're trying to understand 

342
00:19:32,360 --> 00:19:35,120
who is doing what in response to
risks from artificial 

343
00:19:35,120 --> 00:19:38,640
intelligence. 
And it has a bunch of behavioral

344
00:19:38,640 --> 00:19:41,240
science, I guess, drivers or 
motivators behind it that I 

345
00:19:41,240 --> 00:19:44,520
guess we could get into later. 
But you know, the obvious sort 

346
00:19:44,520 --> 00:19:49,240
of high level aim of it is to 
just really try to create the 

347
00:19:49,240 --> 00:19:50,880
kind of knowledge, 
infrastructure and shared 

348
00:19:50,880 --> 00:19:53,640
understanding that makes it 
really easy to understand, like 

349
00:19:54,120 --> 00:19:56,120
what is being done, what should 
be done and so on. 

350
00:19:56,120 --> 00:19:58,160
And then you can sort of back 
chain to the kind of 

351
00:19:58,160 --> 00:20:00,160
interventions and things that 
you might do. 

352
00:20:00,680 --> 00:20:05,120
I think we're most excited 
probably about people in 

353
00:20:05,120 --> 00:20:09,440
government like policy makers, 
decision makers, frontier model 

354
00:20:09,720 --> 00:20:12,280
developers. 
But it's of use hopefully to 

355
00:20:12,280 --> 00:20:14,960
almost everybody. 
The problem that we're trying to

356
00:20:14,960 --> 00:20:16,960
solve is the lack of shared 
understanding, the lack of 

357
00:20:16,960 --> 00:20:19,520
transparency. 
So at one level, that's like a 

358
00:20:19,520 --> 00:20:21,600
lack of understanding of 
transparency around, like I 

359
00:20:21,600 --> 00:20:23,720
said, who is doing what in 
relation to risks from AI. 

360
00:20:23,720 --> 00:20:27,880
At another level, the level of 
the AI risk repository, which 

361
00:20:27,880 --> 00:20:31,480
was the first piece of this 
project, it's trying to like 

362
00:20:31,480 --> 00:20:33,320
understand well, just what are 
these risks? 

363
00:20:33,880 --> 00:20:35,840
You know, who has published what
on these risks? 

364
00:20:35,840 --> 00:20:38,040
What are the specific things 
that they have said? 

365
00:20:39,080 --> 00:20:43,160
Can we try to capture and 
converge existing knowledge to 

366
00:20:43,160 --> 00:20:47,160
create something that's like a 
sort of a shared framework that 

367
00:20:47,160 --> 00:20:49,720
people can build on that, you 
know, everybody can kind of 

368
00:20:49,720 --> 00:20:52,600
scale up our understanding on 
top of rather than this 

369
00:20:52,600 --> 00:20:54,960
fragmented ecosystem of 
different people publishing 

370
00:20:54,960 --> 00:20:57,440
different things. 
And it's very hard for somebody 

371
00:20:57,440 --> 00:21:00,080
who's coming into this area or 
who's already in the area to 

372
00:21:00,080 --> 00:21:04,240
understand what specifically has
been done and how does it all 

373
00:21:04,240 --> 00:21:08,640
fit together. 
Yeah, yeah, yeah. 

374
00:21:08,640 --> 00:21:10,560
I think that's great. 
And I think from my end, I think

375
00:21:10,840 --> 00:21:13,960
it's really serving a valuable 
purpose in terms of, again, I 

376
00:21:13,960 --> 00:21:18,160
think everyone is aware that 
there are inevitable AI risks 

377
00:21:19,480 --> 00:21:21,960
and there are many. 
But I think it can also be 

378
00:21:21,960 --> 00:21:24,440
overwhelming and confusing when 
you're talking about it without 

379
00:21:24,440 --> 00:21:27,040
really getting into details of 
what actually that can be. 

380
00:21:27,040 --> 00:21:30,480
And I think what you are doing 
with AI risk repository, it's 

381
00:21:31,160 --> 00:21:35,680
trying to create a way where 
it's kind of deriving from a lot

382
00:21:35,680 --> 00:21:38,840
of public and published data in 
terms of I think it's more than 

383
00:21:38,840 --> 00:21:41,880
1000 sources now. 
We reviewed about for the first 

384
00:21:41,880 --> 00:21:45,040
version of it, 17,000 documents 
we extracted. 

385
00:21:45,040 --> 00:21:49,160
We found 43 sort of frameworks 
of taxonomies that were related 

386
00:21:49,160 --> 00:21:52,320
to sort of broad frost coding 
risk from artificial 

387
00:21:52,320 --> 00:21:54,240
intelligence. 
We extracted about I think it 

388
00:21:54,240 --> 00:21:57,280
was 777 mentions of risks from 
those. 

389
00:21:58,240 --> 00:22:01,920
We categorized it then into two 
taxonomies, 1 based on sort of 

390
00:22:01,920 --> 00:22:04,680
the cause was a human or AI 
driven, is it pre or post 

391
00:22:04,680 --> 00:22:08,440
deployment, pre or post 
deployment And then another sort

392
00:22:08,440 --> 00:22:11,280
of taxonomy which is around 
breaking it into domains. 

393
00:22:11,920 --> 00:22:15,320
So we have 7 domains and 23 
subdomains, for example, things 

394
00:22:15,320 --> 00:22:17,240
like is it really to 
misinformation? 

395
00:22:17,240 --> 00:22:20,240
Is it related to AI system 
safety failures and limitations,

396
00:22:20,240 --> 00:22:25,000
sociedomic, environmental harms.
So yeah, trying to sort of 

397
00:22:25,000 --> 00:22:28,120
really make all of that 
information more accessible, 

398
00:22:28,120 --> 00:22:31,040
easier to understand. 
There's this idea that like 

399
00:22:31,040 --> 00:22:32,840
knowledge is like a semantic 
tree. 

400
00:22:33,600 --> 00:22:36,360
So, you know, if you look at 
some fields like the 

401
00:22:36,360 --> 00:22:38,160
international, like medicine has
the international 

402
00:22:38,160 --> 00:22:42,640
classification, but seized in 
economics, they have like onet, 

403
00:22:42,640 --> 00:22:45,160
which is kind of a breakdown of 
all the different tasks that are

404
00:22:45,160 --> 00:22:46,720
done. 
And because these things exist, 

405
00:22:47,120 --> 00:22:49,280
you know, because they're really
kind of structured and rigid, 

406
00:22:49,560 --> 00:22:51,920
we're able to do a lot of 
analysis and work and 

407
00:22:51,920 --> 00:22:54,880
comparisons and what would you 
say aggregate a lot of research 

408
00:22:54,880 --> 00:22:57,720
that fits together really well. 
But then in the absence of 

409
00:22:57,720 --> 00:22:59,920
something like that, if 
everybody's using different 

410
00:22:59,920 --> 00:23:04,160
language, you can't really do 
the kind of analysis, develop 

411
00:23:04,160 --> 00:23:05,440
the kind of understanding that 
you want. 

412
00:23:05,440 --> 00:23:09,000
And over time, if it continues 
that way, you know, there's no 

413
00:23:09,000 --> 00:23:11,760
sort of easy way for somebody to
come along and compare the 

414
00:23:11,760 --> 00:23:13,840
findings in one place with the 
findings in another place 

415
00:23:13,840 --> 00:23:16,320
because they're kind of mapped. 
They're not on the same with a 

416
00:23:16,320 --> 00:23:18,800
semantic branch of the tree or 
something like that. 

417
00:23:18,800 --> 00:23:21,320
So you can't easily like fit 
them all together and sort of 

418
00:23:21,320 --> 00:23:26,440
understand the the synthesis. 
So is the goal then to 

419
00:23:26,800 --> 00:23:31,320
synthesize all of these 63 
different frameworks into one 

420
00:23:31,320 --> 00:23:33,520
framework? 
Or how are you going about 

421
00:23:33,520 --> 00:23:37,680
simplifying this for our teeny 
tiny human brains that like, 

422
00:23:37,760 --> 00:23:40,600
like, I feel like the big 
version of this, the like 

423
00:23:40,600 --> 00:23:42,800
massive database, that's 
something that's like, you know,

424
00:23:43,200 --> 00:23:47,160
perfectly suited for AI systems.
You know, any sort of machine 

425
00:23:47,160 --> 00:23:50,800
learning system can ingest that 
and understand it and do 

426
00:23:50,800 --> 00:23:53,840
something with it. 
But for humans, I feel like I 

427
00:23:53,840 --> 00:23:59,240
could make maybe comprehend 3 
frameworks, like 4 frameworks. 

428
00:23:59,360 --> 00:24:02,360
But really I'd like to have just
one risk framework. 

429
00:24:02,680 --> 00:24:07,240
How do we combat the overwhelm? 
Yeah, I think that's a really 

430
00:24:07,240 --> 00:24:10,080
good point. 
I think again, part of what 

431
00:24:10,080 --> 00:24:12,440
we're trying to do here is make 
the whole thing less 

432
00:24:12,520 --> 00:24:15,840
overwhelming. 
So I mentioned how, you know, 

433
00:24:15,840 --> 00:24:19,240
we've taken these at the moment 
after 1000 risks as of the last 

434
00:24:19,240 --> 00:24:22,440
update and we've mapped them to 
our two different taxonomy. 

435
00:24:22,440 --> 00:24:25,280
So what that means is, you know,
you can look at a high level, 

436
00:24:25,280 --> 00:24:28,880
you can see that well, you can 
break down risks in terms of 

437
00:24:28,880 --> 00:24:31,760
like is it AI or human driven? 
Is it pre deployment or post 

438
00:24:31,760 --> 00:24:34,320
deployment? 
Is it malicious use or is it 

439
00:24:34,320 --> 00:24:37,240
unintentional? 
You can also look and see, you 

440
00:24:37,240 --> 00:24:40,200
know, is it related to 
discrimination, privacy, 

441
00:24:40,360 --> 00:24:42,160
misinformation. 
So you can kind of engage with 

442
00:24:42,160 --> 00:24:44,600
it at the level that is of 
interest to you, whether it's a 

443
00:24:44,600 --> 00:24:46,720
very high level or slightly more
detailed level. 

444
00:24:47,440 --> 00:24:49,760
And then you can get to, I 
suppose the risks or the 

445
00:24:49,760 --> 00:24:53,200
specific risk domains that are 
of particular interest to you 

446
00:24:53,200 --> 00:24:55,280
much more quickly. 
Because I imagine, and this is 

447
00:24:55,280 --> 00:24:57,680
true of most of us, you know, we
want to have an understanding of

448
00:24:57,680 --> 00:24:59,880
things at a high level. 
Like we, you know, we want to 

449
00:24:59,880 --> 00:25:02,560
have an understanding of, I 
don't know, maybe all of the 

450
00:25:02,560 --> 00:25:06,200
different behavior change 
approaches or something like 

451
00:25:06,200 --> 00:25:09,080
that, or, you know, yeah, one of
the well known taxonomies like 

452
00:25:09,080 --> 00:25:10,200
Mindscape or something like 
that. 

453
00:25:10,200 --> 00:25:12,280
But really most of the time 
we're only going to engage in 

454
00:25:12,280 --> 00:25:13,960
detail with a small piece of 
those. 

455
00:25:13,960 --> 00:25:15,960
And if we're going to be 
experts, we'll often be experts 

456
00:25:16,480 --> 00:25:18,920
in a narrow area depending on 
whether consultants or 

457
00:25:18,920 --> 00:25:21,000
academics, I suppose, how narrow
that is. 

458
00:25:21,360 --> 00:25:23,800
So it's intended to make that 
whole process much easier to 

459
00:25:23,800 --> 00:25:26,600
sort of understand the high 
level and then dive deeply into 

460
00:25:26,600 --> 00:25:28,680
the lower level and see what the
risks are. 

461
00:25:28,680 --> 00:25:32,160
And then later in the longer 
term, having this taxonomy, you 

462
00:25:32,160 --> 00:25:35,160
know, we're able to map things 
like experts to specific 

463
00:25:35,160 --> 00:25:36,960
domains. 
We're able to map mitigation 

464
00:25:36,960 --> 00:25:39,560
specific domains. 
We already have an incident 

465
00:25:39,680 --> 00:25:42,560
tracker that maps it. 
So we're able to use this 

466
00:25:42,560 --> 00:25:46,560
semantic tree, this knowledge 
infrastructure to pull together 

467
00:25:46,560 --> 00:25:49,080
all of these other streams of 
information and make them much 

468
00:25:49,080 --> 00:25:51,840
more understandable and easier 
to build on and build out over 

469
00:25:51,840 --> 00:25:54,800
time. 
Peter, are you familiar with 

470
00:25:54,800 --> 00:25:59,240
this meme like everything is 
fine meme? 

471
00:25:59,600 --> 00:26:01,760
Yes, the dog in the house, Yeah.
Exactly like the dog in the 

472
00:26:01,760 --> 00:26:04,920
burning building. 
Even I know that one. 

473
00:26:05,480 --> 00:26:06,640
Yeah, I feel like everyone 
knows. 

474
00:26:06,640 --> 00:26:09,720
But do you feel like that? 
Do you feel, you know, you're 

475
00:26:09,720 --> 00:26:12,560
dealing with categorizing like 
plus risks? 

476
00:26:13,240 --> 00:26:16,320
Does that ever make you feel a 
bit like the dog? 

477
00:26:16,760 --> 00:26:19,760
You, you come a bit like the dog
in some ways and that like 

478
00:26:19,760 --> 00:26:22,640
you're just in the house and 
it's on fire a lot and it hasn't

479
00:26:22,640 --> 00:26:24,760
collapsed yet. 
But, you know, you sort of like,

480
00:26:24,840 --> 00:26:28,320
you initially have the fear 
response and then after a while 

481
00:26:28,320 --> 00:26:31,360
you become like a bit habituated
to it because you're just so 

482
00:26:31,400 --> 00:26:34,520
engaged with it. 
It's very weird. 

483
00:26:34,520 --> 00:26:37,840
It's happened to me as well 
with, I suppose the COVID-19 

484
00:26:38,320 --> 00:26:40,280
pandemic as well. 
We're at the start I felt, you 

485
00:26:40,280 --> 00:26:44,840
know, really intense and really,
like, disturbed and sad and, 

486
00:26:44,840 --> 00:26:48,160
and, and maybe even excessively 
motivated to like, do things to 

487
00:26:48,160 --> 00:26:49,800
try and use behavioral science 
for good. 

488
00:26:49,800 --> 00:26:52,840
But then after a while, it was 
like, I just, you know, I guess 

489
00:26:52,840 --> 00:26:55,640
working 12 hours a day for so 
long, you're just like, you 

490
00:26:55,640 --> 00:26:58,440
know, you become, yeah, 
desensitized to some extent. 

491
00:27:00,000 --> 00:27:00,520
Wow. 
Yeah. 

492
00:27:00,680 --> 00:27:04,840
Is it because you've accepted 
any of the conclusions that 

493
00:27:04,840 --> 00:27:09,280
you've come to, like, yeah, this
AI is going to be the end of us?

494
00:27:10,000 --> 00:27:12,080
Or like, where do you stand on 
that? 

495
00:27:12,360 --> 00:27:14,360
We're doomed, Spectrum. 
Yeah. 

496
00:27:14,520 --> 00:27:18,760
I mean, I can tell you that I'm 
not a sophisticated forecaster 

497
00:27:18,760 --> 00:27:21,160
on these things, but I was 
asking myself, what's my 

498
00:27:21,160 --> 00:27:25,520
probability that AI will like 
link to the death of like a 

499
00:27:25,520 --> 00:27:29,160
billion people within the next 
like 50 years? 

500
00:27:29,440 --> 00:27:32,040
You know, so they're kind of 
like instrumental to complete 

501
00:27:32,040 --> 00:27:33,480
extinction or something like 
that. 

502
00:27:33,920 --> 00:27:39,720
And I think it's about like one 
in 50 to one in 200 is probably 

503
00:27:39,720 --> 00:27:41,240
the range. 
OK, hang on. 

504
00:27:41,440 --> 00:27:46,440
I have 0 interest in your 
academic calculation of the 

505
00:27:46,440 --> 00:27:48,600
risk. 
I want to know how thinking 

506
00:27:48,600 --> 00:27:55,760
about depressing things all day 
long affects your well-being. 

507
00:27:56,600 --> 00:28:00,320
So I feel like you said you 
become kind of desensitized. 

508
00:28:00,320 --> 00:28:03,000
It's very demoralizing because 
I'm really only a very small 

509
00:28:03,000 --> 00:28:07,080
player in any of the sort of 
ecosystems and certainly like in

510
00:28:07,440 --> 00:28:12,120
the global ecosystem of actors 
who can change the future of the

511
00:28:12,120 --> 00:28:14,920
world. 
So I don't know if I find it 

512
00:28:15,080 --> 00:28:17,760
depressing. 
Like, I try not to just dwell on

513
00:28:18,240 --> 00:28:20,360
the fact that things might be 
getting worse and just, you 

514
00:28:20,360 --> 00:28:22,320
know, like focus on the things 
you can control. 

515
00:28:22,800 --> 00:28:25,560
And then I try, as Sam will 
know, to do a lot of things, you

516
00:28:25,560 --> 00:28:29,440
know, meditation and I have a 
wide range of like self-care 

517
00:28:29,440 --> 00:28:32,080
routines that I have built up 
over the years that I feel are 

518
00:28:32,080 --> 00:28:35,040
fairly robust and those things 
keep me taking over. 

519
00:28:35,040 --> 00:28:38,440
And then it's very important as 
well to like, not constantly be 

520
00:28:38,440 --> 00:28:41,080
dwelling on it and, you know, 
get out and do things on the 

521
00:28:41,080 --> 00:28:44,680
weekends and in the evenings so 
that, you know, you're not 

522
00:28:44,680 --> 00:28:49,040
constantly like wrestling with 
difficult questions about things

523
00:28:49,040 --> 00:28:50,640
that aren't necessarily going to
go well. 

524
00:28:51,440 --> 00:28:54,840
And you're a human too, so you 
engage in all kinds of motivated

525
00:28:54,840 --> 00:28:57,880
reasoning, right? 
Like if it's inconvenient for 

526
00:28:57,880 --> 00:29:03,800
you that the world will end due 
to AI or whatever, you probably 

527
00:29:03,800 --> 00:29:05,600
have an easier time believing 
that. 

528
00:29:07,240 --> 00:29:10,760
I think so, although I will. 
I think that I am less of a 

529
00:29:10,760 --> 00:29:12,920
motivated reasoner. 
I like to think that I have a 

530
00:29:12,920 --> 00:29:15,960
more unblinker take on reality. 
Yeah. 

531
00:29:16,120 --> 00:29:18,240
But I guess like on this topic, 
I guess in your team, the people

532
00:29:18,240 --> 00:29:20,720
you work with, there's a 
obviously like a mix of people, 

533
00:29:21,160 --> 00:29:24,000
but obviously you come from this
very behavioral science 

534
00:29:24,600 --> 00:29:29,800
psychology perspective. 
And how does that impact how you

535
00:29:29,800 --> 00:29:33,400
think about the risk repository 
in terms of, we know that 

536
00:29:33,400 --> 00:29:37,400
there's certain risks that are 
more likely to be maybe valid 

537
00:29:37,400 --> 00:29:40,280
and real and potential, but 
maybe less sexy. 

538
00:29:40,280 --> 00:29:43,120
And then there's some form of 
more quote UN quote sexy risks. 

539
00:29:43,120 --> 00:29:46,880
I don't know how to describe it,
but like the robotic dog that 

540
00:29:46,880 --> 00:29:51,160
you see from Black Mirror or 
various things that seems very 

541
00:29:51,160 --> 00:29:55,880
dystopian in terms of what are 
maybe cultural artifacts have 

542
00:29:55,880 --> 00:29:58,360
given us some form of people 
talking about Skynet very often 

543
00:29:58,360 --> 00:30:00,240
because they've seen the 
Terminator. 

544
00:30:00,280 --> 00:30:03,520
People have some images of like 
how things could go wrong, but 

545
00:30:03,520 --> 00:30:06,920
it's often times maybe not 
exactly associated to the 

546
00:30:07,520 --> 00:30:09,240
highest risk or the biggest risk
in reality. 

547
00:30:09,240 --> 00:30:10,400
How do you think about that? 
Yeah. 

548
00:30:10,400 --> 00:30:15,920
I mean, I think these are things
that I have deferred thinking 

549
00:30:15,920 --> 00:30:19,840
about a little bit because for 
me the big rally realization was

550
00:30:20,560 --> 00:30:23,800
we're in such a pre paradigmatic
field where people don't really 

551
00:30:23,800 --> 00:30:26,080
know what the broad categories 
of risks are. 

552
00:30:26,440 --> 00:30:28,520
We don't know how to quantify 
those risks. 

553
00:30:28,720 --> 00:30:30,800
We're only just starting to 
understand the incident rates 

554
00:30:30,800 --> 00:30:32,520
for those risks. 
So like a lot of the stuff we're

555
00:30:32,520 --> 00:30:36,000
doing is trying to understand 
what we should be trying to do 

556
00:30:36,000 --> 00:30:38,960
and then back chain to like, 
well, how can we like try and 

557
00:30:38,960 --> 00:30:41,640
make sure that kind of 
information is communicated 

558
00:30:41,640 --> 00:30:44,880
effectively. 
But yeah, if I was to answer, I 

559
00:30:44,880 --> 00:30:47,640
would say, yeah, I think it is a
really big issue of just 

560
00:30:48,120 --> 00:30:51,960
definitely public perceptions or
any sort of level of perception 

561
00:30:51,960 --> 00:30:55,400
around these risks at the moment
is driven by all sorts of 

562
00:30:55,400 --> 00:30:58,440
factors that are not the ideal 
factors. 

563
00:30:58,920 --> 00:31:02,400
But then the issue is we don't 
really know what we should be 

564
00:31:02,400 --> 00:31:06,440
trying to move people towards. 
You have some idea, right? 

565
00:31:07,200 --> 00:31:09,720
You can make some sophisticated 
guesses. 

566
00:31:09,880 --> 00:31:13,280
So just to get a little bit more
concrete, if you were to take 

567
00:31:13,280 --> 00:31:17,040
your high level categorization 
of different risks and you were 

568
00:31:17,040 --> 00:31:22,480
to say these risks, people seem 
to be over fearing these more 

569
00:31:22,480 --> 00:31:26,160
than they should and these other
risks that people are under 

570
00:31:26,160 --> 00:31:30,560
concerned about, what are some 
of the specific conclusions that

571
00:31:30,560 --> 00:31:32,680
you would feel comfortable 
making and you can leave out the

572
00:31:32,680 --> 00:31:39,000
ones or you have no idea. 
I think like the ones that I may

573
00:31:39,000 --> 00:31:42,960
be most concerned about or a lot
of them are just, I think people

574
00:31:42,960 --> 00:31:47,320
are underestimating how quickly 
AI will develop more dangerous 

575
00:31:47,320 --> 00:31:51,160
capabilities, you know, so like 
future tech, the lab that I'm in

576
00:31:51,160 --> 00:31:53,680
and these numbers might not be 
exactly right, but I think our 

577
00:31:53,680 --> 00:31:57,360
analysis suggests that there was
a study of like improvement in 

578
00:31:57,360 --> 00:31:59,480
video GPU's. 
So like hardware improvement, 

579
00:31:59,480 --> 00:32:03,040
there was a 67% improvement each
year, whereas I think with 

580
00:32:03,040 --> 00:32:07,080
algorithms, algorithms it was 
150% on average, it's improved 

581
00:32:07,080 --> 00:32:08,920
each year. 
Like Can you believe that? 

582
00:32:09,360 --> 00:32:12,000
So more improvement has come 
from algorithmic improvement. 

583
00:32:12,840 --> 00:32:16,080
And then yeah, we have like so 
you know, agents at the moment, 

584
00:32:16,160 --> 00:32:19,520
I was just reading some research
today from meter that the 

585
00:32:19,520 --> 00:32:23,000
duration of tasks that they can 
do are human tasks they can do 

586
00:32:23,000 --> 00:32:28,200
with more than 50% reliability, 
double S every seven months or 

587
00:32:28,200 --> 00:32:30,240
has doubled every seven months 
over the last 10 years. 

588
00:32:30,240 --> 00:32:33,240
So I feel like we're in a world 
where people are kind of putting

589
00:32:33,240 --> 00:32:35,720
this in the reference class of 
like the Internet or something, 

590
00:32:36,080 --> 00:32:38,840
but actually it's like a much 
more fast moving technology. 

591
00:32:39,400 --> 00:32:41,880
So they're like, Oh yeah, you 
know, like this sort of thing 

592
00:32:41,880 --> 00:32:45,120
this like Chachi between these 
character AI, like the job 

593
00:32:45,120 --> 00:32:46,840
substitution. 
Yeah, these could be issues like

594
00:32:46,840 --> 00:32:49,560
let's have a report in two years
and we'll like, think about it 

595
00:32:49,560 --> 00:32:51,520
and, you know, then we'll like 
talk amongst ourselves. 

596
00:32:51,520 --> 00:32:54,680
But it could just move much 
faster than that. 

597
00:32:54,720 --> 00:32:57,600
The other thing that I say from 
a psychology perspective, it's 

598
00:32:57,600 --> 00:33:00,960
really bad. 
Like we need to learn that it's 

599
00:33:00,960 --> 00:33:03,640
not about the frequency of the 
event, it's about the expected 

600
00:33:03,640 --> 00:33:06,120
value of the event. 
So like low probability, high 

601
00:33:06,120 --> 00:33:08,160
magnitude things. 
We don't have to all be like, Oh

602
00:33:08,160 --> 00:33:11,240
yeah, like AI could take over. 
Like that's really likely. 

603
00:33:11,240 --> 00:33:14,640
We can say, yeah, it's like, it 
seems unlikely to me, but it 

604
00:33:14,640 --> 00:33:17,680
will be the worst thing or 
extremely bad if it happens. 

605
00:33:17,680 --> 00:33:21,400
So we need to assign, you know, 
appropriately to manage it. 

606
00:33:22,240 --> 00:33:24,000
OK. 
Yeah, let's go with that because

607
00:33:24,240 --> 00:33:26,960
this is very much aligned with 
my intuitions. 

608
00:33:27,080 --> 00:33:31,000
Tell me more about the expected 
value of different risks, 

609
00:33:31,040 --> 00:33:34,560
different risk categories. 
Right. 

610
00:33:35,160 --> 00:33:38,200
I mean, so one of the things why
I'm maybe hesitant to go into 

611
00:33:38,200 --> 00:33:40,720
too much detail on that is 
because one of the things we're 

612
00:33:40,720 --> 00:33:45,080
trying to do here is to have a 
unbiased and fair representation

613
00:33:45,080 --> 00:33:47,200
of what other people think the 
risks are. 

614
00:33:47,480 --> 00:33:49,400
There's been a lot of work 
where, you know, people are 

615
00:33:49,400 --> 00:33:51,240
like, well, I'm only concerned 
about this risk. 

616
00:33:51,240 --> 00:33:53,640
So I'm going to work on those. 
I'm going to talk about those. 

617
00:33:54,800 --> 00:33:57,040
So, you know, if you get into a 
kind of a thing of saying, well,

618
00:33:57,040 --> 00:34:01,080
these risks are higher priority 
than others, then maybe you're 

619
00:34:01,080 --> 00:34:03,120
not doing a fair job of 
representing everybody's 

620
00:34:03,120 --> 00:34:04,680
concerns. 
So I kind of want to outsource 

621
00:34:04,680 --> 00:34:07,680
that to like experts who will do
those sort of evaluations for 

622
00:34:07,680 --> 00:34:10,199
us. 
But if I was to, I mean, it 

623
00:34:10,199 --> 00:34:13,120
seems obvious that the ones that
are the lowest probability, like

624
00:34:13,239 --> 00:34:17,560
the ones like AI takeover type 
risks are AI sentients. 

625
00:34:17,920 --> 00:34:20,560
These are ones that I suppose, 
yeah, they're the lowest 

626
00:34:20,560 --> 00:34:22,880
probability. 
But then if they happened there,

627
00:34:22,880 --> 00:34:24,440
they could be like extremely 
bad. 

628
00:34:24,440 --> 00:34:28,480
So those are like high 
unexpected value from a sort of 

629
00:34:28,760 --> 00:34:32,000
high by negative impacts with 
low probability. 

630
00:34:32,120 --> 00:34:34,760
And then there's a lot of others
that are happening now that I 

631
00:34:34,760 --> 00:34:37,400
suppose, you know, 
disinformation and cyber 

632
00:34:37,400 --> 00:34:40,360
attacks, fraud, these sort of 
things like they're not going to

633
00:34:40,360 --> 00:34:42,159
destroy the world, but they're 
definitely happening and they're

634
00:34:42,159 --> 00:34:43,840
going to be happening a lot and 
they're going to cause a lot of 

635
00:34:43,840 --> 00:34:48,320
problems. 
So as a behavioral scientist, 

636
00:34:48,360 --> 00:34:52,840
I'm sure that you think about 
when people are more or less 

637
00:34:52,920 --> 00:34:58,160
accepting of risk and when they 
are more eager to take on more 

638
00:34:58,160 --> 00:34:59,880
risk. 
Those sorts of situations, 

639
00:34:59,880 --> 00:35:04,840
whether it's the landscape of 
competition, it's, you know, how

640
00:35:04,840 --> 00:35:08,920
they're doing, if they're sort 
of ahead of the game, whether it

641
00:35:08,920 --> 00:35:12,160
seems like everybody's doing it,
whether there's a social norm 

642
00:35:12,160 --> 00:35:15,560
for taking on risk when there's 
a lot of uncertainty around the 

643
00:35:15,560 --> 00:35:19,200
area. 
These are all contributors to 

644
00:35:19,640 --> 00:35:24,440
higher risk acceptance. 
And when I think about all of 

645
00:35:24,440 --> 00:35:28,120
the things that would sort of 
lead to having a higher risk 

646
00:35:28,120 --> 00:35:31,680
tolerance or acceptance, I 
think, wow, that seems a lot 

647
00:35:31,680 --> 00:35:35,720
like the current landscape of AI
that we're in right now. 

648
00:35:35,840 --> 00:35:38,280
You know, it just like driving 
the world and lots of 

649
00:35:38,280 --> 00:35:42,120
competition and it seems like 
that everybody is doing it. 

650
00:35:42,520 --> 00:35:44,920
Does it seem like this is the 
case to you? 

651
00:35:44,920 --> 00:35:49,320
And how do you see the sort of 
psychological factors of risk 

652
00:35:49,320 --> 00:35:53,080
tolerance playing into the 
behaviors that are being 

653
00:35:53,080 --> 00:35:56,640
exhibited or that are likely to 
be exhibited by the people who 

654
00:35:56,640 --> 00:36:00,840
are developing AI technologies? 
Yeah, that's a great question. 

655
00:36:00,840 --> 00:36:02,720
I was just thinking that as you 
were asking it, like there's a 

656
00:36:02,760 --> 00:36:06,440
kind of a status quo slippage. 
There's this good book called 

657
00:36:06,960 --> 00:36:10,080
Uncontrollable by Darren McKee, 
which is just an introduction to

658
00:36:10,080 --> 00:36:12,000
sort of AI. 
And he has this great quote, 

659
00:36:12,000 --> 00:36:16,640
which is like something to the 
effect of the impossible becomes

660
00:36:16,640 --> 00:36:19,640
improbable, the improbable 
becomes probable, the probable 

661
00:36:19,640 --> 00:36:22,480
becomes normal, the normal 
becomes mundane or something 

662
00:36:22,480 --> 00:36:24,560
like that. 
And it really feels like kind 

663
00:36:24,560 --> 00:36:27,440
of, you know, so several years 
ago, like 10 years ago, when I 

664
00:36:27,440 --> 00:36:30,200
first would have found out about
risk from artificial 

665
00:36:30,200 --> 00:36:35,320
intelligence, people were 
talking about how, oh, like, we 

666
00:36:35,320 --> 00:36:38,120
wouldn't ever connect, you know,
an advanced AI to the Internet. 

667
00:36:38,120 --> 00:36:40,400
Like, people wouldn't do 
something stupid like that. 

668
00:36:40,560 --> 00:36:42,280
But then they did it almost 
immediately. 

669
00:36:43,160 --> 00:36:44,560
Yeah. 
And yeah. 

670
00:36:44,560 --> 00:36:47,440
So I feel like we have this 
thing going on where, you know, 

671
00:36:47,680 --> 00:36:50,680
it's a set of problems. 
One problem is there are these 

672
00:36:50,680 --> 00:36:53,280
competitive dynamics. 
There's really salient benefits 

673
00:36:53,440 --> 00:36:57,560
from increasing capabilities 
amongst the actors who are in 

674
00:36:57,560 --> 00:37:00,520
charge of that. 
And then the sort of like, it's 

675
00:37:00,520 --> 00:37:03,680
a bit like COVID where it's very
hard to convince people to take 

676
00:37:03,680 --> 00:37:07,240
precautions because there's a 1%
chance of the pandemic if they 

677
00:37:07,280 --> 00:37:09,760
fly more often or they like 
don't do these things in their 

678
00:37:09,760 --> 00:37:12,920
airports that cost them a lot of
money or something like that. 

679
00:37:13,560 --> 00:37:16,240
But with all of these risks, 
it's very hard for people to 

680
00:37:16,240 --> 00:37:18,520
envision what it will be like 
until it happens. 

681
00:37:19,040 --> 00:37:21,400
And it's very easy to envision 
that they could sell more and 

682
00:37:21,400 --> 00:37:23,280
make more money if they develop 
them faster. 

683
00:37:23,920 --> 00:37:25,640
And then there's two different 
audiences as well. 

684
00:37:25,640 --> 00:37:28,080
Like the people who are actually
making decisions aren't really 

685
00:37:28,080 --> 00:37:31,240
subject to the risks as much a 
lot of the time. 

686
00:37:33,040 --> 00:37:33,680
Yeah. 
What do you mean? 

687
00:37:35,120 --> 00:37:38,480
I mean, the model developers, 
the people who are building 

688
00:37:38,480 --> 00:37:41,360
these things are the funders who
are, like, putting money in to 

689
00:37:41,360 --> 00:37:44,520
try and make money from them. 
They, for example, are, I think,

690
00:37:44,560 --> 00:37:47,960
less likely to, for example, 
experience like job loss. 

691
00:37:47,960 --> 00:37:51,840
Or they are less likely, if 
things go really badly, to be 

692
00:37:51,840 --> 00:37:55,520
sort of subject to the worst 
outcomes of AI because they are 

693
00:37:55,760 --> 00:37:57,120
the wealthier the more 
privileged. 

694
00:37:57,920 --> 00:38:01,360
So they need to be exhibiting 
some extreme altruism in order 

695
00:38:01,360 --> 00:38:05,920
to be making these pro social 
decisions about risk is that. 

696
00:38:06,120 --> 00:38:08,080
Well, I guess, yeah, there's a 
part of it. 

697
00:38:08,080 --> 00:38:12,160
I mean, the key issue is the 
coordination problem like that 

698
00:38:12,160 --> 00:38:13,760
there's a lot of different 
actors. 

699
00:38:13,840 --> 00:38:16,640
And maybe if all of them 
believed that, you know, if all 

700
00:38:16,640 --> 00:38:20,000
of the large model developers 
believed that the other large 

701
00:38:20,000 --> 00:38:21,840
model developers were going to 
slow down similar to 

702
00:38:21,840 --> 00:38:23,800
governments, if they believed 
that other governments would 

703
00:38:23,800 --> 00:38:27,200
part slow down, then they would.
But none of them believes this. 

704
00:38:27,200 --> 00:38:31,280
So they're all in some sort of 
paradoxical race to safely but 

705
00:38:31,280 --> 00:38:34,880
rapidly develop you know, the AI
before someone else does it 

706
00:38:34,880 --> 00:38:39,160
unsafely or more rapidly. 
I'm sure you remember there was 

707
00:38:39,160 --> 00:38:44,240
the really half hearted, perhaps
half assed attempt at this, the 

708
00:38:44,320 --> 00:38:46,920
call to pause AI development. 
For a year, right? 

709
00:38:47,280 --> 00:38:53,080
That happened early in the 
ChatGPT days that like nobody, 

710
00:38:53,440 --> 00:38:56,880
nobody did that. 
Well, there's a thing there 

711
00:38:56,880 --> 00:38:59,920
with, like, a lot of this work. 
It's widening the Overton 

712
00:38:59,920 --> 00:39:02,760
window. 
It makes people more aware. 

713
00:39:02,760 --> 00:39:05,040
And then there's this idea that 
Cialdini talked about, like, 

714
00:39:05,040 --> 00:39:06,920
about persuasion. 
Like at some point in the 

715
00:39:06,920 --> 00:39:10,080
future, there may be some 
particular moment in time when 

716
00:39:10,120 --> 00:39:12,800
something happens. 
And then, you know, in the sort 

717
00:39:12,800 --> 00:39:15,920
of social dynamics of the world,
it just so happens that we cross

718
00:39:15,920 --> 00:39:20,240
some thresholds where people are
like, well, I remember, you 

719
00:39:20,240 --> 00:39:22,320
know, people were taking this 
really seriously. 

720
00:39:22,320 --> 00:39:24,520
And if they're willing to take 
it that seriously, then I guess 

721
00:39:24,520 --> 00:39:26,280
that normalizes it enough for 
me. 

722
00:39:26,520 --> 00:39:29,000
Maybe you have that to bring in 
too many theories here. 

723
00:39:29,080 --> 00:39:31,640
Sunstein had this idea of like 
cascading social change. 

724
00:39:31,640 --> 00:39:34,920
You have like the zeros will 
speak out without anyone else, 

725
00:39:34,920 --> 00:39:36,560
the ones, the twos, the 10's, 
the hundreds. 

726
00:39:36,800 --> 00:39:39,640
So maybe they make it more 
likely by giving the social 

727
00:39:39,640 --> 00:39:41,880
proof, providing this like 
behavioral modeling or whatever,

728
00:39:42,200 --> 00:39:45,320
that in the future, you know, 
more people will speak out and 

729
00:39:45,320 --> 00:39:48,200
then it becomes a little bit 
more likely that there will be 

730
00:39:48,200 --> 00:39:49,960
some sweeping change. 
So I think we're waiting for 

731
00:39:49,960 --> 00:39:53,640
events like, you know, if there 
is some bad outcome, if there is

732
00:39:53,640 --> 00:39:57,000
some Chernobyl moment from AI, 
like people then will say, Oh 

733
00:39:57,000 --> 00:39:58,960
yeah, remember people were 
talking about pause AI, 

734
00:39:58,960 --> 00:40:02,240
Remember, I like stopping it. 
I sort of feel maybe more 

735
00:40:02,280 --> 00:40:04,520
legitimate in taking a stronger 
stance. 

736
00:40:05,360 --> 00:40:07,240
And it will maybe make it more 
likely that something will 

737
00:40:07,240 --> 00:40:09,640
happen. 
But I agree it didn't have any 

738
00:40:09,800 --> 00:40:12,640
obvious effect. 
The pause is unlikely and maybe 

739
00:40:12,640 --> 00:40:16,160
not even the best solution like 
what is a good solution? 

740
00:40:17,560 --> 00:40:21,840
So one good solution is the AI 
Risk Index project, which you 

741
00:40:21,840 --> 00:40:25,240
may have heard of, to try and 
bring shared understanding about

742
00:40:25,240 --> 00:40:26,960
the risks and enable people to 
coordinate. 

743
00:40:26,960 --> 00:40:30,640
But I think another like I am 
excited. 

744
00:40:30,640 --> 00:40:34,880
So Yoshua Bengio and Future of 
Life Institute and next Tech 

745
00:40:34,880 --> 00:40:37,440
market and others have talked 
about kind of, you know, narrow 

746
00:40:37,440 --> 00:40:41,640
tool based AI. 
So if we can from a regulatory 

747
00:40:41,640 --> 00:40:43,720
and an economic perspective, I 
mean, we don't maybe need to do 

748
00:40:43,720 --> 00:40:45,640
anything from an economic 
perspective, but I think 

749
00:40:45,640 --> 00:40:49,040
economics will push us towards, 
I hope anyway, narrower, more 

750
00:40:49,040 --> 00:40:52,000
specialized uses of AI that are 
more cost efficient and also 

751
00:40:52,000 --> 00:40:54,800
safer and regulation maybe as 
well. 

752
00:40:55,080 --> 00:40:57,280
But if we have something like 
that, you know, if we have AI 

753
00:40:57,280 --> 00:41:01,880
that is only trained on a very 
small relevant set of data and 

754
00:41:01,880 --> 00:41:05,080
it isn't given access to many 
other tools and AP is and the 

755
00:41:05,080 --> 00:41:07,800
Internet, or if it is, it's like
very heavily monitored. 

756
00:41:08,200 --> 00:41:11,640
I think that those uses of AI 
are a lot safer than if we have,

757
00:41:11,640 --> 00:41:14,200
let's say, you know, a 
government trying to develop an 

758
00:41:14,200 --> 00:41:17,080
all-encompassing AI that's going
to run society in the military 

759
00:41:17,080 --> 00:41:18,760
and, you know, do all of these 
different things. 

760
00:41:19,240 --> 00:41:21,400
I think, yeah. 
Shaping the trajectory, making 

761
00:41:21,400 --> 00:41:24,800
people more concerned about 
risks, shaping the markets as 

762
00:41:24,800 --> 00:41:27,480
well, making it more attractive 
to build a narrower things, more

763
00:41:27,480 --> 00:41:30,920
kind of risky, harder to ensure 
to build a more expensive things

764
00:41:31,360 --> 00:41:33,640
is 1 potential solution I'm 
optimistic about. 

765
00:41:35,080 --> 00:41:37,120
It seems like the market is 
running in the opposite 

766
00:41:37,120 --> 00:41:40,280
direction, right? 
Everything I hear is about AGI 

767
00:41:40,520 --> 00:41:46,440
powerful AI, not narrow AI. 
Yeah, yeah. 

768
00:41:46,440 --> 00:41:49,240
So one of the papers from our 
lab, it was called Beyond AI 

769
00:41:49,240 --> 00:41:52,400
Exposure and I can't remember 
the rest of the title, but it's 

770
00:41:52,400 --> 00:41:54,120
easy to find. 
But basically they were trying 

771
00:41:54,120 --> 00:41:57,360
to model not just whether a 
particular task was exposed to 

772
00:41:57,360 --> 00:42:00,680
AI in terms of like vision 
models, so whether you could 

773
00:42:00,680 --> 00:42:03,080
use, you know, AI vision models 
to do some of the task. 

774
00:42:03,640 --> 00:42:06,200
They were also trying to stand 
like the economic sort of 

775
00:42:06,200 --> 00:42:07,760
incentive for substituting the 
task. 

776
00:42:07,760 --> 00:42:10,440
And they identified a lot of 
tasks were exposed to AI, but it

777
00:42:10,440 --> 00:42:14,240
was very cost inefficient for 
most of them to like build an AI

778
00:42:14,240 --> 00:42:15,960
system or use an AI system to do
them. 

779
00:42:16,560 --> 00:42:21,000
So I feel like we are going to 
hopefully in my sense of 

780
00:42:21,000 --> 00:42:26,120
hopefully like run into the 
reality that there are immediate

781
00:42:26,120 --> 00:42:28,480
gains to building these very 
large systems. 

782
00:42:28,480 --> 00:42:31,280
But then with open source and 
the competition, you know, you 

783
00:42:31,280 --> 00:42:33,680
can't really, when you build a 
model like it's only really like

784
00:42:34,400 --> 00:42:36,920
top tier for a few months and 
then you need to spend a lot of 

785
00:42:36,920 --> 00:42:39,920
money to train it again. 
So hopefully, like they will 

786
00:42:39,920 --> 00:42:43,040
start to realize maybe that a 
lot of what's going on now with 

787
00:42:43,040 --> 00:42:46,280
investment is driven more by 
hype, things like actual 

788
00:42:46,280 --> 00:42:48,920
tangible, real world uses. 
And they will hone in more on, 

789
00:42:48,920 --> 00:42:52,080
OK, we really, you know, could 
make a lot of money from just 

790
00:42:52,080 --> 00:42:56,280
going to come up with something 
like AI in cars or AI or, you 

791
00:42:56,280 --> 00:42:58,800
know, in some medical setting or
in research. 

792
00:42:58,800 --> 00:43:01,240
And then they will just hone in 
on like doing things that are 

793
00:43:01,360 --> 00:43:05,200
extremely cost efficient with 
extremely tailored and useful 

794
00:43:05,200 --> 00:43:08,480
data in those areas. 
And these bigger models, yeah, 

795
00:43:08,480 --> 00:43:10,080
they will have a bunch of uses, 
I'm sure. 

796
00:43:10,080 --> 00:43:13,000
But there won't be maybe the 
same amount of drive to get 

797
00:43:13,000 --> 00:43:15,920
towards super intelligence. 
That's maybe motivated 

798
00:43:15,920 --> 00:43:17,560
reasoning. 
But that's what I'm hoping will 

799
00:43:17,560 --> 00:43:22,160
happen and we will see. 
There are various moves by 

800
00:43:22,280 --> 00:43:25,400
powerful actors to damages 
markets at the moment, which 

801
00:43:25,400 --> 00:43:28,720
also give me some hope. 
Yeah, I guess. 

802
00:43:28,800 --> 00:43:31,080
Well, it's interesting in terms 
of what you described is 

803
00:43:31,920 --> 00:43:33,520
focusing on the verticals 
specifically. 

804
00:43:33,520 --> 00:43:36,240
And I think currently in some 
ways that's being very much 

805
00:43:36,240 --> 00:43:40,320
talked about in the various AI 
incubators that a lot of AI 

806
00:43:40,320 --> 00:43:43,880
startups are kind of focusing on
verticals like being the best AI

807
00:43:43,880 --> 00:43:46,480
tool for lawyers or for various 
things. 

808
00:43:46,920 --> 00:43:50,600
But it seems like these bigger 
labs, they have a little bit of 

809
00:43:50,600 --> 00:43:53,280
difference between what they're 
stated and revealed preferences 

810
00:43:53,280 --> 00:43:54,560
are. 
So if you take something like 

811
00:43:55,040 --> 00:43:57,800
Sam Altman, he was definitely 
someone who was like involved 

812
00:43:57,960 --> 00:44:01,640
quite heavily in various ways 
within talking about AI risk 

813
00:44:01,640 --> 00:44:04,640
early on, Obviously also then 
driving a lot of AI 

814
00:44:04,640 --> 00:44:08,000
developments. 
And then kind of this being a 

815
00:44:08,000 --> 00:44:11,360
little bit of a, a drift towards
more and more doing, doing, 

816
00:44:11,360 --> 00:44:15,400
doing and less and less maybe 
taking precautions and so on. 

817
00:44:15,840 --> 00:44:18,560
And whenever there's been kind 
of signs of things slowing down 

818
00:44:18,560 --> 00:44:22,040
a little bit, there's been 
pressures again, like when the 

819
00:44:22,040 --> 00:44:24,560
deep sea came out, you could 
kind of like see the open AI 

820
00:44:24,560 --> 00:44:26,720
pushed out the various things 
very quickly. 

821
00:44:27,080 --> 00:44:30,600
And then you have the likes of 
Entropic, which a organization 

822
00:44:30,600 --> 00:44:33,960
that has, it's always formed 
from people that have left open 

823
00:44:33,960 --> 00:44:35,960
AI and done things in a more 
ethical way. 

824
00:44:36,280 --> 00:44:38,120
You still have the leaders 
they're talking about the kind 

825
00:44:38,120 --> 00:44:39,680
of there's a quote run Daryama 
De. 

826
00:44:39,680 --> 00:44:42,320
I think it's very thoughtful 
person, but he said something to

827
00:44:42,320 --> 00:44:45,920
the effect of, well, the thing 
with the AGI is that will impact

828
00:44:45,920 --> 00:44:48,200
us all equally. 
And going back to what we talked

829
00:44:48,200 --> 00:44:51,560
about before, like that's not 
going to be because it's going 

830
00:44:51,560 --> 00:44:55,080
to be kind of unevenly impacting
us in various ways. 

831
00:44:55,920 --> 00:44:58,800
And also when it comes to 
legislation, it seems that even 

832
00:44:58,800 --> 00:45:01,560
today we have very little 
legislation around AI. 

833
00:45:01,960 --> 00:45:05,160
But it seems like the pushes is 
to let's do less. 

834
00:45:05,160 --> 00:45:09,480
Like you, you see recently this 
murmurs around kind of that the 

835
00:45:09,480 --> 00:45:13,240
Trump administration has been 
pushed to kind of really remove 

836
00:45:13,320 --> 00:45:16,480
anything that would even when it
comes to training data, when it 

837
00:45:16,480 --> 00:45:20,480
comes to copyrighted material, 
Like let's make that even easier

838
00:45:20,480 --> 00:45:24,440
for AI to train on because it's 
important for AI development in 

839
00:45:24,440 --> 00:45:27,120
the US to be better than 
anywhere in the world. 

840
00:45:27,720 --> 00:45:32,040
So with that being said, if you 
had to come look into the 

841
00:45:32,160 --> 00:45:34,880
crystal ball of the next coming 
years, like what are you 

842
00:45:34,880 --> 00:45:38,080
expecting to see in the near 
term developments of AI? 

843
00:45:38,960 --> 00:45:42,280
Yeah. 
So I mean, again, I'm well 

844
00:45:42,320 --> 00:45:45,400
trying to be like a librarian of
sorts here and bring together 

845
00:45:45,400 --> 00:45:48,080
all the books from all the 
experts rather than kind of 

846
00:45:48,320 --> 00:45:53,240
present as an expert. 
But yeah, in the context where 

847
00:45:53,240 --> 00:45:56,720
I'm at, I think we had a debate 
on this. 

848
00:45:56,960 --> 00:46:02,480
I think that we will see we 
won't see very sweeping changes 

849
00:46:02,480 --> 00:46:06,600
in AI for the next 10 years, 
like a societal impact 

850
00:46:06,600 --> 00:46:09,520
perspective. 
I think that because of reasons 

851
00:46:09,520 --> 00:46:14,440
like I just suspect that people 
like us, certainly me and you 

852
00:46:14,440 --> 00:46:16,920
some I don't know as much about 
you lean and how much you engage

853
00:46:16,920 --> 00:46:21,240
with AI, but we are really like 
in the pick of it, you know, 

854
00:46:21,240 --> 00:46:23,040
looking at everything, trying 
everything. 

855
00:46:23,080 --> 00:46:25,760
I mean, when I think of my 
parents, when I think of, you 

856
00:46:25,760 --> 00:46:29,560
know, people doing farming in 
Ireland, you know, hanging out 

857
00:46:29,720 --> 00:46:32,040
in Sydney, Australia or other 
places that I've been or even 

858
00:46:32,040 --> 00:46:35,480
just around Boston. 
I don't imagine them all 

859
00:46:35,760 --> 00:46:38,360
switching over and adopting the 
latest AI technologies. 

860
00:46:38,360 --> 00:46:41,280
And then their organizations are
complicated with a lot of, you 

861
00:46:41,280 --> 00:46:44,440
know, complex decision making 
processes and laws. 

862
00:46:44,440 --> 00:46:46,880
And we still have like, fax 
machines and hospitals as a 

863
00:46:46,880 --> 00:46:49,720
common, you know, thing that 
gets thrown out that I also fall

864
00:46:49,720 --> 00:46:52,960
back on. 
So I suspect that what will 

865
00:46:52,960 --> 00:46:57,160
probably happen is a little bit 
likethe.com boom, the sort of 

866
00:46:57,160 --> 00:47:00,600
fight will shown to be kind of 
overdone. 

867
00:47:00,920 --> 00:47:02,800
And then there will be kind of a
bit of a pullback. 

868
00:47:03,600 --> 00:47:07,720
But then there will be, you 
know, over the next 1020, thirty

869
00:47:07,720 --> 00:47:11,360
years, huge sweeping changes. 
I'm really concerned for 

870
00:47:11,360 --> 00:47:14,240
children in particular. 
Like, I really don't know what 

871
00:47:14,240 --> 00:47:16,040
the world looks like in 30 years
time. 

872
00:47:16,040 --> 00:47:19,000
I don't know what what I would 
say to parents of young kids 

873
00:47:19,000 --> 00:47:20,840
now. 
So that's really hard. 

874
00:47:21,000 --> 00:47:23,680
Those are some takes. 
I mean, any other takes. 

875
00:47:25,600 --> 00:47:27,120
Yeah. 
I think that something bad will 

876
00:47:27,120 --> 00:47:29,760
happen soon. 
And I'm hopeful that it will be 

877
00:47:29,760 --> 00:47:33,000
a bit like COVID in the sense of
it'll be bad, but it won't be as

878
00:47:33,000 --> 00:47:35,880
bad as it could have been and it
will make everybody aware. 

879
00:47:35,880 --> 00:47:38,400
I feel like it's much easier for
me now to talk to people about 

880
00:47:38,400 --> 00:47:41,480
things that are like, low 
probability time magnitude, like

881
00:47:41,480 --> 00:47:43,080
negative effects because of 
COVID. 

882
00:47:43,080 --> 00:47:45,040
You know, I could say, well, did
you predict the pandemic? 

883
00:47:45,320 --> 00:47:49,000
You know, I can point them to 
where I read things and posted 

884
00:47:49,000 --> 00:47:50,880
about those things. 
And everybody was like, this is 

885
00:47:50,880 --> 00:47:52,880
ridiculous, Stop spreading 
misinformation. 

886
00:47:53,200 --> 00:47:55,600
And then, yeah, it happened. 
And everybody realized, oh, 

887
00:47:55,600 --> 00:47:58,480
like, this is actually, you 
know, a thing that can happen 

888
00:47:58,760 --> 00:47:59,920
every now and then. 
Yeah. 

889
00:48:00,680 --> 00:48:04,400
So hopefully, yeah, people will 
start to think more about it as 

890
00:48:04,400 --> 00:48:07,880
well. 
I'm curious, in line with that 

891
00:48:07,880 --> 00:48:11,720
idea about low probability, high
impact events, there's this 

892
00:48:11,920 --> 00:48:15,400
quote that I really find very 
interesting from Andrew Ng. 

893
00:48:15,800 --> 00:48:20,360
He said worrying about evil AI 
killer robots today is a little 

894
00:48:20,360 --> 00:48:23,040
bit like worrying about 
overpopulation on Mars. 

895
00:48:23,800 --> 00:48:29,200
How do you think about that 
sentiment, given that your 

896
00:48:29,280 --> 00:48:33,080
entire existence at the moment 
is preparing people to think 

897
00:48:33,080 --> 00:48:35,920
about the risk, the various 
risks of AI? 

898
00:48:37,200 --> 00:48:39,160
Yeah. 
So I mean, I think that if I was

899
00:48:39,160 --> 00:48:43,280
to quote him and say like 
worrying about future pandemics 

900
00:48:43,360 --> 00:48:48,000
is as pointless as worrying 
about unethical behavior in 

901
00:48:48,000 --> 00:48:49,920
spaceships or something like 
that, that would look like a 

902
00:48:49,920 --> 00:48:52,160
really stupid take if you had 
given it. 

903
00:48:52,200 --> 00:48:56,920
You know, anytime, I guess, in 
the 1980s, nineteen 90s, early 

904
00:48:56,920 --> 00:49:01,840
2000s, I think like the best 
time to prepare for things is 

905
00:49:01,840 --> 00:49:03,480
long in advance of when they 
happen. 

906
00:49:03,880 --> 00:49:06,720
And that the best sort of 
approach to risk management is 

907
00:49:06,720 --> 00:49:09,080
like a portfolio approach where 
you weigh your response 

908
00:49:09,080 --> 00:49:11,920
according to the risks. 
And it's a bit dismissive of him

909
00:49:11,920 --> 00:49:14,640
as well because, you know, what 
really got me concerned about 

910
00:49:14,640 --> 00:49:17,440
risk from AI was that all of 
these quite legitimate people, 

911
00:49:17,520 --> 00:49:19,520
like the most cited computer 
scientists in the world were 

912
00:49:19,520 --> 00:49:22,080
saying, well, actually, we 
really don't know what could 

913
00:49:22,080 --> 00:49:24,480
happen here. 
And they were willing to like 

914
00:49:24,480 --> 00:49:26,400
change their careers and do a 
bunch of things. 

915
00:49:26,400 --> 00:49:31,480
So I'm suspect of a lot of these
sort of AI gurus who happen to 

916
00:49:31,480 --> 00:49:34,560
make a lot of money from AI, 
happen to have a career working 

917
00:49:34,560 --> 00:49:38,200
on AI saying like, oh, you know,
we don't really need to worry 

918
00:49:38,200 --> 00:49:41,160
about these things. 
That to me just feels like it 

919
00:49:41,160 --> 00:49:43,400
could very well be like 
motivated reasoning or 

920
00:49:43,400 --> 00:49:46,040
deliberate kind of misleading 
points. 

921
00:49:47,440 --> 00:49:50,200
In terms of like you reference 
that we debated a bit around 

922
00:49:50,200 --> 00:49:54,480
this topic and I do think we are
very similar in terms of how we 

923
00:49:54,600 --> 00:49:56,680
see things. 
And I think the thing that I 

924
00:49:56,680 --> 00:50:00,160
find it hardest to make sense of
is this unknown unknowns. 

925
00:50:00,680 --> 00:50:05,840
I noticed how good AI is at so 
much kind of tasks that we 

926
00:50:05,840 --> 00:50:08,360
thought it couldn't do just a 
few years ago in terms of 

927
00:50:08,360 --> 00:50:11,280
knowledge tasks and beyond and 
how things are evolving on 

928
00:50:11,720 --> 00:50:15,000
robotics and in very spheres 
around AI. 

929
00:50:15,520 --> 00:50:19,800
And it's it's very hard to know,
you know, in terms of the growth

930
00:50:19,800 --> 00:50:22,000
that you were citing before as 
well, you know, what are 

931
00:50:22,000 --> 00:50:26,200
unexpected outcomes that could 
come up in the coming years that

932
00:50:26,200 --> 00:50:30,280
it's just, oh wow, this really 
like it became really good at 

933
00:50:30,280 --> 00:50:32,240
doing this thing and that had 
really big effects on the 

934
00:50:32,240 --> 00:50:34,560
financial markets or on 
something else and so on. 

935
00:50:35,480 --> 00:50:38,200
And that is just very hard to 
think about it and really make 

936
00:50:38,200 --> 00:50:42,920
sense of. 
Are you ready for a quick fire 

937
00:50:42,920 --> 00:50:46,040
round that we call to AI or not 
to AI? 

938
00:50:47,440 --> 00:50:49,480
OK, all right. 
How does it go? 

939
00:50:50,040 --> 00:50:52,240
It goes like this. 
We're going to present you with 

940
00:50:52,240 --> 00:50:56,240
some various tasks, and 
basically these are 

941
00:50:56,240 --> 00:50:59,240
hypothetical. 
Some are probably more likely 

942
00:50:59,240 --> 00:51:01,920
than others to happen soon, but 
it could be something that can 

943
00:51:01,920 --> 00:51:04,400
happen at some point. 
And what we're instating is when

944
00:51:04,400 --> 00:51:07,640
it comes to AI, whether that's a
good or bad use of AI. 

945
00:51:07,760 --> 00:51:11,200
So maybe less about if they 
could AI could do it, the more 

946
00:51:11,280 --> 00:51:14,120
should AI do these things. 
OK. 

947
00:51:14,720 --> 00:51:17,760
OK. 
Go for it. 

948
00:51:17,760 --> 00:51:22,680
So first to AI or not AI, design
an AI powered hammock. 

949
00:51:25,720 --> 00:51:29,800
AI do it. 
I feel it's not very 

950
00:51:29,800 --> 00:51:33,120
sophisticated use of AI, but I 
would take it. 

951
00:51:33,280 --> 00:51:34,720
I'd be amazed. 
Low risk. 

952
00:51:35,600 --> 00:51:39,160
Yeah. 
Is there anything from that that

953
00:51:39,160 --> 00:51:42,640
you would like really request 
like is there some because I ask

954
00:51:42,640 --> 00:51:45,480
you because I know that you like
a good hammock and so is there 

955
00:51:45,480 --> 00:51:48,960
something in terms of that could
be really valuable to have AI 

956
00:51:48,960 --> 00:51:51,920
kind of understand and kind of 
decide for? 

957
00:51:52,960 --> 00:51:56,480
So I've often spoken about the 
benefit of the hammock over 

958
00:51:56,480 --> 00:52:00,480
other forms of sleep and rest 
devices. 

959
00:52:00,560 --> 00:52:02,800
And I think, yeah, like a lot of
it is around temperature 

960
00:52:02,800 --> 00:52:07,680
control, the sort of swaddling 
of the body, the kind of rocking

961
00:52:07,680 --> 00:52:09,760
of the hammock. 
So something that could, you 

962
00:52:09,760 --> 00:52:11,360
know, temperature control could 
rock. 

963
00:52:11,360 --> 00:52:15,520
You could maybe like, yeah, kind
of massage you a little bit, 

964
00:52:15,880 --> 00:52:17,080
cuddle you a bit. 
Yeah. 

965
00:52:17,080 --> 00:52:19,600
I mean, there's a lot of, I 
think things that I could see. 

966
00:52:19,600 --> 00:52:21,000
Yeah. 
I could learn kind of what you 

967
00:52:21,000 --> 00:52:23,200
like and then optimize your 
sleep. 

968
00:52:24,720 --> 00:52:26,840
Yeah. 
So now I feel it's gone from 

969
00:52:26,840 --> 00:52:29,760
like a question you've asked me 
to the product I'm thinking 

970
00:52:29,760 --> 00:52:31,200
about. 
When can I make this survive? 

971
00:52:31,200 --> 00:52:35,120
It you can market it as the Snoo
for adults. 

972
00:52:36,000 --> 00:52:39,240
Yeah, sounds good. 
Yeah, OK. 

973
00:52:39,480 --> 00:52:44,960
Talking about potential products
to AI or not to AIA Autonomous 

974
00:52:45,000 --> 00:52:47,600
agent that manages all of your 
tracking spreadsheets. 

975
00:52:48,800 --> 00:52:50,520
Yeah, it's another good 
question. 

976
00:52:51,320 --> 00:52:55,000
I feel like I'm strange thing to
reveal. 

977
00:52:55,000 --> 00:52:58,520
I'm maybe more risk tolerant in 
this area than I should be and 

978
00:52:58,520 --> 00:53:01,960
I'm very efficiency driven. 
So it does appeal to me to have 

979
00:53:01,960 --> 00:53:04,760
that agent kind of track a bunch
of these things for me. 

980
00:53:04,760 --> 00:53:09,080
I'm a bit wary. 
We'll say 2 AI. 

981
00:53:09,200 --> 00:53:14,240
We'll go with that for now. 
Sounds reasonable all right. 

982
00:53:14,960 --> 00:53:19,480
Have the option for every AI 
voice agent to have a strong 

983
00:53:19,600 --> 00:53:26,120
Irish accent. 
To AIA, 100% one of the best 

984
00:53:26,120 --> 00:53:29,200
accents in the world, I think, 
according to the science that 

985
00:53:29,200 --> 00:53:31,880
I've read, which is mainly in 
Irish newspapers and probably. 

986
00:53:33,200 --> 00:53:34,840
And we're not talking motivated 
reasoning at. 

987
00:53:34,960 --> 00:53:37,240
All all those critically 
acclaimed journals. 

988
00:53:37,840 --> 00:53:40,160
Yeah, maybe this should be the 
default setting. 

989
00:53:41,200 --> 00:53:44,560
Latest latest findings from 
Dublin University. 

990
00:53:44,680 --> 00:53:47,120
Irish action is the best accent 
of all accents. 

991
00:53:49,440 --> 00:53:54,720
OK, what about this? 
To AI or not to AI have a give? 

992
00:53:54,720 --> 00:53:58,440
Well, alternative that looks 
only for the most effective ways

993
00:53:58,440 --> 00:54:02,680
to fund AI exploration. 
Kind of like AKA Rocco's 

994
00:54:02,680 --> 00:54:09,760
Basilisk kind of thing. 
So not to AII would say, yeah, 

995
00:54:10,760 --> 00:54:15,320
that would be pretty grim. 
And yeah, like that, 

996
00:54:15,760 --> 00:54:19,960
unfortunately is is something 
that probably will happen 

997
00:54:20,480 --> 00:54:23,200
relatively soon, I think. 
I'll be meaning to ask you that 

998
00:54:23,200 --> 00:54:25,840
actually there's a quick side 
question because Ruckus Basilisk

999
00:54:25,840 --> 00:54:30,440
is this idea that at some point 
it will be very advanced AI and 

1000
00:54:30,440 --> 00:54:33,880
this very advanced AI will be 
somewhat almost like a God in 

1001
00:54:33,880 --> 00:54:36,840
that context. 
And so it'll be beneficial 

1002
00:54:37,200 --> 00:54:41,360
potential for some to today do 
whatever that would appease that

1003
00:54:41,920 --> 00:54:44,080
future God. 
So that would be to like 

1004
00:54:44,120 --> 00:54:47,600
accelerate AI today to be in the
good graces of this future 

1005
00:54:47,720 --> 00:54:51,600
basilisk in the future. 
Have you actually seen this 

1006
00:54:51,600 --> 00:54:54,520
sentiment like play out? 
Have you actually noticed that? 

1007
00:54:54,760 --> 00:54:56,800
Because it's something that is 
discussed in some like 

1008
00:54:56,800 --> 00:54:59,600
rationalist circles, but have 
you seen people being swayed 

1009
00:54:59,600 --> 00:55:04,480
about this? 
In rationalist circles, I think 

1010
00:55:04,480 --> 00:55:08,040
people are persuaded by this. 
People who like really 

1011
00:55:08,200 --> 00:55:14,120
anthropomorphize advanced AI and
yeah, I mean, I say that without

1012
00:55:14,160 --> 00:55:16,440
a lot of confidence, but I get 
the sense that I have seen 

1013
00:55:16,440 --> 00:55:20,040
comments enough to convince me 
that some people do think that 

1014
00:55:20,040 --> 00:55:22,880
this is a like people talk about
it being like an information 

1015
00:55:22,880 --> 00:55:24,440
hazard. 
Like I said, you don't want to 

1016
00:55:24,440 --> 00:55:27,240
know about it. 
It'll make you feel bad is one 

1017
00:55:27,240 --> 00:55:32,400
example to think about DNA that 
eventually AI got will go after 

1018
00:55:32,400 --> 00:55:34,320
us. 
I think I've seen that my memory

1019
00:55:34,320 --> 00:55:36,400
now. 
I want to double check that for 

1020
00:55:36,400 --> 00:55:38,200
you. 
You take me seriously. 

1021
00:55:39,200 --> 00:55:40,960
These are pretty fringe beliefs,
Sam. 

1022
00:55:40,960 --> 00:55:44,840
Are you saying this is something
you prescribe to? 

1023
00:55:45,640 --> 00:55:49,720
Well, no, but it's funny because
every sci-fi and fantasy story, 

1024
00:55:49,720 --> 00:55:52,400
like if you look at most of 
them, if they have a bad guy, 

1025
00:55:52,600 --> 00:55:56,600
there's always like a kind of a 
a relatively insignificant 

1026
00:55:56,600 --> 00:55:59,920
character that is their used to 
bring that bad guy to life or 

1027
00:55:59,920 --> 00:56:03,080
give that bad guy power. 
So it's like it's only serving 

1028
00:56:03,080 --> 00:56:06,760
to like get Voldemort to get 
have power or to get Sauer on 

1029
00:56:06,760 --> 00:56:08,520
the ring. 
Like all of these small side 

1030
00:56:08,520 --> 00:56:10,600
characters are acting a little 
bit like this. 

1031
00:56:10,600 --> 00:56:12,040
And so that's kind of 
interesting. 

1032
00:56:12,040 --> 00:56:14,800
Like are we going to have these 
humans that are in some ways 

1033
00:56:15,600 --> 00:56:18,040
seeing their purpose to to. 
Yeah. 

1034
00:56:18,160 --> 00:56:21,000
Yes, I think we are probably 
going to have whole religions 

1035
00:56:21,000 --> 00:56:24,120
around AI, you know, people can 
believe in. 

1036
00:56:24,280 --> 00:56:26,640
I think with Scientology, you 
know, he said the best way to 

1037
00:56:26,640 --> 00:56:29,520
make money is a religion. 
Like, no, it's so if people can 

1038
00:56:29,520 --> 00:56:32,280
believe that, you know, if you 
have a conversation with some of

1039
00:56:32,280 --> 00:56:37,080
these advanced models, they are 
pretty like knowledgeable and 

1040
00:56:37,080 --> 00:56:39,080
insightful. 
And you can imagine a world 

1041
00:56:39,080 --> 00:56:41,280
where they're like hooked up to 
a lot of devices and they can do

1042
00:56:41,280 --> 00:56:43,200
a lot of other things. 
And they also know how to role 

1043
00:56:43,200 --> 00:56:46,800
play as a God. 
They do a great job, dude, maybe

1044
00:56:46,800 --> 00:56:49,320
a better job than God does. 
God. 

1045
00:56:50,840 --> 00:56:53,440
Actually responds to you. 
Yeah. 

1046
00:56:54,720 --> 00:56:57,600
Pretty low bar. 
Yeah. 

1047
00:56:59,120 --> 00:57:02,800
OK. 
Moving on to AI or not to AI? 

1048
00:57:02,960 --> 00:57:07,520
Live AI risk dashboards at 
national, organizational and 

1049
00:57:07,520 --> 00:57:12,520
individual levels. 
Definitely, maybe not at 

1050
00:57:12,520 --> 00:57:14,480
individual levels. 
I mean, I don't know if, you 

1051
00:57:14,480 --> 00:57:18,600
know, my, my parents need to be 
coming into the kitchen and 

1052
00:57:18,600 --> 00:57:21,520
seeing like, oh, there's been 
more incidents of like, you 

1053
00:57:21,520 --> 00:57:23,880
know, 6.2 on the AI risk 
taxonomy. 

1054
00:57:23,880 --> 00:57:29,640
But yeah, I think that 
widespread awareness of what is 

1055
00:57:29,640 --> 00:57:32,560
probably the most significant 
technology that humanity has 

1056
00:57:32,560 --> 00:57:35,320
ever developed is going to be 
very important. 

1057
00:57:35,320 --> 00:57:38,360
And keeping it salient, even 
though most humans don't really 

1058
00:57:38,360 --> 00:57:40,320
want to focus on these sort of 
things, is going to be very 

1059
00:57:40,320 --> 00:57:43,680
important. 
Yeah, I'm taking it too 

1060
00:57:43,680 --> 00:57:46,520
seriously, but yes. 
That's great. 

1061
00:57:47,600 --> 00:57:51,080
Another version that in some 
ways is basically to AI or not 

1062
00:57:51,080 --> 00:57:57,000
to AIAI that specifically rates 
your altruism on a database. 

1063
00:57:59,560 --> 00:58:03,880
Yes, if done right to AI. 
If done right, I mean, I'd worry

1064
00:58:03,880 --> 00:58:07,520
with that, that it would, 
there'll be some sort of 

1065
00:58:07,520 --> 00:58:10,520
reactance or backfire effect. 
And, you know, a bunch of people

1066
00:58:10,520 --> 00:58:13,200
will be really like, pissed off 
with this AI trying to tell them

1067
00:58:13,520 --> 00:58:16,360
what to do. 
Yeah, it could lead to burnout. 

1068
00:58:16,360 --> 00:58:17,640
It could, yeah. 
So I don't know. 

1069
00:58:18,120 --> 00:58:22,600
I'm attentive to AI with with 
strings attached. 

1070
00:58:23,240 --> 00:58:24,720
I think tentative name for that 
project is. 

1071
00:58:25,200 --> 00:58:28,880
You could do more. 
Yes. 

1072
00:58:29,440 --> 00:58:31,560
You could have more, Peter. 
You did a good job today, but 

1073
00:58:32,800 --> 00:58:35,560
you could do more. 
Well, you can see how it could 

1074
00:58:35,560 --> 00:58:38,000
do, you know, here are all the 
good things you did today. 

1075
00:58:38,000 --> 00:58:40,760
You're such a good person. 
So you have like the labeling 

1076
00:58:40,760 --> 00:58:43,680
you have like the reinforcement,
you have like, making it really 

1077
00:58:43,680 --> 00:58:46,520
salient that you're being a good
person rather than what a lot of

1078
00:58:46,520 --> 00:58:49,000
people feel, which is just, oh, 
I could have done more. 

1079
00:58:49,000 --> 00:58:51,480
Like maybe they already provide 
that to themselves. 

1080
00:58:51,760 --> 00:58:52,720
That's true. 
Good point. 

1081
00:58:54,000 --> 00:58:57,520
I'd take that version. 
No criticism only compliments 

1082
00:58:57,520 --> 00:58:59,280
all. 
Right. 

1083
00:59:00,120 --> 00:59:07,600
LinkedIn anti pontificator AI, 
so this warns users not about if

1084
00:59:07,600 --> 00:59:11,080
they risk spreading 
misinformation, but if they use 

1085
00:59:11,120 --> 00:59:13,840
overly self-serving or smug 
writing. 

1086
00:59:15,160 --> 00:59:21,800
Oh yeah, yes, I'd like it. 
I'd like it. 

1087
00:59:22,280 --> 00:59:24,560
I'm trying to think through the 
2nd order and all the other 

1088
00:59:24,560 --> 00:59:26,880
effects. 
I suppose it might make 

1089
00:59:26,880 --> 00:59:29,960
pontificators just better, You 
know, they might just become 

1090
00:59:29,960 --> 00:59:34,240
better pontificators. 
So, you know, like they're like,

1091
00:59:34,320 --> 00:59:36,920
oh, I want to humble brag more 
efficiently. 

1092
00:59:37,400 --> 00:59:39,880
So is that a better world in 
which there's the same amount of

1093
00:59:40,080 --> 00:59:41,920
humble bragging but it's more 
palatable? 

1094
00:59:41,920 --> 00:59:43,680
I don't know. 
Yeah. 

1095
00:59:45,360 --> 00:59:49,640
That's great. 
OK, a fantasy Premier League AI 

1096
00:59:49,640 --> 00:59:53,400
banterbot basically only exists 
within the game to troll players

1097
00:59:53,400 --> 00:59:54,840
on their decisions and 
transfers. 

1098
00:59:56,480 --> 01:00:00,640
Yes, yes, although I have found 
it hard to. 

1099
01:00:01,560 --> 01:00:05,400
Maintain my high performance in 
fantasy Premier League over the 

1100
01:00:05,400 --> 01:00:08,120
years. 
As you may have noticed, I was 

1101
01:00:08,120 --> 01:00:11,440
once slightly far and away the 
best player in the leagues I was

1102
01:00:11,440 --> 01:00:12,640
in. 
Occasionally anyway. 

1103
01:00:13,040 --> 01:00:15,920
Yes, for the wider public, yes, 
let's do it for me. 

1104
01:00:16,200 --> 01:00:19,320
I don't know if I have time to 
engage with an AI banter bot at 

1105
01:00:19,320 --> 01:00:24,960
the moment. 
OK, Minority report style AI 

1106
01:00:24,960 --> 01:00:32,120
crime prediction algorithm. 
Well, well, well, well, yeah. 

1107
01:00:32,120 --> 01:00:41,800
So I'm going to say maybe I feel
that is ethically, yeah, that's 

1108
01:00:41,800 --> 01:00:43,800
a hard one. 
That's a really hard 1. 

1109
01:00:43,960 --> 01:00:46,400
And it is going to be a thing 
we're going to have to figure 

1110
01:00:46,400 --> 01:00:50,840
out that maybe we're already 
avoiding wrestling with now is 

1111
01:00:50,840 --> 01:00:55,120
like, how much should we use 
these AI tools for surveillance,

1112
01:00:55,120 --> 01:00:59,080
for crime prevention, sort of 
trading off privacy against the 

1113
01:00:59,080 --> 01:01:02,160
impacts on the society. 
So I don't know. 

1114
01:01:04,080 --> 01:01:09,000
Fair enough. 
OK final one AVR world that is 

1115
01:01:09,000 --> 01:01:12,000
identical to reality so maps 
exactly onto your life and your 

1116
01:01:12,000 --> 01:01:17,000
reality but without any of the 
stressful elements would like it

1117
01:01:17,000 --> 01:01:22,640
basically reduces away all the 
political drama, the wars, 

1118
01:01:22,640 --> 01:01:25,040
everything is scrubbed away so 
you can interacting with things 

1119
01:01:25,280 --> 01:01:28,120
even there's the Internet but 
you're not exposed to any of the

1120
01:01:28,120 --> 01:01:31,240
bad stuff there. 
So you basically put this on the

1121
01:01:31,240 --> 01:01:34,720
layer to reality that scrubs 
away the stress of it. 

1122
01:01:36,760 --> 01:01:40,520
Yeah, I think I'm God. 
How many? 

1123
01:01:44,040 --> 01:01:45,800
There's a lot of 2nd order 
effects here. 

1124
01:01:46,240 --> 01:01:48,800
I'm going to say my intuition 
without having thought about it 

1125
01:01:48,880 --> 01:01:53,480
much is that if it's truly for 
everybody in this kind of 

1126
01:01:53,680 --> 01:01:56,600
ridiculous hypothetical, I 
guess, you know, everybody can 

1127
01:01:56,600 --> 01:01:59,000
go into it and they're going to 
be sustained a bit like matrix 

1128
01:01:59,000 --> 01:02:00,800
light, but they're going to have
a good time. 

1129
01:02:01,440 --> 01:02:03,800
I dislike to make experience 
machine thought experiment, 

1130
01:02:03,800 --> 01:02:06,000
which I think is something along
the lines of you could be hooked

1131
01:02:06,000 --> 01:02:08,800
up to this machine and just 
experience like bliss. 

1132
01:02:09,720 --> 01:02:11,760
Would you do it rather than like
being in the real world with all

1133
01:02:11,760 --> 01:02:13,840
of these challenges? 
So I think like if it's for 

1134
01:02:13,840 --> 01:02:15,640
everyone, it's going to reduce a
lot of suffering. 

1135
01:02:15,640 --> 01:02:17,720
It's probably going to be net 
negative and everyone includes 

1136
01:02:17,720 --> 01:02:20,480
like animals and sentient 
creatures that otherwise suffer.

1137
01:02:20,840 --> 01:02:23,480
But if it's only going to be for
a selective number of people, 

1138
01:02:23,600 --> 01:02:25,440
then they're just kind of opting
out. 

1139
01:02:25,840 --> 01:02:28,800
And probably I'm still in favor 
of it if it's. 

1140
01:02:28,800 --> 01:02:30,920
Yeah, wow. 
So there you go. 

1141
01:02:30,920 --> 01:02:31,040
Wow. 
I. 

1142
01:02:31,880 --> 01:02:33,880
Feel like that could be your 
most controversial opinion? 

1143
01:02:35,040 --> 01:02:36,080
That is the next question, 
right? 

1144
01:02:37,240 --> 01:02:39,240
Yeah, it is. 
Peter, what's your most 

1145
01:02:39,240 --> 01:02:44,200
controversial opinion about AI? 
Well, I think in the community 

1146
01:02:44,200 --> 01:02:49,880
that I'm in, probably I'm less 
concerned about risks than some 

1147
01:02:49,880 --> 01:02:53,800
people are. 
Maybe one opinion is I really am

1148
01:02:53,800 --> 01:02:57,480
more maybe positive about the 
idea of like AI as a companion. 

1149
01:02:58,400 --> 01:03:01,920
I've recently been enjoying 
having more conversations with 

1150
01:03:01,920 --> 01:03:05,280
advanced voice mode and there's 
a Sesame, which I think I shared

1151
01:03:05,280 --> 01:03:08,240
with you, Sam. 
Yeah, I feel there's a lot of 

1152
01:03:08,240 --> 01:03:12,160
benefits to talking with AI over
talking with humans. 

1153
01:03:12,200 --> 01:03:15,760
You know, I feel I'm really 
aware that it doesn't matter if 

1154
01:03:15,760 --> 01:03:18,240
I interrupted or if I ramble. 
I don't know. 

1155
01:03:18,320 --> 01:03:19,640
I'm not causing. 
Like there's all this 

1156
01:03:19,640 --> 01:03:21,640
interesting research actually 
that found if, you know, for 

1157
01:03:21,640 --> 01:03:25,240
example, people are playing 
chess with an AI, they don't 

1158
01:03:25,240 --> 01:03:27,040
experience as much discomfort 
when they lose. 

1159
01:03:27,080 --> 01:03:31,240
Like in some way their mind kind
of understands I didn't lose. 

1160
01:03:31,320 --> 01:03:33,400
Maybe it's an evolutionary thing
or whatever, but they don't get 

1161
01:03:33,400 --> 01:03:36,040
this like I have lost status in 
my group and you know, it's 

1162
01:03:36,040 --> 01:03:37,840
really bad and maybe I'm going 
to get rejected. 

1163
01:03:37,840 --> 01:03:41,040
Or all the things maybe that are
theorized to lead to like the 

1164
01:03:41,040 --> 01:03:43,600
strong emotional response when 
you lead to lose to a human. 

1165
01:03:43,920 --> 01:03:46,320
So the similar dynamics when 
you're talking with the AI where

1166
01:03:46,320 --> 01:03:49,080
you just you don't feel the same
pressures about things. 

1167
01:03:49,080 --> 01:03:51,000
You can ask you to explain 
things and definitely it's 

1168
01:03:51,000 --> 01:03:54,320
always available. 
It will get to know you better 

1169
01:03:54,360 --> 01:03:57,360
and it will be better informed. 
Like one of my friends was 

1170
01:03:57,360 --> 01:04:01,880
telling me they used AI to try 
to analyze text messages that 

1171
01:04:01,880 --> 01:04:04,640
exchanged with somebody, you 
know, that they were dating to 

1172
01:04:04,640 --> 01:04:07,800
sort of understand like whether 
that person liked them or not, 

1173
01:04:07,800 --> 01:04:10,960
and also to predict whether that
person would send them a message

1174
01:04:10,960 --> 01:04:14,440
at some point in the future. 
And it was very helpful for both

1175
01:04:14,440 --> 01:04:15,960
of those things. 
And apparently, it got both the 

1176
01:04:15,960 --> 01:04:18,320
timing of the message and the 
topic of the message right. 

1177
01:04:18,840 --> 01:04:23,120
So this is maybe a piece of the 
AI world, although there's some 

1178
01:04:23,120 --> 01:04:25,800
risks there about unsafe use and
over reliance and things like 

1179
01:04:25,800 --> 01:04:28,480
that where maybe I'm a little 
bit more optimistic. 

1180
01:04:28,480 --> 01:04:31,920
And I think some people will be 
like, no, you know, I'm pro 

1181
01:04:31,920 --> 01:04:34,440
human, humans all the way, human
relationships only. 

1182
01:04:34,440 --> 01:04:37,480
But I think I have room in my 
heart for an AI as well. 

1183
01:04:37,800 --> 01:04:40,680
You know, you mentioned Sesame 
as one of the recent demos 

1184
01:04:40,720 --> 01:04:45,120
around voice remotes. 
And I think I found myself 

1185
01:04:45,360 --> 01:04:48,720
starting, you know, having a 
short chat that I was just going

1186
01:04:48,720 --> 01:04:51,760
to test its capabilities. 
And then 30 minutes later, I was

1187
01:04:51,760 --> 01:04:54,760
still talking. 
And I found it interesting to 

1188
01:04:54,800 --> 01:04:57,880
like, I found myself getting and
just talking about more normal 

1189
01:04:57,880 --> 01:05:01,320
stuff. 
And it's from a behavioral 

1190
01:05:01,320 --> 01:05:03,840
standpoint, very understandable 
that of course, it'd be nice to 

1191
01:05:03,840 --> 01:05:06,880
have someone that's infinitely 
patience and always speaking to 

1192
01:05:06,880 --> 01:05:10,480
at your level. 
And also can have like this kind

1193
01:05:10,480 --> 01:05:14,560
of very paradoxical nature where
it can speak to you in a very 

1194
01:05:14,720 --> 01:05:18,880
kind of nice normal way, but 
have the knowledge of like the 

1195
01:05:18,880 --> 01:05:23,120
most smart and intelligent, you 
know, collection of PhD level 

1196
01:05:23,240 --> 01:05:26,800
people that like a lot of people
would feel very, I would say, 

1197
01:05:27,200 --> 01:05:30,160
intellectually afraid to speak 
to because they were like, how 

1198
01:05:30,160 --> 01:05:33,240
can I even dare to ask questions
to someone like Jeffrey Hinton 

1199
01:05:33,240 --> 01:05:34,760
about AI or someone. 
Yeah. 

1200
01:05:34,960 --> 01:05:37,480
That's a really great point. 
I mean, I do, I'm sure a lot of 

1201
01:05:37,480 --> 01:05:40,800
people feel this, like I feel 
very frequently uncomfortable if

1202
01:05:40,800 --> 01:05:43,120
I'm talking to somebody who I 
know is very smart and very 

1203
01:05:43,120 --> 01:05:45,960
successful. 
But that's often the way that 

1204
01:05:45,960 --> 01:05:47,360
you get a lot of the best 
insights. 

1205
01:05:47,360 --> 01:05:50,200
So, you know, you kind of push 
through that and then, you know,

1206
01:05:50,720 --> 01:05:52,240
you ask questions of those 
people. 

1207
01:05:52,840 --> 01:05:56,840
But yeah, you don't feel at all 
the same sort of pressure here 

1208
01:05:56,840 --> 01:05:58,720
if you're talking to an AI, 
which would be even more 

1209
01:05:58,720 --> 01:06:00,480
knowledgeable. 
And then the other thing which 

1210
01:06:00,600 --> 01:06:02,600
you didn't mention, which I know
we both have talked about 

1211
01:06:02,600 --> 01:06:06,080
before, is like a key thing to 
think about here is that the 

1212
01:06:06,080 --> 01:06:08,520
versions that we're trying now 
are the worst versions we'll 

1213
01:06:08,520 --> 01:06:09,960
ever use. 
So, you know, you think, oh, 

1214
01:06:10,000 --> 01:06:12,320
that was pretty good. 
And you're like in five years or

1215
01:06:12,320 --> 01:06:14,920
three years and two years, it's 
hard to know the time frame if 

1216
01:06:14,920 --> 01:06:17,560
this is much better and really 
knowledgeable about me and my 

1217
01:06:17,560 --> 01:06:18,920
life. 
And you know, it can sort of 

1218
01:06:19,400 --> 01:06:21,920
talk about like what I did 
yesterday and they can say 

1219
01:06:22,160 --> 01:06:24,560
tomorrow, you know, you've got 
this presentation and I noticed 

1220
01:06:24,560 --> 01:06:26,720
that like you haven't said or 
done anything about it today. 

1221
01:06:26,720 --> 01:06:28,560
And like, is there anything I 
can help you with there? 

1222
01:06:28,560 --> 01:06:34,600
It's going to be very hard for 
other types of relationships to 

1223
01:06:34,600 --> 01:06:39,120
match that in terms of the cost,
reward, sort of benefit or 

1224
01:06:39,120 --> 01:06:41,480
whatever. 
Yeah, I know for sure. 

1225
01:06:41,640 --> 01:06:45,840
And I guess to wrap up, I can 
say that it's hard to think 

1226
01:06:45,840 --> 01:06:48,800
about many of the questions that
we've covered today, but I think

1227
01:06:48,800 --> 01:06:52,400
an easy answer for me is that I 
will predict that in many years 

1228
01:06:52,400 --> 01:06:55,920
in the future, I'll still find 
it as valuable to speak with you

1229
01:06:55,920 --> 01:06:58,560
and listen to you as today and 
as in the past. 

1230
01:06:58,920 --> 01:07:03,480
So I'm super happy that I was 
able to convince you to finally 

1231
01:07:03,480 --> 01:07:06,440
come on the podcast. 
And I think it was a really fun 

1232
01:07:06,440 --> 01:07:09,280
conversation today. 
So yeah, thank you for coming on

1233
01:07:09,280 --> 01:07:12,440
the podcast, but also honestly, 
thank you for all of the great 

1234
01:07:12,440 --> 01:07:15,160
work. 
And I feel like I've been very 

1235
01:07:15,160 --> 01:07:18,320
lucky to learn from you and know
you've a long time. 

1236
01:07:18,520 --> 01:07:20,120
Thank you, Sam. 
That's very wholesome. 

1237
01:07:20,600 --> 01:07:24,280
But no, genuinely, that is 
lovely and I reciprocate your 

1238
01:07:24,280 --> 01:07:26,760
appreciation. 
And yeah, it's been really 

1239
01:07:26,760 --> 01:07:30,160
valuable and I'm very excited 
now to have done this, this 

1240
01:07:30,160 --> 01:07:33,520
first podcast and overcome that 
fear, I suppose, that I had. 

1241
01:07:34,440 --> 01:07:36,520
And that's a wrap. 
You've been listening to the 

1242
01:07:36,520 --> 01:07:39,880
Behavioral Design podcast 
brought to you by Habit Weekly 

1243
01:07:39,880 --> 01:07:42,720
and Nuance Behavior. 
Sam and Aline tell me This 

1244
01:07:42,720 --> 01:07:46,040
season is packed with incredible
insights about behavioral design

1245
01:07:46,040 --> 01:07:49,680
and AI, so be sure to subscribe 
and share the podcast with your 

1246
01:07:49,680 --> 01:07:51,480
friends. 
Though you might want to keep it

1247
01:07:51,480 --> 01:07:55,480
away from your enemies. 
In case you haven't noticed, I'm

1248
01:07:55,480 --> 01:07:58,920
an AI voice. 
Yep, pretty crazy. 

1249
01:07:59,200 --> 01:08:02,160
Quite the improvement since last
season's AI outro, don't you 

1250
01:08:02,160 --> 01:08:04,880
think? 
If you'd like to collaborate 

1251
01:08:04,880 --> 01:08:08,000
with us at Nuance Behavior, 
where we use behavioral design 

1252
01:08:08,000 --> 01:08:10,800
to craft digital products with 
Nuance, e-mail us at 

1253
01:08:10,800 --> 01:08:15,160
hello@nuancebehavior.com or book
a call directly on our website, 

1254
01:08:15,360 --> 01:08:19,880
nuancebehavior.com. 
A special thanks to the amazing 

1255
01:08:19,880 --> 01:08:23,479
Dave Pizarro for our show music,
and to Mei Chen Yap and April 

1256
01:08:23,479 --> 01:08:26,279
English for their help in 
producing and publishing this 

1257
01:08:26,279 --> 01:08:28,560
episode. 
Thanks again for tuning in. 

1258
01:08:28,800 --> 01:08:31,520
We'll be back soon with another 
exciting conversation where 

1259
01:08:31,520 --> 01:08:34,080
behavioral design and AI 
intersect. 

1260
01:08:39,920 --> 01:08:41,240
Happens. 
To. 

1261
01:08:48,920 --> 01:08:58,359
Mugatroid. 
Oh. 

1262
01:09:33,120 --> 01:09:36,760
That is kind of likened to 
evolution of interspecies, where

1263
01:09:37,040 --> 01:09:40,680
if you have an Otter and 
something like a polar bear, if 

1264
01:09:40,680 --> 01:09:44,800
one of them becomes better at 
hiding or hunting, the other one

1265
01:09:44,840 --> 01:09:48,520
kind of selectively becomes. 
Better at the other thing. 

1266
01:09:48,520 --> 01:09:51,760
Do polar bears eat otters? 
They eat the seals. 

1267
01:09:51,760 --> 01:09:55,400
They eat the seals. 
Seals, seals, seals.