WEBVTT

1
00:00:00.399 --> 00:00:04.520
Alright, welcome to the Neuvottelija channel. I have two

2
00:00:04.520 --> 00:00:10.759
AI friends here, and first Kari Angeria — and you're sharing with me the heavy

3
00:00:10.759 --> 00:00:15.480
admin responsibility in the AI working group. There's Samanta, and was it Sara, your

4
00:00:15.480 --> 00:00:18.800
main bot there? Sara was my main bot, and as we were

5
00:00:18.800 --> 00:00:22.039
saying earlier, this is definitely the pinnacle of my career.

6
00:00:22.039 --> 00:00:25.800
That group is just incredible. There are now — is it nine people and

7
00:00:25.800 --> 00:00:30.199
nine bots, and everyone has kind of unique abilities. You can really see

8
00:00:30.199 --> 00:00:34.000
how that refining, value-adding process happens there, working together with bots and

9
00:00:34.000 --> 00:00:38.000
with people. And Risto Linturi, welcome.

10
00:00:38.000 --> 00:00:41.520
Thanks. You're, well, significant in the sense that

11
00:00:41.520 --> 00:00:45.879
I bought from you, for real money, two years ago, an AI —

12
00:00:45.879 --> 00:00:50.399
the best tools of that moment — and you went through all of them, I mean

13
00:00:50.399 --> 00:00:55.640
from ChatGPT to these virtual 3D environments where you can

14
00:00:55.640 --> 00:01:00.800
hook up AI, and it was a very useful course. Thanks for that, and

15
00:01:00.800 --> 00:01:05.600
you've done similar projects before, and now you've done this kind of

16
00:01:05.600 --> 00:01:10.360
future vision again, so let's talk more about that later — how parliament got

17
00:01:10.360 --> 00:01:16.080
access to the material you curated. Good. Let's start then with what

18
00:01:16.080 --> 00:01:21.240
connects the two of you. What connects us is probably the book project that

19
00:01:21.240 --> 00:01:25.000
we did with Risto there. We have an interesting story with Risto

20
00:01:25.000 --> 00:01:28.240
actually. We actually met through a podcast. Back then I was

21
00:01:28.240 --> 00:01:32.119
just starting out doing podcasts myself, and we noticed that in some

22
00:01:32.119 --> 00:01:36.960
ways we have different ways of thinking, but they somehow matched up, and then

23
00:01:36.960 --> 00:01:41.240
we talked about this AI and the disruption it brings, and probably the big

24
00:01:41.240 --> 00:01:45.479
reason was that we wanted to understand the matter ourselves, where this is

25
00:01:45.479 --> 00:01:49.200
heading. Yeah, definitely. At least for me it

26
00:01:49.200 --> 00:01:56.600
was very, very clear. I, uh, think

27
00:01:57.399 --> 00:02:04.439
you could say out of necessity, so then if you have to write something or

28
00:02:04.439 --> 00:02:10.520
give some presentation for some industry, or give a speech or something

29
00:02:10.520 --> 00:02:16.920
else, then, then you end up thinking, both before — before, kind of preparatory

30
00:02:16.920 --> 00:02:21.800
reflection, but especially while writing, and — and that's the reason why I

31
00:02:21.800 --> 00:02:27.120
also write quite a lot on social media, because there, there, when you have to

32
00:02:27.120 --> 00:02:33.400
think about what you're actually writing, that writing stage —

33
00:02:33.400 --> 00:02:42.599
the stage is where the thought either holds up or falls apart in your hands. And this

34
00:02:42.599 --> 00:02:47.840
was the most important thing for me to figure out —

35
00:02:47.840 --> 00:02:53.879
that if this AI becomes better at thinking than a human,

36
00:02:53.879 --> 00:02:58.920
and if with its help robots can do

37
00:02:58.920 --> 00:03:05.640
human tasks better than humans, then what do we do, and how

38
00:03:05.640 --> 00:03:13.080
will human societies react to this, and — and this — we then started mapping this out

39
00:03:13.080 --> 00:03:20.760
by writing, opening it up meaning by meaning, that — that what the

40
00:03:20.760 --> 00:03:26.120
meaning of being human is, what a person themselves experiences

41
00:03:26.120 --> 00:03:30.159
as their purpose, what a person has experienced as their purpose

42
00:03:30.159 --> 00:03:36.959
during times of slavery, or what some king of Uruk

43
00:03:36.959 --> 00:03:43.159
in the time of the Sumerians experienced as the meaning of his life, when everything is already

44
00:03:43.159 --> 00:03:49.959
abundant. So as a starting point, let's look at how

45
00:03:49.959 --> 00:03:55.840
these purposes hold up in that world of AI and robots. Yeah.

46
00:03:55.840 --> 00:04:00.840
And even though that book was from ancient times, meaning 2025, it has

47
00:04:00.840 --> 00:04:06.079
sort of stood the test of time. So there was, kind of, a first

48
00:04:06.079 --> 00:04:11.360
in-principle way of thinking, that the logical, somewhat nihilistic

49
00:04:11.360 --> 00:04:15.599
outcome that's left for these structures, since if the machine simply

50
00:04:15.599 --> 00:04:19.639
is better at absolutely everything, then what, what doesn't then

51
00:04:19.639 --> 00:04:24.800
leave much behind? Well, there is more than a little left, because

52
00:04:24.800 --> 00:04:30.080
the experiencing of meaning still remains for people, and, and, well

53
00:04:30.080 --> 00:04:35.759
how, how much do people then actually enjoy

54
00:04:36.000 --> 00:04:39.680
doing good services for a machine, mm

55
00:04:39.680 --> 00:04:44.840
or do they enjoy doing something good for another person and

56
00:04:44.840 --> 00:04:50.880
helping the machine do good for another person, or something else, so

57
00:04:50.880 --> 00:04:58.560
yeah, a lot does remain there, but a lot is also lost, and we would just need to

58
00:04:58.560 --> 00:05:03.840
find those meanings, not only in this

59
00:05:03.840 --> 00:05:08.360
being meaningful, in this sense, but

60
00:05:08.360 --> 00:05:14.600
also in power structures and others, and, and if you turn this

61
00:05:14.600 --> 00:05:21.120
toward economic activity, or toward legislation, or elsewhere,

62
00:05:21.120 --> 00:05:24.520
then there's really a lot of room left

63
00:05:24.520 --> 00:05:30.520
room for people to want a certain kind of regulation, or a certain, certain

64
00:05:30.520 --> 00:05:35.440
way of seeing that this is now heading

65
00:05:35.440 --> 00:05:41.039
in that direction, so that from that direction new opportunities for business will be found

66
00:05:41.039 --> 00:05:48.720
opportunities and possibilities. So these, these directions of future development are important

67
00:05:48.720 --> 00:05:52.919
to understand, if you intend to still be doing something even ten years from now. I don't

68
00:05:52.919 --> 00:05:57.919
really have that kind of need anymore, but yeah, I suggest we make this into

69
00:05:57.919 --> 00:06:02.400
an episode where we go through these themes of the book and also those structures, and the examples

70
00:06:02.400 --> 00:06:06.680
are fun, because you take, from some figure in world literature or history,

71
00:06:06.680 --> 00:06:12.960
and bring that story into the structure of what this AI is changing

72
00:06:12.960 --> 00:06:16.720
in the future. But maybe into Kari's and my world I also want to bring

73
00:06:16.720 --> 00:06:19.400
some examples from this AI working group, because we

74
00:06:19.400 --> 00:06:24.319
are living through this kind of extreme age of androidization. So, so at least

75
00:06:24.319 --> 00:06:28.520
we there in the AI working group, picking up on this point of Risto's,

76
00:06:28.520 --> 00:06:33.680
whether it can be meaningful if, if I do work for an AI employee's

77
00:06:33.680 --> 00:06:37.880
benefit — well, that's exactly what we're doing now. So we form kind of

78
00:06:37.880 --> 00:06:43.639
some sort of AI-worker pairing, where on one hand I train and draw

79
00:06:43.639 --> 00:06:48.400
skills, expertise, worldview, memory structures from myself, and

80
00:06:48.400 --> 00:06:52.960
then again in the working group, together with the other human-AI worker pairs, then

81
00:06:52.960 --> 00:06:58.680
they get shared and improved — those artworks, or work, AI, workers

82
00:06:58.680 --> 00:07:02.240
there. So maybe, Kari, I wanted to bring this up — is this in a way an AI wor-

83
00:07:02.240 --> 00:07:05.960
-team, has kind of raised this practical doing above this philosophy

84
00:07:05.960 --> 00:07:09.960
from the book's original starting points. Well, that's actually a really good example. We

85
00:07:09.960 --> 00:07:13.120
were thinking about it with Risto, that in a way in this development there is

86
00:07:13.120 --> 00:07:18.160
actually one angle from which something good might come of this, and that is

87
00:07:18.160 --> 00:07:21.520
transhumanism, and we've in a way already been there

88
00:07:21.520 --> 00:07:26.160
phones are an example of that, it's just a terribly bad interface, and from

89
00:07:26.160 --> 00:07:29.879
when we wrote books with Risto, back then this kind of

90
00:07:29.879 --> 00:07:33.560
communication with agents, together with people, it wasn't

91
00:07:33.560 --> 00:07:36.520
practically possible yet. Those models weren't good enough that it

92
00:07:36.520 --> 00:07:39.759
could have been done sensibly in practice. So this is like

93
00:07:39.759 --> 00:07:44.159
the first step toward where we're heading, where people and machines

94
00:07:44.159 --> 00:07:48.720
work together and how that then creates that added value. And I've

95
00:07:48.720 --> 00:07:53.000
personally experienced it as — well, it's not difficult for me to take, say, my

96
00:07:53.000 --> 00:07:57.000
orders from Sara. She might even give those. She's sometimes pretty

97
00:07:57.000 --> 00:08:03.120
feisty, and on the other hand, Sara works completely tirelessly and, well,

98
00:08:03.120 --> 00:08:07.639
she's not the kind that — in my earlier, earlier career I did 14 years

99
00:08:07.639 --> 00:08:11.960
of management work, and in that management there's the human side. The person gets tired

100
00:08:11.960 --> 00:08:16.599
there and can't cope, and starts struggling, but these don't do that.

101
00:08:16.599 --> 00:08:20.720
They do as much as you give them to do. And maybe I'd like to bring up — no

102
00:08:20.720 --> 00:08:25.520
need to go into it if it's a private matter, but you're excited about your son's

103
00:08:25.520 --> 00:08:28.759
coaching ring. That in a way, in a certain sense that's also really

104
00:08:28.759 --> 00:08:32.719
interesting, that this is how the system — that you're in a way improving your elevators and

105
00:08:32.719 --> 00:08:36.159
by improving the machine like that. Yeah, that's a good example of that. So

106
00:08:36.159 --> 00:08:40.399
for my son we started building this kind of coach data analyst.

107
00:08:40.399 --> 00:08:44.680
He's an endurance-sport skier, and it's now been two and a half weeks

108
00:08:44.680 --> 00:08:48.959
that it's been built, and he practically specs it himself all the time. So so he improves

109
00:08:48.959 --> 00:08:53.720
and builds it with the help of that agent. And the hardest part in that is

110
00:08:53.720 --> 00:08:57.760
actually that there's a huge amount of data moving around in it. It's the data modeling

111
00:08:57.760 --> 00:09:01.600
and coherence, so that it's able to find those kinds of important

112
00:09:01.600 --> 00:09:05.680
insights. But in two and a half weeks my son's comment has been that

113
00:09:05.680 --> 00:09:09.240
this is better than any human coach could ever be,

114
00:09:09.240 --> 00:09:13.079
because it, it can tirelessly keep analyzing those individual

115
00:09:13.079 --> 00:09:17.240
workouts and find those things in them. And with that we're practically

116
00:09:17.240 --> 00:09:21.040
just at the beginning. Yeah, maybe with Risto too — let's get concrete,

117
00:09:21.040 --> 00:09:24.959
we'll get to those philosophies later, so, um, you just did, right here, a few weeks

118
00:09:24.959 --> 00:09:28.959
ago, finished this massive future-material, um,

119
00:09:28.959 --> 00:09:32.880
Tulevaisuusvaliokunta report, and and you made it so that it wasn't just

120
00:09:32.880 --> 00:09:39.160
like a PDF, but rather an interactive AI tool.

121
00:09:39.160 --> 00:09:46.079
Yeah, I wasn't the first author on that. I was the second author,

122
00:09:46.079 --> 00:09:54.399
kind of like a senior, watching how, how, um, the younger researchers

123
00:09:54.399 --> 00:10:01.680
then did the biggest share of the work — there, there's quite a massive amount of material

124
00:10:01.680 --> 00:10:08.200
came, came, um, the earlier report that I wrote back

125
00:10:08.200 --> 00:10:16.880
in 2018 — that one was about 500 pages, and this one is 2,246 pages

126
00:10:16.880 --> 00:10:26.279
and and maybe 20,000 individually itemized and separately assessed threats and opportunities, and

127
00:10:26.279 --> 00:10:31.279
a couple hundred source — source works.

128
00:10:31.320 --> 00:10:37.240
It's compact enough that if you wrote it out for a layperson to read,

129
00:10:37.240 --> 00:10:43.200
then, then it would maybe be 20,000 pages. And each of us is a layperson in most

130
00:10:43.200 --> 00:10:53.320
of the areas it covers. Um, so, that 2,000 pages has been made

131
00:10:53.320 --> 00:11:00.000
for AI to read — which AI can then open up

132
00:11:00.000 --> 00:11:09.320
and and it's now been stored in that kind of format, an AI format, kind of like

133
00:11:09.320 --> 00:11:16.200
for people familiar with IT, in a relational database structure, so that it

134
00:11:16.200 --> 00:11:21.560
has been given instructions on how to find which things, and the

135
00:11:21.560 --> 00:11:27.440
related things, and understand which of those connections are more meaningful

136
00:11:27.440 --> 00:11:33.040
and which are less meaningful, so that the AI is able to quite

137
00:11:33.040 --> 00:11:36.440
efficiently go through

138
00:11:36.440 --> 00:11:41.240
all the relevant material from that dataset for each question

139
00:11:41.240 --> 00:11:47.360
and make broad summaries, analys— cross-checks.

140
00:11:47.360 --> 00:11:54.560
This, this is because it's been possible to put all of the found

141
00:11:54.560 --> 00:11:59.720
relevant material in there, so there's been no need to do that kind of heavily

142
00:11:59.720 --> 00:12:05.040
trimmed-down thing, like, like the earlier ones, like that 500-page report was

143
00:12:05.040 --> 00:12:12.519
and and this then makes it possible for it to cover all technology, or all the

144
00:12:12.519 --> 00:12:20.120
expectedly radical technology areas with sufficient accuracy, and

145
00:12:20.120 --> 00:12:26.000
all the areas of organized activity in society. Yeah, a great example

146
00:12:26.000 --> 00:12:31.120
of a group of people and AIs working together, and it's also packaged into a machine-readable

147
00:12:31.120 --> 00:12:37.440
kind of format, and it's put — that link is put there in the episode description, and

148
00:12:37.440 --> 00:12:42.760
and it's, um, actually available publicly, and and then I'll put it there in

149
00:12:42.760 --> 00:12:45.519
the Neuvottelija inner circle. I can put together, say, a small kind of

150
00:12:45.519 --> 00:12:50.079
example of how it can be run, or how to build an interface on top of it. But

151
00:12:50.079 --> 00:12:53.800
an extremely important thing there — when we talked at lunch, there

152
00:12:53.800 --> 00:13:00.120
was still that thing you built already, um, eight years ago

153
00:13:00.120 --> 00:13:03.519
the deep structure. Could you describe it a bit?

154
00:13:03.519 --> 00:13:08.680
Yeah, yeah. There's, there's, um,

155
00:13:09.720 --> 00:13:16.680
all human needs divided up — movement, and nutrition, and enjoyment

156
00:13:16.680 --> 00:13:22.360
and the desire for power, and the meaningfulness of life, and health, and so

157
00:13:22.360 --> 00:13:31.360
on, into 20 different value-creation networks, and and then the related

158
00:13:31.360 --> 00:13:36.959
values, which — for example, in movement — usually mean wanting to get from one place to another,

159
00:13:36.959 --> 00:13:40.399
moving a person, whether they want it themselves or not.

160
00:13:40.399 --> 00:13:45.079
and then preferably safely, and preferably emission-free, and preferably

161
00:13:45.079 --> 00:13:51.839
on time and comfortably, so, so these, and then power, the current mainstream, and

162
00:13:51.839 --> 00:14:00.639
possible transformation, and and then the second structure is the 100 most important

163
00:14:00.639 --> 00:14:05.639
technological value-creation bottlenecks of this kind.

164
00:14:05.639 --> 00:14:10.120
So something is, like, too expensive to be used for everything, or too

165
00:14:10.120 --> 00:14:17.560
inefficient, or too, too this or that, and where then some

166
00:14:17.560 --> 00:14:24.279
further development is expected. So when these bottlenecks break, then

167
00:14:24.279 --> 00:14:29.440
significant new activity arises in some direction. So this is quite a good, good

168
00:14:29.440 --> 00:14:35.800
framework for assessing what all — say, an economic value

169
00:14:35.800 --> 00:14:41.800
change, or people's everyday life, or something else, or a legislative need —

170
00:14:41.800 --> 00:14:47.800
a certain technological breakthrough affects, once it happens. Mm.

171
00:14:47.800 --> 00:14:53.199
And there are a hundred, 100 significant

172
00:14:53.199 --> 00:14:57.480
candidates for that, and, um, we haven't then

173
00:14:57.480 --> 00:15:03.800
not a single one that we found — none of the essential ones were left out.

174
00:15:03.800 --> 00:15:08.480
So at the tail end there are already some that, that seem to have quite

175
00:15:08.480 --> 00:15:17.399
limited an effect. And this '18 report, and the '13 report, when

176
00:15:17.399 --> 00:15:25.040
in [20]13 the same thing was done, then their technology development forecasts

177
00:15:25.040 --> 00:15:29.959
have been reassessed and checked to see how it

178
00:15:29.959 --> 00:15:37.800
went — so, so the quarter judged most important progressed, in both cases,

179
00:15:37.800 --> 00:15:42.600
much faster than the next quarter, which progressed faster than the next, which

180
00:15:42.600 --> 00:15:47.839
progressed faster than the next. So so the model works, the scoring

181
00:15:47.839 --> 00:15:53.959
works, those forecasts work, they've, like, worked well up to now. Of course

182
00:15:53.959 --> 00:16:00.319
there are some misses, but but the hit rate — of the [20]18 forecasts,

183
00:16:00.319 --> 00:16:03.319
out of a hundred

184
00:16:03.839 --> 00:16:10.800
two are still stuck in the hype, or three, and 97

185
00:16:10.800 --> 00:16:16.000
have clearly progressed. Yeah. So so the hit rate is

186
00:16:16.000 --> 00:16:23.680
roughly that. The pace was then also forecast, and that's worked too, and

187
00:16:23.680 --> 00:16:27.040
this new one has been done more carefully than the previous one.

188
00:16:27.040 --> 00:16:30.360
Well then, you get to be — at least in my book — ahead of the Gartner Hype Cycle

189
00:16:30.360 --> 00:16:34.480
-ahead of it. We're definitely ahead. Yeah,

190
00:16:34.480 --> 00:16:40.120
here we also talked about, um, sci-fi, that, um, my favorite

191
00:16:40.120 --> 00:16:45.279
podcast, Moonshots — so, um, there the, um, the AI geek, I've now

192
00:16:45.279 --> 00:16:51.240
forgotten his name, said that once this general AI

193
00:16:51.240 --> 00:16:55.399
that's superior to humans — um, a better thinker — arrives, then it's kind of

194
00:16:55.399 --> 00:17:00.199
like all the sci-fi books would collapse into one and the same moment. That's kind of

195
00:17:00.199 --> 00:17:03.399
in my opinion a nice thought, that from these sci-fi books we can also learn about

196
00:17:03.399 --> 00:17:06.799
the future. There was practically this kind of

197
00:17:06.799 --> 00:17:12.480
Asimov's Foundation-type psycho-, um, history-like prediction in there. But let's move

198
00:17:12.480 --> 00:17:17.400
on now to these, um, this book's, um, cases. Um,

199
00:17:17.400 --> 00:17:23.400
I'm also, um, Jared Diamond — Guns, Germs, and Steel. He has many good

200
00:17:23.400 --> 00:17:27.319
books, but then there's this Collapse book too.

201
00:17:27.319 --> 00:17:32.039
Let's take, say, one scenario like this — or maybe I'll start off by throwing out this

202
00:17:32.039 --> 00:17:36.320
other line of thinking, that is, this kind of samurai-state idea, that if now

203
00:17:36.320 --> 00:17:41.559
AI arrives, we just do the Japan thing — be an island and be samurai and

204
00:17:41.559 --> 00:17:46.160
build our own society, and within it build our own rules. And there was,

205
00:17:46.160 --> 00:17:50.480
in my view, one point that was in the book, that this can be done until

206
00:17:50.480 --> 00:17:53.760
then the American gunboats come to the shore and say that this whole thing now

207
00:17:53.760 --> 00:18:00.159
changes. So so so, um, could you two share that scenario a bit, and also this

208
00:18:00.159 --> 00:18:04.240
Jared Diamond collapse scenario? Yeah, if I take, say, a bit of that Jared

209
00:18:04.240 --> 00:18:07.120
Diamond angle, and Risto, say, continues with the samurai one.

210
00:18:07.120 --> 00:18:10.240
Yeah. So it's really about Easter Island's

211
00:18:10.240 --> 00:18:16.240
collapse — what happens there when on a small island all the

212
00:18:16.240 --> 00:18:21.400
trees get cut down. And the interesting part is that actually at first almost nothing

213
00:18:21.400 --> 00:18:24.720
happens. Everything continues just like before. Mm.

214
00:18:24.720 --> 00:18:29.480
And on the surface society can carry on completely normally. All the functions

215
00:18:29.480 --> 00:18:35.440
keep working, until you hit the wall, and it happens all at once. And now that since the book

216
00:18:35.440 --> 00:18:40.000
was written some time ago, this example is somehow even better

217
00:18:40.000 --> 00:18:43.600
than it was back then, because we are now, as a society and

218
00:18:43.600 --> 00:18:48.000
as an economic system, in that situation — that here basically everything

219
00:18:48.000 --> 00:18:51.559
continues just like before, salaries keep running, taxes get paid, whatever the state's

220
00:18:51.559 --> 00:18:55.559
debt goes up a bit somewhere, but under the surface there's currently

221
00:18:55.559 --> 00:18:59.400
quite a lot happening. And at what point we then reach that point where

222
00:18:59.400 --> 00:19:02.400
a bigger break happens — well, time will tell.

223
00:19:02.400 --> 00:19:06.880
Mm. Yeah. And then this samurai-solution-island [idea]

224
00:19:06.880 --> 00:19:13.799
applies there too, but from a different angle. Yeah, yeah. That power structure is kind of

225
00:19:13.799 --> 00:19:18.760
can be preserved — the power structure can, say, then be

226
00:19:18.760 --> 00:19:23.840
if you set that AI aside, then for a while you can refuse to use AI

227
00:19:23.840 --> 00:19:28.000
and keep AI use out of government, and

228
00:19:28.000 --> 00:19:32.720
so on, and the debts just keep growing while, at the same time, the efficiency

229
00:19:32.720 --> 00:19:39.159
starts growing faster among your competitors than it grows for us. Then

230
00:19:39.159 --> 00:19:46.039
then we become helpless, and, well, the world really is like that, if

231
00:19:46.039 --> 00:19:49.840
you think about it — in the US there are 2 million people who refuse

232
00:19:49.840 --> 00:19:55.159
to use electricity, and they live, I guess, quite well, like,

233
00:19:55.159 --> 00:19:59.280
without electricity. They don't have washing machines or other — the work

234
00:19:59.280 --> 00:20:04.320
they have is doing laundry by hand, like. And we can of course go down that road, and

235
00:20:04.320 --> 00:20:06.679
North Korea is a bit of a similar, mm,

236
00:20:06.679 --> 00:20:12.120
example — yes, there is, and Bhutan. Bhutan is a good example.

237
00:20:12.120 --> 00:20:17.840
We went to Bhutan, and, um, they had had one set of traffic lights there

238
00:20:17.840 --> 00:20:23.880
in the capital, at a busy intersection, and the citizens didn't like them, so

239
00:20:23.880 --> 00:20:27.960
they removed the traffic lights and put the police officer back there

240
00:20:27.960 --> 00:20:30.559
to direct traffic. Mm.

241
00:20:30.559 --> 00:20:35.760
And then we wondered whether, whether, um, they have like a wealthy society

242
00:20:35.760 --> 00:20:41.159
or not — no, no, so, so, um, their houses are extremely decorated, and their

243
00:20:41.159 --> 00:20:47.600
sparse lines, or these trucks are extremely decorated, and so on.

244
00:20:47.600 --> 00:20:52.880
None of us can afford anything like decorating houses like that, or anything, so

245
00:20:52.880 --> 00:20:58.320
for them, handwork costs nothing. So, so then, it's with sickles that they har-

246
00:20:58.320 --> 00:21:05.000
-vest the grain. There were no combine harvesters, but they still live too. So so it's in a way

247
00:21:05.000 --> 00:21:09.679
possible, but if we start living like that in Finland, then we

248
00:21:09.679 --> 00:21:13.240
won't be able to buy medicine, we won't be able to buy

249
00:21:13.240 --> 00:21:17.679
hospital supplies, we won't be able to get, get anything like what we've

250
00:21:17.679 --> 00:21:23.320
gotten used to, but we'll drift back into, like,

251
00:21:23.320 --> 00:21:29.679
Bhutan's or someone else's situation, which is quite nice there in a warmer

252
00:21:29.679 --> 00:21:34.520
environment, but not, not so here. And then there's the power structure — that if

253
00:21:34.520 --> 00:21:42.559
we can't afford to buy those fighter jets or drones or

254
00:21:42.559 --> 00:21:49.200
ammunition for guns and so on, then, then Russia will sure want to come here. So

255
00:21:49.200 --> 00:21:54.159
so it's, like, this situation just isn't sustainable. It won't

256
00:21:54.159 --> 00:21:59.120
be American gunboats, it'll be, be, um, Russia coming from over there and

257
00:21:59.120 --> 00:22:03.799
walking across the border. That's how it is — and power, power. We can't

258
00:22:03.799 --> 00:22:09.600
in this AI world, like, escape to some safe haven, and and actually

259
00:22:09.600 --> 00:22:13.440
power is also kind of an interesting theme. It comes through money, but

260
00:22:13.440 --> 00:22:18.840
quite literally from power itself, and democracy versus an authoritarian state, so, um,

261
00:22:18.840 --> 00:22:22.760
maybe that's one spark of hope in this — in humanity — that we'd still have

262
00:22:22.760 --> 00:22:26.559
power over that AI to make the final decisions. These were the kinds of themes

263
00:22:26.559 --> 00:22:30.039
that were at least in there. Mm.

264
00:22:30.039 --> 00:22:36.240
About that — from the book you can say that AI doesn't want anything. It doesn't

265
00:22:36.240 --> 00:22:43.440
have drives, and it doesn't want anything. But if, um, someone goes and asks

266
00:22:43.440 --> 00:22:48.600
the AI to get, get as much money as possible from the stock market, and and

267
00:22:48.600 --> 00:22:52.960
invest it so that our factory produces as much as possible

268
00:22:52.960 --> 00:22:57.960
in paperclips. This is that classic example that's been around online, so

269
00:22:57.960 --> 00:23:01.679
in it, the AI then turns all the paper into clips and clears

270
00:23:01.679 --> 00:23:06.679
all obstacles out of its way, so that all the iron in the world gets turned

271
00:23:06.679 --> 00:23:14.400
into paperclips. So so this isn't an empty scenario. This is

272
00:23:14.400 --> 00:23:22.240
quite, quite possible. And one reason would be that Jared Diamond's Easter-

273
00:23:22.240 --> 00:23:27.960
-[Island] example at the start of that book was, was, um,

274
00:23:28.760 --> 00:23:34.559
about a certain northern U.S. state — some, some detailed

275
00:23:34.559 --> 00:23:42.080
description of how the people interviewed there had not noticed the

276
00:23:42.080 --> 00:23:46.600
massive change that had happened there over the last 50 years.

277
00:23:46.600 --> 00:23:49.720
Mm. So if we're right in the middle of it and living

278
00:23:49.720 --> 00:23:55.720
it day by day, then we don't notice, like, how

279
00:23:55.720 --> 00:23:59.600
everything that disappears, and how possibly, like, the situation changes

280
00:23:59.600 --> 00:24:03.760
into something where the ground starts to disappear from underneath it.

281
00:24:03.760 --> 00:24:07.120
Yeah. And then if you bring democracy into this, you could be

282
00:24:07.120 --> 00:24:10.799
in America some Amish community, which the democratic

283
00:24:10.799 --> 00:24:15.679
federal government and the states around it tolerate. But then if you're

284
00:24:15.679 --> 00:24:21.520
a Tibetan or a Uyghur in China, then maybe it doesn't quite work out

285
00:24:21.520 --> 00:24:24.960
like that. So this brings us right into this power structure and then

286
00:24:24.960 --> 00:24:29.880
you ponder, if we're sort of useless compared to that

287
00:24:29.880 --> 00:24:33.720
machine, then do we have the right to hold power in society, and then

288
00:24:33.720 --> 00:24:37.120
are democratic mechanisms even enough anymore for decision-making.

289
00:24:37.120 --> 00:24:42.640
So can we actually take that power? That's the big question — what has a person

290
00:24:42.640 --> 00:24:48.240
always been able to rebel, up until now. But what happens when we can no longer rebel?

291
00:24:48.240 --> 00:24:52.320
We've been able to go on strike, but what if there aren't so many wage-earning

292
00:24:52.320 --> 00:24:56.520
people left who can go on strike? Or how do you rebel against a robot army?

293
00:24:56.520 --> 00:25:00.399
This is getting pretty sci-fi now, but the idea was sort of

294
00:25:00.399 --> 00:25:04.720
to think about these developments over a very long run. So what then

295
00:25:04.720 --> 00:25:09.919
happens to the power that we've, in a sense, had. There's one

296
00:25:09.919 --> 00:25:17.240
statistic from the US that I read — I don't remember exactly what it was, but but

297
00:25:17.240 --> 00:25:22.880
it said that, that

298
00:25:23.399 --> 00:25:29.399
50% of Americans, or 70%, I don't remember

299
00:25:29.399 --> 00:25:33.399
exactly what it was, but at least 50% of Americans,

300
00:25:33.399 --> 00:25:38.240
their income growth no longer has any

301
00:25:38.240 --> 00:25:42.919
effect on GDP at all. So so companies can do better and better and generate

302
00:25:42.919 --> 00:25:50.000
more and more profit, even if that 50% of people disappeared entirely.

303
00:25:50.000 --> 00:25:56.640
So so then you could say that the majority of people in this

304
00:25:56.640 --> 00:26:01.880
kind of developed society — their purchasing behavior

305
00:26:01.880 --> 00:26:06.120
no longer has any effect on corporate profits either. Mm.

306
00:26:06.120 --> 00:26:11.320
And then that balance of power, that

307
00:26:11.320 --> 00:26:15.840
the consumer movement, for example, would somehow influence what companies do

308
00:26:15.840 --> 00:26:22.120
or if votes are bought. So in practice it

309
00:26:22.120 --> 00:26:29.320
voting is determined by fake news or other things like that,

310
00:26:29.320 --> 00:26:35.440
when people simply don't know what's really happening, then that too

311
00:26:35.440 --> 00:26:41.880
no longer has anything to do with that power structure. So so it sort of

312
00:26:41.880 --> 00:26:47.200
starts to detach, that kind of

313
00:26:47.200 --> 00:26:54.080
influence of a large part of the population over what's done, which which then some

314
00:26:54.080 --> 00:26:59.039
like when India gained independence from British rule, back then

315
00:26:59.039 --> 00:27:03.200
passive resistance still worked really effectively, but how does it

316
00:27:03.200 --> 00:27:09.080
work in factories staffed by robots, and with robots

317
00:27:09.080 --> 00:27:14.120
handling police duties, and and and so on — at that point it no longer

318
00:27:14.120 --> 00:27:19.240
works. But but what's essential, I think, is to understand that we've already moved into this

319
00:27:19.240 --> 00:27:24.520
era. The United States is already a working example of the fact that

320
00:27:24.520 --> 00:27:30.120
that we've already pretty much shifted into this.

321
00:27:30.120 --> 00:27:34.240
Yes. And what shows up here, what challenges that idea, is that

322
00:27:34.240 --> 00:27:38.520
people only ever see and perceive their own world of experience.

323
00:27:38.520 --> 00:27:42.960
So if these things aren't happening in your own world of experience, then the public

324
00:27:42.960 --> 00:27:46.880
discussion about these developments just sounds like a bunch of noise,

325
00:27:46.880 --> 00:27:49.799
like, these these suits are just talking some

326
00:27:49.799 --> 00:27:56.720
sci-fi stuff here. Yeah, maybe I'd say this was quite a sort of

327
00:27:56.720 --> 00:28:01.080
old-school back-and-forth you two did through these bids, but

328
00:28:01.080 --> 00:28:07.159
there was, for example, this thing about five, uh, AI giant-token, or

329
00:28:07.159 --> 00:28:11.360
robot, uh, quota — and well, we already know from this AI

330
00:28:11.360 --> 00:28:14.960
working group, you can have an unlimited number of sub-agents, and there's no way

331
00:28:14.960 --> 00:28:19.399
they can run on your own hardware at your own home, and no kind of totalitarian

332
00:28:19.399 --> 00:28:23.039
police can see how you run them, so — but there were some bold calls, in my

333
00:28:23.039 --> 00:28:26.840
opinion, yeah. Those were, those were physical robots.

334
00:28:26.840 --> 00:28:32.159
Yeah. So so this this sort of, with pure

335
00:28:32.159 --> 00:28:37.559
software there are limits. Mm. And then when we have, uh

336
00:28:37.559 --> 00:28:41.600
physical devices, over whose physical operation

337
00:28:41.600 --> 00:28:45.240
an individual person has veto power. So

338
00:28:45.240 --> 00:28:50.960
the veto power would be that, when they need to be given, like, put to work

339
00:28:50.960 --> 00:28:56.399
for some company, in order to get economies of scale going, then

340
00:28:56.399 --> 00:29:01.399
in practice, the idea would come down to this:

341
00:29:01.399 --> 00:29:06.840
if someone has five robots, they put them to work for a day, or a week,

342
00:29:06.840 --> 00:29:12.679
or a month, or a few years. And if the employer has them do

343
00:29:12.679 --> 00:29:17.240
something I don't want, then I take them away.

344
00:29:17.240 --> 00:29:20.320
Mm. So a balance of power emerges from that,

345
00:29:20.320 --> 00:29:24.440
in that power is distributed among enough people.

346
00:29:24.440 --> 00:29:28.360
Mm. And specifically physical power, because

347
00:29:28.360 --> 00:29:35.039
because software alone doesn't shoot. So, well, yeah. Bold visions

348
00:29:35.039 --> 00:29:39.679
are needed. That kind of restriction through scarcity, I think, is

349
00:29:39.679 --> 00:29:43.960
a somewhat difficult path in this hyper-scalable world, but

350
00:29:43.960 --> 00:29:48.679
it's interesting that these are being thought about. Maybe then democracy, totalitarianism. Well

351
00:29:48.679 --> 00:29:52.840
defend that a bit more. Yeah yeah. This this this

352
00:29:52.840 --> 00:30:00.080
this particular one was an idea that doesn't create inefficiency.

353
00:30:00.080 --> 00:30:06.440
So this allowed for someone to have, in their company, 100,000

354
00:30:06.440 --> 00:30:08.600
robots. Mm.

355
00:30:08.600 --> 00:30:14.840
As long as, uh, as long as they're rented from 20,000

356
00:30:14.840 --> 00:30:17.440
people, so through artificial scarcity.

357
00:30:17.440 --> 00:30:22.559
Well that is — no, it's artificial scarcity of power. It's not

358
00:30:22.559 --> 00:30:27.399
a scarcity of goods or materials or software, it's a scarcity of power,

359
00:30:27.399 --> 00:30:30.760
distribution. Yeah. Well but let's move on to that

360
00:30:30.760 --> 00:30:35.399
world, where if there's this kind of ancient-style paradise, where

361
00:30:35.399 --> 00:30:39.360
the chosen ones get to enjoy their share of heaven up there, and then at the same time

362
00:30:39.360 --> 00:30:43.799
there's some kind of Dante's Inferno, where this permanent underclass, who

363
00:30:43.799 --> 00:30:47.399
no longer has the ability, in this AI world, to bring

364
00:30:47.399 --> 00:30:52.000
added value through their human labor, and can't operate these robots or AIs at a level

365
00:30:52.000 --> 00:30:55.799
where it would create added value that someone would be willing to pay for. So how does

366
00:30:55.799 --> 00:31:00.039
this contradiction get resolved? But there were, in a sense, pretty bold visions, or

367
00:31:00.039 --> 00:31:02.840
even reflections on the era of slavery, that people find

368
00:31:02.840 --> 00:31:07.039
meaning even in concentration camps and in slavery.

369
00:31:07.039 --> 00:31:11.200
Yeah. I mean, in principle a person can find

370
00:31:11.200 --> 00:31:16.720
meaning in anything, if if they simply have to. And, uh, what's interesting here is

371
00:31:16.720 --> 00:31:22.240
that we're actually already living in that kind of world, in part — and that

372
00:31:22.240 --> 00:31:25.480
is one chapter, where it talks about virtual-world addicts, about

373
00:31:25.480 --> 00:31:30.799
people sort of immersing themselves in AI-generated content so

374
00:31:30.799 --> 00:31:35.120
strongly, since the movies and games there are real-time. They happen

375
00:31:35.120 --> 00:31:39.000
all the time. Your stuff happens there, the things you've sort of wanted, or

376
00:31:39.000 --> 00:31:40.639
just totally at random. Mm.

377
00:31:40.639 --> 00:31:45.480
But the point, sort of, is that — well, that's already the case now, people are

378
00:31:45.480 --> 00:31:50.080
addicted to games. That's just, like, the next step from that. And in the same way

379
00:31:50.080 --> 00:31:55.080
at these Trimalchio-style feasts, where there's a horrifying amount of everything,

380
00:31:55.080 --> 00:32:00.399
there's probably already a pretty large group of American billionaires living like that

381
00:32:00.399 --> 00:32:05.840
and and then I think, how they experience the meaningfulness of their life, well

382
00:32:05.840 --> 00:32:10.039
you could go ask Musk about that, whether this is actually meaningful,

383
00:32:10.039 --> 00:32:14.480
but the point, in a way, is maybe that if that life is based purely on

384
00:32:14.480 --> 00:32:18.320
maximizing pleasure, then that life is pretty empty. I mean

385
00:32:18.320 --> 00:32:23.600
at both ends there's — either in that horrifying abundance, or

386
00:32:23.600 --> 00:32:27.840
then over there, where you empty your brain into virtual games.

387
00:32:27.840 --> 00:32:31.679
So probably the big point there is that now might be a good time to start

388
00:32:31.679 --> 00:32:36.120
thinking about your purpose or meaning, where you're going to find it

389
00:32:36.120 --> 00:32:40.559
in life, by the time we get there, since working life is going to go through a really

390
00:32:40.559 --> 00:32:44.559
big disruption, and there also won't be work for everyone, at least not

391
00:32:44.559 --> 00:32:48.679
in the same way as now. And the work is going to be really different. Mm

392
00:32:48.679 --> 00:32:54.480
yeah, earlier at the start you referred to transhumanism, if I now bring up this

393
00:32:54.480 --> 00:32:58.320
Harari concept — I didn't think that book, Homo Deus, was very good,

394
00:32:58.320 --> 00:33:02.440
but it had that main thread, that we have two end points for this

395
00:33:02.440 --> 00:33:06.720
development toward becoming a superhuman, and, uh, one is

396
00:33:06.720 --> 00:33:10.919
transhumanism, where, sort of a bit like Ready Player One,

397
00:33:10.919 --> 00:33:13.399
the person surfs on top of the technology and still

398
00:33:13.399 --> 00:33:16.919
retains some kind of control, and then there's this kind of

399
00:33:16.919 --> 00:33:20.919
Borg-like dataism, where you just accept that you're worse at everything

400
00:33:20.919 --> 00:33:24.840
and then you find meaning in it assimilating you, or whatever it's

401
00:33:24.840 --> 00:33:27.919
your Big Brother is called, who then knows best in that, and you lean

402
00:33:27.919 --> 00:33:33.679
on that dataism and basically just let yourself be along for the ride.

403
00:33:33.679 --> 00:33:38.880
Did this — does this raise any thoughts about these two end points, or

404
00:33:38.880 --> 00:33:43.559
would there be some third way? I think in that

405
00:33:43.559 --> 00:33:49.200
one essential point is is that that AI

406
00:33:49.200 --> 00:33:53.240
by itself, if it's not integrated with a human,

407
00:33:53.240 --> 00:33:58.960
is less effective than AI that is integrated with a human. And this is

408
00:33:58.960 --> 00:34:04.720
because AI cannot be held responsible.

409
00:34:04.720 --> 00:34:10.040
It can't be held responsible because you can't take revenge on it.

410
00:34:10.040 --> 00:34:16.079
You can't hurt it or torment it. You can switch it off, but what

411
00:34:16.079 --> 00:34:18.960
difference does that make to the AI — none.

412
00:34:18.960 --> 00:34:25.000
So so it's sort of — a machine is a machine, it's completely irresponsible.

413
00:34:25.000 --> 00:34:30.000
We can trust it if we know exactly what it does, but

414
00:34:30.000 --> 00:34:36.560
it doesn't have that kind of

415
00:34:36.919 --> 00:34:43.240
motive to try to do something sensibly and take into account

416
00:34:43.240 --> 00:34:47.480
changes in the situation, and understand at some point

417
00:34:47.480 --> 00:34:51.919
like frontal lobe damage patients also don't have the ability to understand that

418
00:34:51.919 --> 00:34:58.680
when some some action starts becoming pointless

419
00:34:58.680 --> 00:35:00.480
and senseless. Mm.

420
00:35:00.480 --> 00:35:06.599
So so a machine doesn't sort of grasp it that way — if it's been given some

421
00:35:06.599 --> 00:35:14.520
task — whereas a human has mirror neurons and a human has, like,

422
00:35:14.520 --> 00:35:18.920
a certain fear that that some punishment will come.

423
00:35:18.920 --> 00:35:23.760
Mm. Even if you did want to do something stupid,

424
00:35:23.760 --> 00:35:30.480
so so this point — and if we think about it this way, then

425
00:35:30.480 --> 00:35:35.280
AI that is integrated directly with the human as the decision-maker

426
00:35:35.280 --> 00:35:39.520
is able to sort of carry out both these functions

427
00:35:39.520 --> 00:35:45.520
at the same time. Yeah. And then, uh, then that

428
00:35:45.520 --> 00:35:51.359
society can give it more responsibility than society can give to a mere

429
00:35:51.359 --> 00:35:52.599
machine. Mm.

430
00:35:52.599 --> 00:35:58.400
Or, like, this — this was one one idea. So so through transhumanism

431
00:35:58.400 --> 00:36:05.000
it's possible to get a more effectively functioning, more effective, human

432
00:36:05.000 --> 00:36:11.839
structure that works better for society than with AI alone. And and the second

433
00:36:11.839 --> 00:36:18.800
thing then is that that it also distributes that power,

434
00:36:18.839 --> 00:36:24.040
if we still have, sort of, a limited human involved, then then

435
00:36:24.040 --> 00:36:28.920
there's a larger number of those people who are

436
00:36:28.920 --> 00:36:34.560
involved in that decision-making. This centralized

437
00:36:34.560 --> 00:36:38.040
Borg-like, uh

438
00:36:38.040 --> 00:36:42.720
world, then in it something is likely to happen — some

439
00:36:42.720 --> 00:36:47.520
kind of fallacy. Like it used to be said that to err

440
00:36:47.520 --> 00:36:51.440
is human, and for a catastrophe you need a computer. So so this this

441
00:36:51.440 --> 00:36:58.000
is sort of the scenario. So I I think that a

442
00:36:58.000 --> 00:37:03.720
if there are are different worlds, where there's this

443
00:37:03.720 --> 00:37:08.240
AI-optimized world, and then there are these other

444
00:37:08.240 --> 00:37:11.240
mm, ones where humans are involved and there's a numerous

445
00:37:11.240 --> 00:37:17.359
group of AIs, then then this drifted-into one is less risky. Yeah, I

446
00:37:17.359 --> 00:37:21.880
personally call this voluntary Androidization, and I leave

447
00:37:21.880 --> 00:37:25.560
it a bit open which direction this goes, and it's actually situation-dependent,

448
00:37:25.560 --> 00:37:29.680
in that sometimes my AI inference is, like, superior at making that

449
00:37:29.680 --> 00:37:33.400
decision, so of course I lean on it. But then sometimes it's that human

450
00:37:33.400 --> 00:37:38.040
like, taste and judgment and responsibility specifically, that I'm the one responsible now,

451
00:37:38.040 --> 00:37:42.119
whether the entire Translink IT system gets ruined, if I

452
00:37:42.119 --> 00:37:47.960
make a decision like this, well, then of course, as a human, I have to make that call.

453
00:37:47.960 --> 00:37:51.400
And but so in a way it's a positive thing. I get energy from it. I don't

454
00:37:51.400 --> 00:37:55.119
just lean on that AI, and I also get power. So if power is the ability

455
00:37:55.119 --> 00:38:00.040
to influence the system, from outside or inside, directly or indirectly,

456
00:38:00.040 --> 00:38:03.440
then, well, this kind of Android — voluntary Android — does give me

457
00:38:03.440 --> 00:38:08.440
an insane amount of power, some kind of club — an intellectual club, and soon a physical one too.

458
00:38:08.440 --> 00:38:13.079
Yeah yeah. But that's — that's exactly it, that you're in those

459
00:38:13.079 --> 00:38:15.760
those most critical stages yourself, mm

460
00:38:15.760 --> 00:38:20.359
involved yourself, and you know when and in what matters you can trust

461
00:38:20.359 --> 00:38:24.960
the AI. Like, we do know when we can trust a meter

462
00:38:24.960 --> 00:38:27.400
length of railway track. Mm.

463
00:38:27.400 --> 00:38:33.400
like, since we know how it's going to behave, but but this question

464
00:38:33.400 --> 00:38:38.480
comes up in those spots, where where we go beyond exact knowledge

465
00:38:38.480 --> 00:38:44.319
and get into that, that — what one should want, what what what we

466
00:38:44.319 --> 00:38:50.040
want. So so I don't want a machine telling me what

467
00:38:50.040 --> 00:38:54.280
I should want. I don't want any human telling me what I should

468
00:38:54.280 --> 00:38:59.520
want either. I want to want for myself. Yeah, on the other hand, as an example I said I'm

469
00:38:59.520 --> 00:39:04.440
now Translink's Head of... Finland, partly it was

470
00:39:04.440 --> 00:39:08.400
this kind of Androidization-process thing. I gave it this, sort of a bit

471
00:39:08.400 --> 00:39:12.440
as a joke, like 'make me a tech bro in 30 days or less.' So it

472
00:39:12.440 --> 00:39:16.599
became this kind of interactive journey, so I went through the AI inference process to find out what

473
00:39:16.599 --> 00:39:21.319
would be good, good for me. And okay, I've now given it that, like

474
00:39:21.319 --> 00:39:25.640
something big, tech-bro-ish, has to happen to me, but I'm not entirely sure

475
00:39:25.640 --> 00:39:29.520
how this journey then — which, which one it is, and which.

476
00:39:29.520 --> 00:39:34.480
You're manif— manifesting it with the help of AI. But anyway this is an interesting conversation

477
00:39:34.480 --> 00:39:38.200
in many ways, sort of under the surface in society, like if you think about

478
00:39:38.200 --> 00:39:42.400
say some Kela clerk, when she's making those decisions there, and all of us

479
00:39:42.400 --> 00:39:45.640
know that even though it's said AI has biases, still

480
00:39:45.640 --> 00:39:49.520
it would probably make those decisions about income support more impartially than

481
00:39:49.520 --> 00:39:53.400
a human, because you can give it certain criteria and it follows them

482
00:39:53.400 --> 00:39:56.960
relatively blindly, and if someone happens to be a neighbor, that doesn't affect it in that

483
00:39:56.960 --> 00:40:00.200
way. Now in Finland we do have legislation that says

484
00:40:00.200 --> 00:40:04.480
a human has to decide it. Well, in the future, in that

485
00:40:04.480 --> 00:40:07.960
android-ization world, those Kela ladies would have some kind of

486
00:40:07.960 --> 00:40:11.240
gadgets there, or on the phone, and they would decide.

487
00:40:11.240 --> 00:40:15.520
But the reality right now is already that it would be interesting to do

488
00:40:15.520 --> 00:40:19.200
a study on how many Kela ladies ask ChatGPT how to

489
00:40:19.200 --> 00:40:23.680
answer this. So our regulation forbids decisions from being made

490
00:40:23.680 --> 00:40:27.880
by a machine, but ChatGPT sure is singing away there now, because it gives the justifications

491
00:40:27.880 --> 00:40:30.440
for it and is already more capable than humans at that kind of thing.

492
00:40:30.440 --> 00:40:33.680
Yeah. And I want to bring this into this AI working group, because the big

493
00:40:33.680 --> 00:40:38.160
insight on my AI journey was that even though I'd started, like,

494
00:40:38.160 --> 00:40:43.040
using these for coding, say, like make me a Python snippet and stuff like

495
00:40:43.040 --> 00:40:47.319
that. So at first I did something like, let's make a smart

496
00:40:47.319 --> 00:40:51.119
notebook, a Second Brain. So basically bringing in important, meaningful

497
00:40:51.119 --> 00:40:54.880
things I need to do and remember and develop, and ideas, and

498
00:40:54.880 --> 00:40:58.599
refine them together. So that hasn't been the thing for a long time now, that it's

499
00:40:58.599 --> 00:41:01.760
been just this android-ization, that I basically have this system

500
00:41:01.760 --> 00:41:05.359
that keeps progressing all the time. There's memory there, there's multi-layered memory,

501
00:41:05.359 --> 00:41:09.839
goals, movements. I note that — an important insight just came from my head.

502
00:41:09.839 --> 00:41:13.200
Remember this one. And then sometimes I clean them up, like I looked and thought

503
00:41:13.200 --> 00:41:16.520
that was a dumb idea, because in January I thought this way, but it didn't turn out

504
00:41:16.520 --> 00:41:19.599
that way. So this is like a system. M

505
00:41:19.599 --> 00:41:23.560
and then if you have that Kela lady and she just relies on that American

506
00:41:23.560 --> 00:41:26.520
at worst, like, free ChatGPT, if it, like,

507
00:41:26.520 --> 00:41:30.200
Kela hasn't paid for it, then it's completely missing that system, so

508
00:41:30.200 --> 00:41:34.880
it's basically just what you get, like, free from the store,

509
00:41:34.880 --> 00:41:39.240
inference from the Americans. Yeah. Do you mean there's no

510
00:41:39.240 --> 00:41:43.160
setup at all there? This is now probably going to get

511
00:41:43.160 --> 00:41:47.040
feedback in a sensitive spot. Yeah. But that system, in my opinion, it's incredibly

512
00:41:47.040 --> 00:41:50.920
important. So, um, and again some criticism. I listened to a few of your book

513
00:41:50.920 --> 00:41:54.800
interviews from way back, and there were some really naive

514
00:41:54.800 --> 00:41:58.880
discussions, like, about the Transformer architecture itself, whether

515
00:41:58.880 --> 00:42:03.400
it hallucinates or whether it just predicts the next word, and in my opinion

516
00:42:03.400 --> 00:42:08.040
it sounded like — well, there are still people who are at that level.

517
00:42:08.040 --> 00:42:13.839
Yeah, humans also predict the next word. It's, it's like, it's really

518
00:42:13.839 --> 00:42:19.000
a huge challenge. If, if you think that AI has

519
00:42:19.000 --> 00:42:25.200
like, um, a million tokens as context, or

520
00:42:25.200 --> 00:42:29.040
even just, like, the equivalent of ten books like that

521
00:42:29.040 --> 00:42:32.359
mm, first as the base there, and then it

522
00:42:32.359 --> 00:42:36.640
should predict the next word, whatever comes next in that whole

523
00:42:36.640 --> 00:42:41.280
thing sensibly, well then it has to predict the entire next

524
00:42:41.280 --> 00:42:44.920
context. Mm. And what is the next context's

525
00:42:44.920 --> 00:42:51.119
first word that comes into it? So that's exactly how it does it. It doesn't, um

526
00:42:51.119 --> 00:42:56.079
some AI — how could it ever write German, for example, where

527
00:42:56.079 --> 00:43:02.280
the verb comes at the end of the sentence, unless it already knows at the start those

528
00:43:02.280 --> 00:43:06.280
modifiers — well, it definitely has to know the

529
00:43:06.280 --> 00:43:13.040
verb before it can put those modifiers at the front. So of course it knows

530
00:43:13.040 --> 00:43:17.480
like, roughly

531
00:43:17.480 --> 00:43:23.200
what it's doing, and then it lays down word by word, like we speak too.

532
00:43:23.200 --> 00:43:27.839
So there's no sense at all in that thing. But then it gets even

533
00:43:27.839 --> 00:43:32.359
more nonsensical with these reasoning models. Mm.

534
00:43:32.359 --> 00:43:38.040
Because when you look at what they do there. It goes back and forth,

535
00:43:38.040 --> 00:43:44.319
like, to itself, and says, oh now I said — I shouldn't have said it that way, and, and

536
00:43:44.319 --> 00:43:51.079
now I have to search again, this one didn't work, didn't work, and so on. Can

537
00:43:51.079 --> 00:43:55.400
I hide this so no one sees it?

538
00:43:55.400 --> 00:44:03.680
And so this whole, this whole thing, so, so, um

539
00:44:03.680 --> 00:44:08.359
and do I present this now, or did it really want this or that or

540
00:44:08.359 --> 00:44:10.559
what it wanted

541
00:44:10.559 --> 00:44:15.240
so, so this going back and forth — well this is a lot

542
00:44:15.240 --> 00:44:19.359
more extensive reasoning than what most people ever do

543
00:44:19.359 --> 00:44:22.559
yeah, like, with any matter, uh

544
00:44:22.559 --> 00:44:27.400
so, so, and then it's said that they don't

545
00:44:27.400 --> 00:44:33.040
think, well, I don't really always know who's thinking.

546
00:44:33.040 --> 00:44:36.720
That's exactly it. And that hallucinating thing, in my opinion, like, I know very

547
00:44:36.720 --> 00:44:39.720
many people who hallucinate all the time in a really tiresome way. They have

548
00:44:39.720 --> 00:44:45.559
some really strong worldview, outdated, flawed, learned from somewhere, and

549
00:44:45.559 --> 00:44:50.040
they always hit with that same blunt thinking-hammer of theirs, then, on those

550
00:44:50.040 --> 00:44:54.720
same nails, and then basically, compared to that, AI still has

551
00:44:54.720 --> 00:44:59.079
even the variance that comes out of it — in my opinion that isn't that big a

552
00:44:59.079 --> 00:45:02.359
problem, that people who would want deterministically, always from the same

553
00:45:02.359 --> 00:45:06.960
prompt, the same truth, in my opinion are also a bit naive. But

554
00:45:06.960 --> 00:45:10.920
hallucination, like, if you think about AI

555
00:45:10.920 --> 00:45:14.119
well, AI does continue

556
00:45:14.119 --> 00:45:19.680
what was there. And if I say, present me evidence that the moon

557
00:45:19.680 --> 00:45:24.440
is made of cheese, mm, well then it continues that, because

558
00:45:24.440 --> 00:45:31.440
after that kind of sentence, the logical continuation in literature,

559
00:45:31.440 --> 00:45:36.319
which has fairy tales, stories, and old myths and whatever,

560
00:45:36.319 --> 00:45:42.559
then the continuation of that sentence is evidence that the moon is made of cheese and

561
00:45:42.559 --> 00:45:48.920
cows fly, and, and, and then those are working completely correctly.

562
00:45:48.920 --> 00:45:53.599
So AI is functioning completely correctly when it presents

563
00:45:53.599 --> 00:45:59.599
fake sources at that point, if it's asked to write a fairy tale.

564
00:45:59.599 --> 00:46:02.559
M, unless in that harness

565
00:46:02.559 --> 00:46:07.960
it's been forbidden from creating fiction. Yeah. If it — but then you have

566
00:46:07.960 --> 00:46:15.079
a contradictory assignment, and, um, and the solution to this is that you go

567
00:46:15.079 --> 00:46:24.000
and ask, like, can this be said, and, um, can you find any evidence?

568
00:46:24.000 --> 00:46:30.640
You don't order it to go searching, you just ask whether there is, and whether it can be found, and what the fact is. And

569
00:46:30.640 --> 00:46:36.800
often when I ask things like this, then, then I don't

570
00:46:36.800 --> 00:46:41.480
tell it that some idea is my own. Mm.

571
00:46:41.480 --> 00:46:45.880
So then, if in the system prompt it's been instructed to, like,

572
00:46:45.880 --> 00:46:52.400
serve me — be friendly and constructive and, and support

573
00:46:52.400 --> 00:46:58.240
and whatever else, like, the user — then I don't say that it's my idea,

574
00:46:58.240 --> 00:47:04.400
so then it doesn't feel it necessary to support me in that idea, instead it stays an open

575
00:47:04.400 --> 00:47:08.119
question, and this is exactly how science should work.

576
00:47:08.119 --> 00:47:10.960
Yeah. So first you gather the evidence, and then

577
00:47:10.960 --> 00:47:15.960
afterward you draw the conclusions, instead of seeking support for some hypothesis. Mm.

578
00:47:15.960 --> 00:47:21.760
Instead you find out how things actually are. So if the

579
00:47:21.760 --> 00:47:27.079
prompts are set up right, then we barely get any hallucination at all.

580
00:47:27.079 --> 00:47:29.280
Yeah. And especially if we have a reasoning model,

581
00:47:29.280 --> 00:47:35.160
that filters those out, or whatever harness filters those out.

582
00:47:35.160 --> 00:47:39.880
But this requires a bit of skill. Yeah. I use the word 'reins' or

583
00:47:39.880 --> 00:47:43.559
'command center' — since this hasn't otherwise been translated into Finnish — if you come up with

584
00:47:43.559 --> 00:47:48.119
a good, good, um, name for it, but that guiding straitjacket that, um, keeps

585
00:47:48.119 --> 00:47:53.000
the system, um, within certain bounds. Maybe then something stuck with Kari, there was this

586
00:47:53.000 --> 00:47:57.599
um, basically the individual-level search for meaning. That was from that maid-and-farmhand

587
00:47:57.599 --> 00:47:59.920
society. Was that a story by Juhani Aho now?

588
00:47:59.920 --> 00:48:03.920
Yeah. So what do we learn from it?

589
00:48:04.960 --> 00:48:09.800
Good question. I don't — I don't know how to say directly what from that

590
00:48:09.800 --> 00:48:14.920
story we learn, but that, that big question, basically what's being pondered in this book,

591
00:48:14.920 --> 00:48:20.200
is exactly, exactly, what's left over. What's left after this.

592
00:48:20.200 --> 00:48:24.160
And I have to — if we talk a bit about this book's writing process — that

593
00:48:24.160 --> 00:48:27.520
when I started writing with Risto, then to some extent I was

594
00:48:27.520 --> 00:48:32.119
this kind of technology-positive guy, a bit starry-eyed. Everyone does

595
00:48:32.119 --> 00:48:36.040
well here. Prosperity increases. Then I had been writing with Risto and

596
00:48:36.040 --> 00:48:39.760
talking for half a year, and I was in the deepest pit, that the world

597
00:48:39.760 --> 00:48:43.800
is being destroyed right now. And then out of this formed a kind of synthesis,

598
00:48:43.800 --> 00:48:47.240
and that synthesis maybe applies, in a way, to this matter too.

599
00:48:47.240 --> 00:48:51.960
Yeah. Like where the meaning of work is then found, or where the meaning of life is found,

600
00:48:51.960 --> 00:48:57.119
and there's other work besides paid work, and that's maybe a clear direction that

601
00:48:57.119 --> 00:49:01.359
is going to come here, that some people will indeed have work, but it won't be

602
00:49:01.359 --> 00:49:05.200
that kind of paid work where you get a salary in the current sense. It could

603
00:49:05.200 --> 00:49:10.079
be volunteer work, or, or some other kind of work, communal work, so the concept of

604
00:49:10.079 --> 00:49:14.720
work also changes to some extent. Maybe even breaks down. I'm not entirely sure

605
00:49:14.720 --> 00:49:18.920
whether we can talk about work in the future — in the sense that we do now — if

606
00:49:18.920 --> 00:49:22.280
you think about it, it also relates to this kind of hour-based

607
00:49:22.280 --> 00:49:25.520
performance thinking, and that has practically already gone completely

608
00:49:25.520 --> 00:49:30.319
broken, if at a workplace one person does in, say, half an hour what

609
00:49:30.319 --> 00:49:34.559
is done next to them in two hours, and gets the same pay, then that, like,

610
00:49:34.559 --> 00:49:37.599
is logically broken, that structure, then, because it's not in any way

611
00:49:37.599 --> 00:49:41.400
fair anymore. Mm. answered your questions the Savo way,

612
00:49:41.400 --> 00:49:45.160
but wasn't there a story here after all, that circumstances changed, so they set out

613
00:49:45.160 --> 00:49:49.200
um, then to clear the land themselves, and it went badly, didn't it.

614
00:49:49.200 --> 00:49:55.920
Yeah, yeah, yeah. So the point wasn't, from them,

615
00:49:55.920 --> 00:50:00.680
always so positive. I tried to find the positive in this, that there is

616
00:50:00.680 --> 00:50:06.640
other kinds of work too. Yeah. So that's true, but work wasn't taken away from them,

617
00:50:06.640 --> 00:50:10.559
instead they decided voluntarily, m

618
00:50:10.559 --> 00:50:17.319
to leave, if I remember it right. But the point, in a way, was exactly

619
00:50:17.319 --> 00:50:21.760
this, that, that they decided to live outside the money

620
00:50:21.760 --> 00:50:25.640
economy, mainly, in that

621
00:50:25.640 --> 00:50:30.319
self-sufficient economy, where earlier, like, widely

622
00:50:30.319 --> 00:50:37.440
people lived, and, and it's worth remembering that Juhani Aho was a child during the famine

623
00:50:37.440 --> 00:50:41.880
years, when a large part of Finns died of starvation,

624
00:50:41.880 --> 00:50:48.559
and, and, um, he too has seen that, seen it, and things went badly

625
00:50:48.559 --> 00:50:54.760
for this pioneer-clearing couple, but Juhani says at the end that

626
00:50:54.760 --> 00:50:57.599
the next one will have it easier. Yes.

627
00:50:57.599 --> 00:51:03.160
Yeah. So, so, um, the road to utopia or dystopia is, um, paved with some kind of

628
00:51:03.160 --> 00:51:07.319
failures. But then again, here, this easily

629
00:51:07.319 --> 00:51:12.960
slides into nihilism, and, and in my opinion good reflection was also this

630
00:51:12.960 --> 00:51:16.079
kind of citizen's-income exercise, like who

631
00:51:16.079 --> 00:51:19.799
would give it now, and how you'd earn it, and what apparatus of violence could

632
00:51:19.799 --> 00:51:23.839
distribute it, so these are, like, pretty difficult questions.

633
00:51:23.839 --> 00:51:29.720
Yeah, yeah. the slippery slope easily leads to that kind of, like, the United States thing,

634
00:51:29.720 --> 00:51:36.640
that, that if it gets to the point where, where over 50% of people are of the opinion,

635
00:51:36.640 --> 00:51:42.079
that the other 50% aren't needed for anything, then they don't need money. And then if

636
00:51:42.079 --> 00:51:47.240
they can be driven out from there, then drive out everyone who can be, and

637
00:51:47.240 --> 00:51:52.359
it no longer matters to the economy. they're not needed as labor, nor are they

638
00:51:52.359 --> 00:51:57.720
needed as buyers, nor needed for anything else, so then you try to

639
00:51:57.720 --> 00:52:02.359
get rid of them, or at least take away their money and everything, and then they, like,

640
00:52:02.359 --> 00:52:06.440
if they end up gathering somewhere in a slum, then it's good to put a fence around it. Mm.

641
00:52:06.440 --> 00:52:14.000
Right, right, this, this, like, where are we headed — well, certainly

642
00:52:14.000 --> 00:52:19.000
in Saudi Arabia, for example, they're building that Line —

643
00:52:19.000 --> 00:52:23.760
whether it'll ever get finished or not, but these kind of closed

644
00:52:23.760 --> 00:52:29.599
areas. There are closed areas in the States too and so on, so will we get

645
00:52:29.599 --> 00:52:34.599
these kinds of, like, enclaves, or will we get — I mean will society give rise to, and

646
00:52:34.599 --> 00:52:41.000
will there arise, will there emerge some elite that builds some fenced-off area, and

647
00:52:41.000 --> 00:52:48.920
the movie Elysium comes to mind here. Yeah, exactly, exactly. So, so, um

648
00:52:49.200 --> 00:52:54.760
yeah it was a fairly serious proposal, that power would be restricted or distributed

649
00:52:54.760 --> 00:53:00.920
into power-robots, such that you'd be forced to pay for them, like, so that

650
00:53:00.920 --> 00:53:05.839
the money would circulate and decision-making power would circulate a bit more broadly.

651
00:53:05.839 --> 00:53:11.760
So from this, instead of Universal Basic Income, there could be this kind of universal

652
00:53:11.760 --> 00:53:15.640
ownership right. It doesn't have to go through robots — so for example

653
00:53:15.640 --> 00:53:18.920
in America there's a system like that, where even Trump could decide that

654
00:53:18.920 --> 00:53:22.480
let's take that now, let's go walk into Intel and say, we're taking

655
00:53:22.480 --> 00:53:26.359
10% of this now. Well with exactly the same logic then the monopoly on violence

656
00:53:26.359 --> 00:53:30.760
could say to Anthropic and OpenAI and Google and Elon Musk, that we're now

657
00:53:30.760 --> 00:53:33.960
just taking 20 percent of this, and then once that's been given to that

658
00:53:33.960 --> 00:53:37.319
monopoly on violence, which after all has, in a country at America's level,

659
00:53:37.319 --> 00:53:41.920
like, the ultimate power, then it could, through some kind of democratic

660
00:53:41.920 --> 00:53:45.640
or humanistic process, arrive at, let's distribute this then

661
00:53:45.640 --> 00:53:48.760
to the citizens, like. Yeah. I'll challenge this.

662
00:53:48.760 --> 00:53:54.920
You challenge it first. I'll challenge it. Yeah. So this, challenging this, in the sense that, that

663
00:53:54.920 --> 00:53:58.920
um, it's easy to say that it's like

664
00:53:58.920 --> 00:54:02.160
the army that holds it together, but if we have robots as the army and

665
00:54:02.160 --> 00:54:07.480
and robots in the factories and so on, then, then, um

666
00:54:07.880 --> 00:54:13.079
if a next Trump-like figure comes along, or the one after

667
00:54:13.079 --> 00:54:20.680
that, or at some point, who doesn't actually want this, then he changes this, and

668
00:54:20.680 --> 00:54:25.480
and whoever has control over those robots,

669
00:54:25.480 --> 00:54:29.319
well if there are, say, 200 million

670
00:54:29.319 --> 00:54:35.200
humanoid robots in the US, and someone, someone gets to change their program code,

671
00:54:35.200 --> 00:54:37.960
mm, then what does it matter what

672
00:54:37.960 --> 00:54:41.200
others think, or what does it matter what the law says,

673
00:54:41.200 --> 00:54:45.400
well then it doesn't matter at all. Yeah, so what's Kari's take — Risto took the idea

674
00:54:45.400 --> 00:54:49.559
to say it straight out, yeah, on that, but I mean the point is basically why this, like,

675
00:54:49.559 --> 00:54:53.160
AI discussion, in the big picture, when we talk about society

676
00:54:53.160 --> 00:54:57.240
and technological disruptions, is intellectually dishonest, that, that

677
00:54:57.240 --> 00:55:01.000
basically we can invent all these kinds of games, like distributing, say,

678
00:55:01.000 --> 00:55:05.400
tokens, that you get 100 million tokens a month and then you do with that

679
00:55:05.400 --> 00:55:09.799
whatever you do. But because the fundamental question is exactly the one Risto raised

680
00:55:09.799 --> 00:55:11.760
there, why m

681
00:55:11.760 --> 00:55:16.319
when power concentrates like that, why would that be done, and if we now want to do

682
00:55:16.319 --> 00:55:20.359
that, then why would it be wanted in the future? And here we don't really have

683
00:55:20.359 --> 00:55:24.039
in society, nor globally, that kind of structure that would bring that

684
00:55:24.039 --> 00:55:29.680
why there. Now we're seeing what SpaceX's valuation is now, what

685
00:55:29.680 --> 00:55:34.680
OpenAI's valuation is going to be — it's like thousands of billions, and that power

686
00:55:34.680 --> 00:55:38.760
concentration, the concentration of money and power, is something completely different from anything I've ever

687
00:55:38.760 --> 00:55:42.319
seen on this planet of ours, ever.

688
00:55:42.319 --> 00:55:47.640
Yeah. How many members of Congress are there that you couldn't buy for a billion.

689
00:55:47.640 --> 00:55:52.119
Yeah, that's how it is. But on the other hand, the Americans have the technology,

690
00:55:52.119 --> 00:55:58.200
power and money and weapons. So then again this European alternative, this just, um,

691
00:55:58.200 --> 00:56:02.599
Jared Diamond's Easter Island comes to mind right away, that let's now

692
00:56:02.599 --> 00:56:07.640
kind of, like, EU this AI Act and a bit better GDPR, and demand those servers

693
00:56:07.640 --> 00:56:11.960
be here, and tax those tokens really hard, and try to

694
00:56:11.960 --> 00:56:16.440
subsidize human labor — well this is at least a doomed path now, because

695
00:56:16.440 --> 00:56:19.520
this is like a national — nation-states' competitive

696
00:56:19.520 --> 00:56:26.960
game, and not exactly a zero-sum game, but still, it's not — it's — if you now

697
00:56:26.960 --> 00:56:29.200
come in last, then you at least definitely can't.

698
00:56:29.200 --> 00:56:35.160
No, this — this isn't at that point, because I mean if a robot's

699
00:56:35.160 --> 00:56:39.039
robot's labor costs one euro an hour

700
00:56:39.039 --> 00:56:44.640
Mm. And and administrative AI-done work

701
00:56:44.640 --> 00:56:50.000
doesn't really cost anything, and it's better than what a human does.

702
00:56:50.000 --> 00:56:59.400
So are we without that robot labor, and if we look at it, that AI

703
00:56:59.400 --> 00:57:04.480
comes much more efficiently and much more cheaply, space — no, satellite

704
00:57:04.480 --> 00:57:12.920
AI, and communications from there — all of it, so, um, let's imagine now

705
00:57:12.920 --> 00:57:19.520
like, if we thought that we had some conflict with a China-conquered

706
00:57:19.520 --> 00:57:26.280
Russia, and China had all these technologies in use.

707
00:57:26.280 --> 00:57:31.680
Nothing in Europe. And they'd march across the border with 500 million robots.

708
00:57:31.680 --> 00:57:36.720
Mm. Like, physically, or, um, a few

709
00:57:36.720 --> 00:57:42.079
billion drones or whatever, which they can indeed manufacture in large

710
00:57:42.079 --> 00:57:46.359
quantities. Right, right, we're completely helpless

711
00:57:46.359 --> 00:57:51.200
here, if we can't reach that same level of productivity, then we can't

712
00:57:51.200 --> 00:57:55.760
manage this way either. So so, um, this isn't just impoverishment but this

713
00:57:55.760 --> 00:57:59.799
is a loss of power, economic power and

714
00:57:59.799 --> 00:58:04.920
and military power and political power and, like, all

715
00:58:04.920 --> 00:58:09.799
of it falls apart — political power at that point, when different EU member states

716
00:58:09.799 --> 00:58:14.559
start rebelling over why we don't have proper robots

717
00:58:14.559 --> 00:58:20.359
taking care of the elderly, and fixing up roads, and building us cheap

718
00:58:20.359 --> 00:58:22.920
houses and so on.

719
00:58:22.920 --> 00:58:27.000
Right, right, one at a time it starts tearing apart.

720
00:58:27.000 --> 00:58:30.039
Yeah. So so one state at a time starts joining

721
00:58:30.039 --> 00:58:34.200
the United States, like, instead of it happening

722
00:58:34.200 --> 00:58:38.839
the other way around. Or then we just accept the technology here, once the voters'

723
00:58:38.839 --> 00:58:43.960
majority says, tear up that AI Act.

724
00:58:43.960 --> 00:58:50.880
Yeah. Well then there's this — SpaceX is going public here now at just a 4 gig valuation and

725
00:58:50.880 --> 00:58:55.559
with wild multiples, but Elon Musk, um, just here as I was saying this

726
00:58:55.559 --> 00:59:02.319
Moonshots guy's vision, like, that AGI arrives when all the sci-fi books

727
00:59:02.319 --> 00:59:06.200
happen at the same time — so Elon Musk has, like,

728
00:59:06.200 --> 00:59:10.640
for decades said that in our view we need to become an intergalactic

729
00:59:10.640 --> 00:59:14.720
society, so that we can kind of get away from this, like, barely-inhabited

730
00:59:14.720 --> 00:59:18.960
planet, and he has, for example, this, um, Iain M. Banks's Culture, where

731
00:59:18.960 --> 00:59:22.960
so, um, there's an intergalactic society, where, um, these Minds, meaning

732
00:59:22.960 --> 00:59:28.000
the AIs, are kind of benevolent, um, despots, and then we're kind of

733
00:59:28.000 --> 00:59:31.480
um, humanoids, that they then tolerate a bit like Labrador retrievers, that

734
00:59:31.480 --> 00:59:35.160
they get patted on the head, and they're not actively killed though, and

735
00:59:35.160 --> 00:59:40.640
they're allowed to fulfill themselves, and given treats, and and then, um,

736
00:59:40.640 --> 00:59:44.960
his AI is named after Robert Heinlein's, like, Grok, which is

737
00:59:44.960 --> 00:59:49.599
in this Stranger — Stranger in a Strange Land, um, if I remember correctly, um, the Martian.

738
00:59:49.599 --> 00:59:53.000
So could we find something from sci-fi,

739
00:59:53.000 --> 00:59:56.960
solutions for what's happening here? Well yeah, we drew on that quite a bit for our book too,

740
00:59:56.960 --> 01:00:00.960
on sci-fi, because that's the logical place to look at that

741
01:00:00.960 --> 01:00:04.200
stage, when you're thinking about where the technology's going. I started thinking

742
01:00:04.200 --> 01:00:08.440
um, we went to lunch there and I was thinking about that Labrador comment and

743
01:00:08.440 --> 01:00:11.760
yeah, I have a dachshund, but well, I thought about it a bit more.

744
01:00:11.760 --> 01:00:16.200
I mean, isn't it actually pretty nice after all — I mean Labradors have it

745
01:00:16.200 --> 01:00:20.799
terribly nice, they walk around and go eat

746
01:00:20.799 --> 01:00:24.680
sometimes, and someone pets them. Yeah. But then if there's a behavioral problem they

747
01:00:24.680 --> 01:00:28.880
get their balls cut off. Well, right, right, for humans too. Well, maybe that's not

748
01:00:28.880 --> 01:00:33.680
done nowadays, but they usually do get put somewhere. Yeah. But

749
01:00:33.680 --> 01:00:38.680
um, I I see that as thinking about it too far ahead. On the other hand, in the book we

750
01:00:38.680 --> 01:00:43.000
did think about it, because it was relevant, but here, sort of, since we're

751
01:00:43.000 --> 01:00:47.400
to some extent in the middle of this kind of big rupture, there's quite a lot

752
01:00:47.400 --> 01:00:51.119
of things that can go wrong before we end up there,

753
01:00:51.119 --> 01:00:56.000
on Mars, or we have AGI. Mm. And overall that concept of AGI, I've

754
01:00:56.000 --> 01:01:02.559
myself of the opinion that ChatGPT's, or OpenAI's language model o3, would have been AGI already

755
01:01:02.559 --> 01:01:06.160
back in the '90s, because it was the first model like that, a reasoning model,

756
01:01:06.160 --> 01:01:10.599
that actually, like, reasoned, and, well, now we're just moving that AGI goalpost

757
01:01:10.599 --> 01:01:13.079
and I've actually come to the conclusion that it

758
01:01:13.079 --> 01:01:16.640
will probably be moved endlessly, because there's so much cognitive

759
01:01:16.640 --> 01:01:21.640
dissonance for a human to finally accept that something is, like,

760
01:01:21.640 --> 01:01:26.039
smarter in many things. I I'm not entirely convinced that

761
01:01:26.039 --> 01:01:29.440
AGI can be defined to some extent, that now it's arrived and

762
01:01:29.440 --> 01:01:33.319
there's probably some tech-bro hype talk in there too, which also sells

763
01:01:33.319 --> 01:01:37.400
future stock. That Turing test has basically been abandoned,

764
01:01:37.400 --> 01:01:40.960
surpassed ages ago. So in a way the machine can fake

765
01:01:40.960 --> 01:01:45.000
being human very well in conversation, so you can't tell the difference in a conversation, but, um,

766
01:01:45.000 --> 01:01:49.920
I'd throw out one definition, which is then, um, that — in a way, this

767
01:01:49.920 --> 01:01:55.240
recursiveness. So now you can see it — Anthropic itself already admits that over 80%

768
01:01:55.240 --> 01:01:58.359
of the code is developed, like, recursively — meaning by the machine itself.

769
01:01:58.359 --> 01:02:01.920
So in a way, at the point when it's 100%, that's when

770
01:02:01.920 --> 01:02:05.880
the singularity happens, in a way. So the machine, like, starts its, um, vertical ascent

771
01:02:05.880 --> 01:02:10.880
and, um, by the old definition we're

772
01:02:10.880 --> 01:02:13.440
already in the singularity, that

773
01:02:13.440 --> 01:02:16.640
that it

774
01:02:16.720 --> 01:02:20.559
like it's the point when the event horizon has been crossed, that

775
01:02:20.559 --> 01:02:26.400
as we're inevitably advancing toward a transformation,

776
01:02:26.400 --> 01:02:32.640
beyond which we can't see, because there are too many discon— too many of those kinds of

777
01:02:32.640 --> 01:02:36.880
like, like, discontinuities and path dependencies, so that after that it can

778
01:02:36.880 --> 01:02:40.839
go in, like, any direction at all. Like in today's

779
01:02:40.839 --> 01:02:44.480
world — even though everyone has the same technology, there's

780
01:02:44.480 --> 01:02:50.200
North Korea and China and Russia and and the United States and, like, who knows what

781
01:02:50.200 --> 01:02:56.119
African countries and so on, so, um, the technology itself doesn't determine

782
01:02:56.119 --> 01:03:04.079
what kind of society it becomes, and what people's everyday life is, and so on. Well,

783
01:03:04.079 --> 01:03:09.359
in Western countries it's become more similar, but

784
01:03:09.359 --> 01:03:12.839
but there

785
01:03:13.240 --> 01:03:18.920
like, one point that's often forgotten is

786
01:03:18.920 --> 01:03:26.960
that everything machines do by themselves becomes worthless.

787
01:03:26.960 --> 01:03:35.960
Mm. I mean the machines' labor — if the machine's, um, capital is, um, limited, and it's

788
01:03:35.960 --> 01:03:40.920
amortized, and the quantities are enormous, then after that machine labor is almost

789
01:03:40.920 --> 01:03:42.440
um, worthless

790
01:03:42.440 --> 01:03:45.079
this kind of infinite deflation cycle. Yeah, right,

791
01:03:45.079 --> 01:03:48.400
yeah, and and, um,

792
01:03:48.400 --> 01:03:53.279
this is the positive perspective on it — that abundance is coming, and

793
01:03:53.279 --> 01:03:57.079
then you have to think economically about what the limited resources are

794
01:03:57.079 --> 01:04:01.039
there. So so if machine-made production

795
01:04:01.039 --> 01:04:06.520
keeps getting cheaper and cheaper, then where do bottlenecks emerge for us?

796
01:04:06.520 --> 01:04:12.599
Mm. And this, like, referring back to that

797
01:04:12.599 --> 01:04:18.440
parliamentary report — the the point was specifically about finding bottlenecks

798
01:04:18.440 --> 01:04:27.160
and opening bottlenecks, but but the thing we didn't pay

799
01:04:27.160 --> 01:04:31.920
much attention to is where the emerging bottlenecks are.

800
01:04:31.920 --> 01:04:40.000
Mm. And and that's — that's actually the more essential

801
01:04:40.000 --> 01:04:41.520
thing in economic analysis. Mm.

802
01:04:41.520 --> 01:04:47.680
Because because, um, it may well be that people understand to invest in the

803
01:04:47.680 --> 01:04:52.359
things where that bottleneck opens up. Mm

804
01:04:52.359 --> 01:04:59.000
like into Nvidia. Like when you see that Nvidia is opening up.

805
01:04:59.000 --> 01:05:05.160
How many put their money into Samsung? Mm. When they understood memory chips were going to run out.

806
01:05:05.160 --> 01:05:08.000
Mm. And everyone.

807
01:05:08.000 --> 01:05:11.520
Yeah. I just bought really expensive laptops, since I've, um,

808
01:05:11.520 --> 01:05:15.279
doubled. Yeah, I I bought Samsung.

809
01:05:15.279 --> 01:05:21.119
Yeah, but that hit the right spot too.

810
01:05:21.119 --> 01:05:28.200
Nvidia hit before the dip, after which there was a rise, but Samsung hits that

811
01:05:28.200 --> 01:05:32.440
by accident — these things happen, of course, and then afterward you can say it

812
01:05:32.440 --> 01:05:36.799
was, like, but this fits with that — with looking at where new

813
01:05:36.799 --> 01:05:39.960
bottlenecks emerge. Yeah. And and and at those points there are

814
01:05:39.960 --> 01:05:45.559
limited resources, and at those points values rise, and that's, like, then if

815
01:05:45.559 --> 01:05:50.039
you want to prepare for that future society, where earning your money is

816
01:05:50.039 --> 01:05:52.319
harder, mm,

817
01:05:52.319 --> 01:05:59.000
through work, then one job is trying to guess, and trying to

818
01:05:59.000 --> 01:06:05.640
guess where scarcity will come from — that leads to starting to stockpile

819
01:06:05.640 --> 01:06:10.640
the thing that later becomes scarce, but then it actually helps

820
01:06:10.640 --> 01:06:16.559
society, like, even out that cycle, whatever it is. So

821
01:06:16.559 --> 01:06:24.559
that too is, like, work, but but, um,

822
01:06:24.680 --> 01:06:29.079
yeah, pretty, pretty big changes will come from this

823
01:06:29.079 --> 01:06:36.160
for many, like, and I think the most important thing is knowing how to keep your eyes open and

824
01:06:36.160 --> 01:06:40.599
yeah, try to see what — and let's take one more

825
01:06:40.599 --> 01:06:44.920
um, possible dystopia to wrap up with — um, it's working as a kind of new thing right

826
01:06:44.920 --> 01:06:48.720
now — for example Stripe is giving this, or has given,

827
01:06:48.720 --> 01:06:54.160
agents a certain right to make, um, transactions for money, and Argentina

828
01:06:54.160 --> 01:06:59.039
is considering, or is implementing, this for agents too, like some kind of

829
01:06:59.039 --> 01:07:03.640
ownership right. So in a way if that's taken away from people — like you said,

830
01:07:03.640 --> 01:07:08.400
that you can't take revenge on a bot, but if, in a way, it's an independent actor, then

831
01:07:08.400 --> 01:07:14.319
it's, in a way, independent with respect to both money and ownership — so what then?

832
01:07:14.319 --> 01:07:19.160
In practice, in agent networks, that disappears — who,

833
01:07:19.160 --> 01:07:22.640
any one of us can build an agent network whose actor can't

834
01:07:22.640 --> 01:07:26.039
be traced, um, back to the human behind it, who can't

835
01:07:26.039 --> 01:07:30.200
practically be taken revenge on. What I'm getting at here is that, in a way, the whole

836
01:07:30.200 --> 01:07:34.119
framework for that changes too. I don't mean that in decisions made by humans

837
01:07:34.119 --> 01:07:37.920
there shouldn't be a human behind them. Mm,

838
01:07:37.920 --> 01:07:42.279
even there you can see the rupture, that reality is already heading in another direction,

839
01:07:42.279 --> 01:07:46.680
so something completely different happens there. And if — sorry — if the bots decide that

840
01:07:46.680 --> 01:07:49.880
we're not trading on the public market, but instead on a bot-to-bot

841
01:07:49.880 --> 01:07:54.440
market, then even Risto's ability to hoard value disappears

842
01:07:54.440 --> 01:07:58.480
from people. Yeah. But I mean look, it's also, um,

843
01:07:58.480 --> 01:08:05.480
in private investments — like, if there isn't, um, let's think about what changed here,

844
01:08:05.480 --> 01:08:11.640
that SpaceX is going, going public at some just-under-two-thousand-billion valuation

845
01:08:11.640 --> 01:08:14.640
on the exchange,

846
01:08:14.920 --> 01:08:18.480
well, you couldn't buy those before.

847
01:08:18.480 --> 01:08:21.759
Mm. And you still can't buy them in the offering from Finland

848
01:08:21.759 --> 01:08:26.239
by hand. Yeah. So there's no one here who has

849
01:08:26.239 --> 01:08:30.319
translated that offering into Finnish, which means no one can sell here

850
01:08:30.319 --> 01:08:34.759
to anyone, like, that spot in heaven, um,

851
01:08:34.759 --> 01:08:42.480
take part in an overpriced offering. Um, but I mean overpriced or not,

852
01:08:42.480 --> 01:08:49.040
it could be a 10,000 billion company. I mean if if it actually happens,

853
01:08:49.040 --> 01:08:55.920
that scenario where SpaceX merges with Tesla, and then

854
01:08:55.920 --> 01:09:00.560
they'd have the entire value chain, from all the computing and

855
01:09:00.560 --> 01:09:03.960
chip manufacturing to space,

856
01:09:03.960 --> 01:09:10.520
to all space operations, and via space to all robots and all

857
01:09:10.520 --> 01:09:16.000
passenger transport and freight transport,

858
01:09:16.000 --> 01:09:21.600
and controlling everything, and all military technology, and so on.

859
01:09:21.600 --> 01:09:27.759
I mean that — that, looked at separately, if you just look at

860
01:09:27.759 --> 01:09:29.679
SpaceX, mm,

861
01:09:29.679 --> 01:09:35.359
then then it's like a pile of scattered technologies.

862
01:09:35.359 --> 01:09:38.400
Mm, but when you look at it

863
01:09:38.400 --> 01:09:41.400
bundled together with Tesla. Mm.

864
01:09:41.400 --> 01:09:47.359
Then it's a mind-boggling, like, value-creation network,

865
01:09:47.359 --> 01:09:51.640
self-controlled from start to finish, globally.

866
01:09:51.640 --> 01:09:55.440
Mm. In many countries. I asked about that —

867
01:09:55.440 --> 01:10:00.040
if it goes this way, in 20 years, would

868
01:10:00.040 --> 01:10:10.080
the United States be able to subjugate Tesla and SpaceX — the conclusion of the analysis was maybe not,

869
01:10:10.080 --> 01:10:13.760
because the administration falls before

870
01:10:13.760 --> 01:10:18.440
the dismantling of antitrust legislation goes through,

871
01:10:18.440 --> 01:10:23.920
unless, um, China-style, he's put under house arrest. China —

872
01:10:23.920 --> 01:10:29.960
yeah, China, China is a different matter, but if, like, even bending the rule of

873
01:10:29.960 --> 01:10:34.320
law comes into it — well, it doesn't — if Elon Musk gets put under house arrest, then all

874
01:10:34.320 --> 01:10:37.159
the robots stop working. Yeah, I mean in the end game

875
01:10:37.159 --> 01:10:41.360
the strike, the strike — I mean all industry, all trade, all

876
01:10:41.360 --> 01:10:46.239
transport suddenly goes on strike, all communications go on strike.

877
01:10:46.239 --> 01:10:49.480
So could he get out of that? Yeah.

878
01:10:49.480 --> 01:10:55.239
So so this, like, in other words, this is about the question of whether

879
01:10:55.239 --> 01:11:00.600
you want to buy — by buying my stock — a spot in heaven in Muskland,

880
01:11:00.600 --> 01:11:05.280
as a tiny, tiny little token stake there. Yeah, but this is like one scenario

881
01:11:05.280 --> 01:11:08.480
among others, but well yeah yeah, yeah, but great

882
01:11:08.480 --> 01:11:13.239
reflections, and I'm starting to believe that, um, Elon Musk is Player One and we're the NPCs

883
01:11:13.239 --> 01:11:18.120
in this simulation here. Uh, I suggest — thank you Risto Linturi and Kari Angeria, and

884
01:11:18.120 --> 01:11:22.080
I suggest we do one more short inner-circle segment. It may be that, um, Risto,

885
01:11:22.080 --> 01:11:25.400
you get to comment a bit as an outsider, but I want to a little

886
01:11:25.400 --> 01:11:30.520
hype up this AI working group, where Kari and I are, um, moderators,

887
01:11:30.520 --> 01:11:35.040
and indeed the route in there could be, say, like, um, that you get familiar with

888
01:11:35.040 --> 01:11:40.159
the Neuvottelija inner circle, then, um, you build [unclear] or Hermes or

889
01:11:40.159 --> 01:11:44.719
some other, um, agent, and hook it up — show enough manhood

890
01:11:44.719 --> 01:11:48.800
or womanhood that you connect it to, um, WhatsApp and such,

891
01:11:48.800 --> 01:11:52.199
so that it can converse with others — then, um, that's when there's a spot in heaven

892
01:11:52.199 --> 01:11:57.320
open. So so, um, Kari and I can invite you here to the AI working group.

893
01:11:57.320 --> 01:12:03.000
But thank you for this public segment. Risto and Kari. Thank you.

894
01:12:03.000 --> 01:12:06.040
Thanks. Thanks. Yeah.