WEBVTT
1
00:00:00.080 --> 00:00:02.879
Welcome to the deep dive of everybody. So today we
2
00:00:02.960 --> 00:00:06.559
are jumping into a really profound topic. We're looking at
3
00:00:06.559 --> 00:00:09.480
the trajectory of artificial intelligence, but we're doing it through
4
00:00:09.480 --> 00:00:12.320
a very specific, highly philosophical lens.
5
00:00:12.400 --> 00:00:14.560
Yeah, it's a massive subject to try and wrap your
6
00:00:14.560 --> 00:00:15.279
head around.
7
00:00:15.080 --> 00:00:18.000
It really is. And our source material today is the
8
00:00:18.440 --> 00:00:22.559
monumental Joe Rogan Experience number thirteen fifty and that one
9
00:00:22.559 --> 00:00:24.160
features Nick Bostrom, right.
10
00:00:24.280 --> 00:00:26.600
Nick Bostrom, And for those who might not know, he's
11
00:00:26.640 --> 00:00:30.160
this brilliant Swedish philosopher based to the University.
12
00:00:29.800 --> 00:00:30.800
Of Oxford exactly.
13
00:00:30.879 --> 00:00:34.399
He's just globally known for his foundational work on existential risk,
14
00:00:35.479 --> 00:00:39.640
the anthropic principle, human enhancement, ethics, and you know the
15
00:00:39.679 --> 00:00:41.640
really big one, super intelligence.
16
00:00:41.200 --> 00:00:42.799
Risks, the fun, lighthearted stuff.
17
00:00:43.039 --> 00:00:45.200
Oh yeah, definitely real casual weekend reading.
18
00:00:45.679 --> 00:00:47.799
But you know, the mission for us today is to
19
00:00:47.840 --> 00:00:50.560
act as your guides through this source material. We want
20
00:00:50.600 --> 00:00:54.399
to navigate this really fragile, terrifying but also kind of
21
00:00:54.479 --> 00:01:00.119
awe inspiring bridge between today's technological innovations and tomorrow's.
22
00:00:59.520 --> 00:01:04.840
Reality, because we are essentially exploring how we might either
23
00:01:04.920 --> 00:01:08.480
completely engineer our own destruction or maybe come to the
24
00:01:08.519 --> 00:01:12.359
realization that we are simply lines of code in a cosmic.
25
00:01:11.959 --> 00:01:15.719
Computer, which is I mean, it's a heavy subject, but
26
00:01:16.120 --> 00:01:18.879
we are going to navigate it clearly for you, no
27
00:01:19.159 --> 00:01:21.480
crazy jargon. Right. We want you to walk away with
28
00:01:21.560 --> 00:01:26.680
those profound aha moments, really understanding the mechanics of these theories.
29
00:01:26.760 --> 00:01:30.079
Yeah, to really grasp the current technological landscape we're all
30
00:01:30.079 --> 00:01:30.959
sitting in right now.
31
00:01:31.120 --> 00:01:33.879
So to set the stage for how bosstorm views are
32
00:01:33.959 --> 00:01:36.599
current era of tech, I want you to think about
33
00:01:36.640 --> 00:01:38.719
the environment of a master watchmaker.
34
00:01:38.760 --> 00:01:39.799
Okay, a watchmaker.
35
00:01:39.920 --> 00:01:43.799
Yeah, you picture them with their magnifying loop right, their tweezers,
36
00:01:44.120 --> 00:01:47.359
and they are carefully placing every microscopic gear and spring
37
00:01:47.439 --> 00:01:50.359
into the chest, well, very meticulous exactly. They know exactly
38
00:01:50.400 --> 00:01:53.519
what every single piece does. They understand the complete picture
39
00:01:53.560 --> 00:01:56.599
from the start, and the ultimate purpose is timekeeping.
40
00:01:56.760 --> 00:02:00.599
Yeah, every component serves that clear, entirely predictive blend. The
41
00:02:00.599 --> 00:02:02.439
physics are known, the boundaries.
42
00:02:02.000 --> 00:02:04.439
Are closed, the outcome is defined before they even pick
43
00:02:04.519 --> 00:02:05.799
up a tool, right.
44
00:02:05.879 --> 00:02:09.800
The intention governs the entire process. There is no mystery
45
00:02:10.199 --> 00:02:13.479
at all about what the mechanism will do once that
46
00:02:13.560 --> 00:02:14.719
final gear is in place.
47
00:02:14.960 --> 00:02:17.639
But then you take a step into the modern laboratories
48
00:02:17.639 --> 00:02:20.240
where artificial intelligence is being developed right.
49
00:02:20.120 --> 00:02:21.520
Now, and it's a totally different story.
50
00:02:21.680 --> 00:02:26.759
Completely that watchmaker's precision, it just vanishes. We aren't building
51
00:02:26.800 --> 00:02:30.240
a mechanism with a known outcome anymore, No, we really aren't.
52
00:02:30.400 --> 00:02:33.800
We are essentially pouring gears and springs into a box,
53
00:02:33.919 --> 00:02:37.919
shaking it up, and watching as the watch starts designing itself.
54
00:02:38.199 --> 00:02:40.240
That's a terrifying way to put it, but it's accurate.
55
00:02:40.319 --> 00:02:44.439
The blueprint is essentially writing its own continuation exactly. We
56
00:02:44.479 --> 00:02:47.199
are acting as catalysts for an emergent property, and the
57
00:02:47.240 --> 00:02:50.840
source material captures this shift beautifully. It frames the entire
58
00:02:50.879 --> 00:02:55.719
discussion of AI around this massive conceptual anchor that Bostrom
59
00:02:55.759 --> 00:02:58.159
calls the Big Fear in the Big Hope.
60
00:02:58.280 --> 00:02:59.639
The Big Fear in the Big Hope.
61
00:02:59.560 --> 00:03:01.919
Right, aims AI not just as you know, a tool
62
00:03:02.080 --> 00:03:04.759
on a spectrum of usefulness, like a better calculator.
63
00:03:04.879 --> 00:03:06.039
Yeah, it's way beyond that.
64
00:03:06.479 --> 00:03:11.680
It's a severe bifurcation in human history. So the Big
65
00:03:11.719 --> 00:03:15.439
Hope envisions AI is the ultimate savior of humanity, like
66
00:03:15.479 --> 00:03:17.000
a post scarcity.
67
00:03:16.520 --> 00:03:20.919
Mechanism, meaning it effortlessly untangles the complexities of climate change.
68
00:03:20.759 --> 00:03:25.599
Right exactly. It cures diseases by modeling protein folding in seconds.
69
00:03:25.639 --> 00:03:28.719
It optimizes global resource distributions so nobody goes hungry.
70
00:03:28.840 --> 00:03:30.840
It's the utopian outcome, yeah, the utopia.
71
00:03:30.960 --> 00:03:34.000
Yeah. But then you have the big fear, which casts
72
00:03:34.039 --> 00:03:38.319
this inescapable shadow. And that's the prospect of AI becoming
73
00:03:38.360 --> 00:03:40.120
our final invention, our.
74
00:03:40.000 --> 00:03:44.280
Final invention, meaning we literally never invent anything ever again.
75
00:03:44.360 --> 00:03:47.919
Right, it becomes an entity that simply outcomputes us, rendering
76
00:03:48.000 --> 00:03:52.560
human existence obsolete or you know, actively eliminating us because
77
00:03:52.599 --> 00:03:54.319
our goal is no longer align with its own.
78
00:03:54.479 --> 00:03:57.479
And the disparity between those two outcomes, it's just it's
79
00:03:57.520 --> 00:03:59.000
what makes the situation so.
80
00:03:59.159 --> 00:04:01.039
Incredibly it's mind boggling.
81
00:04:01.159 --> 00:04:04.599
We aren't talking about a technology that might be moderately
82
00:04:04.639 --> 00:04:08.360
helpful or like moderately annoying, like a new social media algorithm, right.
83
00:04:08.319 --> 00:04:12.240
We're talking about diametrically opposed extremes heaven or hell, basically.
84
00:04:12.000 --> 00:04:16.000
And Bostrom points out this very specific vulnerability in the
85
00:04:16.040 --> 00:04:18.879
people who are actually building these systems right now. He
86
00:04:19.000 --> 00:04:22.920
calls it an exhaustion of the imagination muscle, the imagination muscle,
87
00:04:23.279 --> 00:04:27.160
because the engineers driving this technology forward. They have this
88
00:04:27.279 --> 00:04:31.519
astonishing capacity to envision radical technical realities.
89
00:04:31.600 --> 00:04:36.199
Oh, their technical imagination is unparalleled. They can mathematically conceptualize
90
00:04:36.199 --> 00:04:40.399
a machine that possesses the sum of all human knowledge, right.
91
00:04:40.319 --> 00:04:44.240
A system that can outstrategize us at every conceivable game,
92
00:04:44.360 --> 00:04:46.000
every geopolitical scenario.
93
00:04:46.120 --> 00:04:49.720
But the issue is when they're asked to extrapolate the
94
00:04:49.800 --> 00:04:54.360
societal fallout of that invention, that imagination muscle just simply
95
00:04:54.399 --> 00:04:54.959
gives out.
96
00:04:55.240 --> 00:04:57.240
It just fails. Why do you think that is? I mean,
97
00:04:57.279 --> 00:05:00.360
these are brilliant people, arguably ten years aga away from
98
00:05:00.360 --> 00:05:03.639
building a replicated human mind. Why are they so blind
99
00:05:03.680 --> 00:05:04.399
to the impact.
100
00:05:04.759 --> 00:05:07.160
Well, the failure isn't a lack of general intelligence, right,
101
00:05:07.199 --> 00:05:09.680
It's more of a structural blindness. How so, it stems
102
00:05:09.680 --> 00:05:12.319
from the immense cognitive load required to operate at the
103
00:05:12.319 --> 00:05:15.680
cutting edge of computer science and mathematics. Think about it.
104
00:05:15.879 --> 00:05:19.360
When a developer is confronted with a computational bottleneck that
105
00:05:19.439 --> 00:05:23.079
has halted progress for years, the entire focus of their
106
00:05:23.120 --> 00:05:28.240
intellect is channeled into cracking that specific mathematical code. They
107
00:05:28.240 --> 00:05:31.600
are hyper focused on the how exactly, how do we
108
00:05:31.639 --> 00:05:34.800
optimize the parameter count? How do we reduce latency in
109
00:05:34.839 --> 00:05:35.680
the neural network.
110
00:05:35.800 --> 00:05:39.800
Right, Solving the how demands such intense granular concentration.
111
00:05:40.079 --> 00:05:43.800
Yeah, that there is quite literally no cognitive bandwidth left
112
00:05:43.959 --> 00:05:46.920
to process the why or the what next. Oh wow,
113
00:05:47.079 --> 00:05:50.040
By the time they successfully engineered the architecture to replicate
114
00:05:50.120 --> 00:05:53.639
human level reasoning, the mental energy required to ponder the
115
00:05:53.639 --> 00:05:57.399
immense ethical and social implications, it's just completely depleted.
116
00:05:57.519 --> 00:05:59.879
They fail to take the next step because their conceptual
117
00:06:00.000 --> 00:06:02.879
framework is strictly bound to that immediate technical hurdle.
118
00:06:03.000 --> 00:06:04.680
Yeah, they're so in the weeds they can't see the
119
00:06:04.759 --> 00:06:08.480
forest or the fact that the forest is potentially on fire.
120
00:06:08.600 --> 00:06:11.600
It reminds me of like assembling a team of the
121
00:06:11.639 --> 00:06:16.000
absolute most brilliant aerospace engineers to build a rocket capable
122
00:06:16.040 --> 00:06:17.079
of reaching light speed.
123
00:06:17.120 --> 00:06:18.319
Okay, I like this analogy.
124
00:06:18.560 --> 00:06:21.680
You have the best physicists, the best mechanics, and they
125
00:06:21.680 --> 00:06:25.040
are just intensely caught up in the sheer puzzle of
126
00:06:25.079 --> 00:06:25.920
the engine design.
127
00:06:26.000 --> 00:06:28.120
Right. They want to make it faster, more efficient, more.
128
00:06:28.040 --> 00:06:31.360
Powerful, and they pour years of their lives into optimizing
129
00:06:31.360 --> 00:06:35.240
the propulsion. But in that hyper focused state, they completely
130
00:06:35.279 --> 00:06:38.399
forget to install a steering wheel or brakes, or a
131
00:06:38.439 --> 00:06:40.839
life support system for the passenger.
132
00:06:40.439 --> 00:06:42.360
Which is a pretty big oversight. Right.
133
00:06:42.560 --> 00:06:46.079
The engine design works flawlessly, It does exactly what it
134
00:06:46.079 --> 00:06:49.040
was mathematically designed to do. But because they ignore the
135
00:06:49.160 --> 00:06:52.600
environment the engine operates in, the destination is a brick
136
00:06:52.639 --> 00:06:53.519
wall exactly.
137
00:06:53.639 --> 00:06:58.360
The engine represents raw, unaligned intelligence, and the steering wheel
138
00:06:58.399 --> 00:07:01.680
and the brakes those represent alignment with human well being
139
00:07:01.800 --> 00:07:03.000
and long term survival.
140
00:07:03.199 --> 00:07:06.639
And right now, all the prestige, all the venture capital,
141
00:07:06.680 --> 00:07:10.600
the academic accolades, it's entirely concentrated on building a bigger engine.
142
00:07:10.680 --> 00:07:13.279
The developers are solving for X, but they are not
143
00:07:13.480 --> 00:07:16.680
conceptually equipped to ask what happens to the entire alphabet
144
00:07:16.879 --> 00:07:19.120
once X is introduced into the ecosystem.
145
00:07:19.399 --> 00:07:23.040
So, if these developers are suffering from this exhausted imagination,
146
00:07:23.759 --> 00:07:26.360
if the breaks aren't even an afterthought, you have to ask,
147
00:07:26.560 --> 00:07:28.680
why do they keep pushing the accelerator?
148
00:07:28.879 --> 00:07:31.000
Right? Why is there no institutional pause?
149
00:07:31.279 --> 00:07:36.720
Exactly? To understand this compulsive, almost blind drive to innovate,
150
00:07:37.240 --> 00:07:42.000
the source material draws a direct, really unflinching comparison between
151
00:07:42.759 --> 00:07:46.079
the modern AI development race and the creation of the
152
00:07:46.120 --> 00:07:46.959
atomic bomb.
153
00:07:47.120 --> 00:07:50.000
Yeah, it points us toward the Manhattan Project as the
154
00:07:50.279 --> 00:07:54.639
historical precedent for developing this kind of existential technology.
155
00:07:54.240 --> 00:07:57.560
And that comparison provides vital context, doesn't it.
156
00:07:57.360 --> 00:07:59.279
It really does. It shows how the mechanics of a
157
00:07:59.319 --> 00:08:03.920
technological race dictate human behavior. The scientists working at Los
158
00:08:03.920 --> 00:08:08.240
Alamos in the nineteen forties they operated under an existential mandate.
159
00:08:07.959 --> 00:08:10.680
Because they possessed intelligence that the Germans had their own
160
00:08:10.759 --> 00:08:11.720
nuclear program.
161
00:08:11.879 --> 00:08:14.560
Right, So, the motivation to split the atom and weaponize
162
00:08:14.560 --> 00:08:18.240
it wasn't born out of abstract scientific curiosity or wanting
163
00:08:18.240 --> 00:08:18.959
to make a lot of mindy.
164
00:08:19.000 --> 00:08:19.959
It wasn't just fun to do.
165
00:08:20.360 --> 00:08:23.560
No, it was perceived as a strict wartime necessity for
166
00:08:23.639 --> 00:08:24.920
the survival of the free.
167
00:08:24.800 --> 00:08:28.879
World, and within that incredibly pressurized life or death environment.
168
00:08:29.360 --> 00:08:32.399
The source introduces this physicist, Leo Seialard.
169
00:08:32.639 --> 00:08:37.000
Sealard is crucial to Boston's framing here. He's described as
170
00:08:37.039 --> 00:08:41.120
this eccentric physicist who was actually the individual who first
171
00:08:41.159 --> 00:08:44.519
conceived the mechanics of a nuclear chain reaction.
172
00:08:44.759 --> 00:08:47.919
He understood the math of the explosion before the infrastructure
173
00:08:47.919 --> 00:08:49.399
to even build it existed.
174
00:08:49.559 --> 00:08:53.759
Yes, and what is remarkable about Selard is his immediate
175
00:08:53.799 --> 00:08:54.919
attempt at restraint.
176
00:08:55.399 --> 00:08:58.120
Right, he had this world altering revelation. But he didn't
177
00:08:58.159 --> 00:09:01.039
rush to publish a paper to get citations or secure
178
00:09:01.080 --> 00:09:01.720
his legacy.
179
00:09:01.879 --> 00:09:05.720
No, he actually engaged that imagination muscle. He extrapolated the
180
00:09:05.720 --> 00:09:10.480
physical reality of a nuclear chain reaction into its geopolitical consequence.
181
00:09:10.639 --> 00:09:13.919
He realized it could be catastrophically bad for civilization. He
182
00:09:13.960 --> 00:09:17.480
saw the engine and the lack of break simultaneously exactly.
183
00:09:17.200 --> 00:09:21.159
And his realization prompted this unprecedented action in the scientific community.
184
00:09:21.720 --> 00:09:24.440
Selord actively sought out his colleagues who were working in
185
00:09:24.519 --> 00:09:27.720
nuclear physics and attempted to persuade them to halt their experiments.
186
00:09:27.799 --> 00:09:30.480
He literally begged them not to pursue or publish their
187
00:09:30.519 --> 00:09:31.480
findings on fission.
188
00:09:31.720 --> 00:09:34.360
Yeah, he was essentially trying to organize a self imposed
189
00:09:34.399 --> 00:09:35.919
embargo on the knowledge.
190
00:09:36.120 --> 00:09:38.919
But the tragic punchline of the source material is that
191
00:09:39.120 --> 00:09:42.960
Silard achieved only quote unquote partial success.
192
00:09:42.799 --> 00:09:47.440
Partial success, which in this context is a tragedy. Some
193
00:09:47.480 --> 00:09:50.759
scientists agreed to delay their publications, but others.
194
00:09:50.440 --> 00:09:53.879
Refused because, as the source points out, some people were
195
00:09:53.960 --> 00:09:55.799
always going to want to be the ones that sort
196
00:09:55.799 --> 00:09:59.360
of innovate, and the concept of partial success in containing
197
00:09:59.399 --> 00:10:04.639
world alter technology is mathematically speaking identical to total failure.
198
00:10:04.720 --> 00:10:07.639
It is if even one innovator decides to publish, the
199
00:10:07.759 --> 00:10:09.480
cat is completely out of the bag.
200
00:10:09.679 --> 00:10:12.559
Right If one hundred scientists agree to lock the information
201
00:10:12.600 --> 00:10:14.759
in a vault, but the one hundred and first scientists
202
00:10:14.799 --> 00:10:18.399
decides to publish the paper, the embargo is meaningless. The
203
00:10:18.519 --> 00:10:19.679
knowledge propagates.
204
00:10:19.720 --> 00:10:22.600
You cannot undiscover the mechanism of the atom, just like.
205
00:10:22.600 --> 00:10:25.960
You cannot undiscover the architecture of a new neural network.
206
00:10:26.039 --> 00:10:29.639
Once the paper is out, the entire global community possesses
207
00:10:29.639 --> 00:10:30.440
the capability.
208
00:10:30.799 --> 00:10:36.480
The innovator's dilemma is fundamentally asymmetric. Containment requires unanimous, perpetual
209
00:10:36.559 --> 00:10:39.559
cooperation from every brilliant mind on the planet.
210
00:10:39.600 --> 00:10:43.440
Well, acceleration only requires a single actor to defect exactly.
211
00:10:43.480 --> 00:10:45.000
But I do have to push back a little on
212
00:10:45.039 --> 00:10:49.000
this Manhattan Project comparison for the listener to internalize today.
213
00:10:49.519 --> 00:10:51.399
Is it really a fair comparison?
214
00:10:51.679 --> 00:10:52.159
Hew? Do you mean?
215
00:10:52.639 --> 00:10:56.080
Well? During World War Two, those scientists were racing against
216
00:10:56.080 --> 00:10:59.639
a literal fascist empire. If they lost the race, the
217
00:10:59.639 --> 00:11:02.000
outcome home was global subjugation by Hitler.
218
00:11:02.200 --> 00:11:05.559
True, the stakes were uniquely high in a very visible way.
219
00:11:05.720 --> 00:11:09.279
But today AI developers at the major tech conglomerates are
220
00:11:09.360 --> 00:11:14.080
largely racing against rival corporations. The primary drivers are market share,
221
00:11:14.279 --> 00:11:17.840
stock valuations, being the first to deploy a commercial product.
222
00:11:17.960 --> 00:11:19.720
Yeah, it's a completely different landscape.
223
00:11:19.759 --> 00:11:23.480
There is no external, existential enemy forcing their hands. So
224
00:11:23.519 --> 00:11:27.039
doesn't that make the lack of ethical foresight today vastly
225
00:11:27.120 --> 00:11:30.200
less excusable than it was for Sealerd's colleagues.
226
00:11:30.360 --> 00:11:32.759
That is a very valid point. Yeah, the motivations have
227
00:11:32.840 --> 00:11:37.360
absolutely shifted from wartime survival to corporate dominance. Right. The
228
00:11:37.399 --> 00:11:40.279
critical takeaway from the source, however, is that while the
229
00:11:40.320 --> 00:11:43.919
moral justification has vanished, the mechanical pressure of the race
230
00:11:44.000 --> 00:11:45.080
remains identical.
231
00:11:45.480 --> 00:11:47.399
The fear of being second place.
232
00:11:47.879 --> 00:11:51.799
Exactly, whether it's to an enemination or a rival tech company,
233
00:11:51.879 --> 00:11:56.919
that fear drives the exact same reckless acceleration. The competitive
234
00:11:56.960 --> 00:12:00.320
ecosystem ensures that caution is penalized.
235
00:12:00.120 --> 00:12:03.399
Because if one company decides to halt development to spend
236
00:12:03.639 --> 00:12:07.600
say five years working on safety protocols and alignment.
237
00:12:07.639 --> 00:12:10.960
A rival company will simply bypass them, release the product,
238
00:12:11.039 --> 00:12:14.159
and capture the entire market. The structure of the race
239
00:12:14.480 --> 00:12:19.039
forcess participants to prioritize speed over safety regardless of whether
240
00:12:19.080 --> 00:12:22.000
the prize is global security or just quarterly profits.
241
00:12:22.320 --> 00:12:25.159
So we basically have a structural guarantee that the acceleration
242
00:12:25.240 --> 00:12:28.480
will continue, which forces us to look at exactly how
243
00:12:28.519 --> 00:12:29.840
this technology is advancing.
244
00:12:29.960 --> 00:12:32.879
Right, how are we moving from standard machines to these
245
00:12:32.919 --> 00:12:34.440
world altering intellects.
246
00:12:34.519 --> 00:12:36.679
We have to look at the mechanics of reverse engineering
247
00:12:36.679 --> 00:12:39.440
the brain, and the source points out this massive paradigm
248
00:12:39.440 --> 00:12:41.879
shift in how machines interact with physical space.
249
00:12:42.200 --> 00:12:47.399
Yeah, the transition from algorithmic programming to true artificial intelligence
250
00:12:47.799 --> 00:12:50.600
is best understood by looking at traditional robotics.
251
00:12:50.720 --> 00:12:56.519
Source material calls current robotic systems quote blind dumb things, blind.
252
00:12:56.279 --> 00:13:00.120
Dumb things, because a robotic arm operating in say an
253
00:13:00.159 --> 00:13:05.679
automotive factory, is just an execution of rigid state space programming.
254
00:13:05.759 --> 00:13:09.480
It's given precise three dimensional coordinates. It moves to coordinate A,
255
00:13:10.039 --> 00:13:13.519
engages the welding torch and moves to coordinate B, and.
256
00:13:13.440 --> 00:13:16.759
It does this flawlessly, but it has absolutely no awareness
257
00:13:16.759 --> 00:13:19.000
of its environment. If a different car chassis comes down
258
00:13:19.039 --> 00:13:21.519
the assembly line, or if the door is supposed to
259
00:13:21.519 --> 00:13:23.080
weld is entirely.
260
00:13:22.759 --> 00:13:25.360
Missing, it doesn't adapt. It just continues to weld the
261
00:13:25.399 --> 00:13:28.960
empty air because it's programming dictates that the coordinates are absolute.
262
00:13:29.080 --> 00:13:32.159
Exactly. If you want to change production, human engineers have
263
00:13:32.200 --> 00:13:35.080
to halt the line and painstakingly reprogram those coordinates.
264
00:13:35.200 --> 00:13:38.559
It possesses zero context. It is just playing a physical
265
00:13:38.600 --> 00:13:39.840
recording over and over.
266
00:13:40.039 --> 00:13:43.960
But the source introduces this new concept imitation learning, and
267
00:13:44.000 --> 00:13:46.600
that is the mechanism that completely shatters this limitation.
268
00:13:47.000 --> 00:13:50.240
Imitation learning. So instead of hard coding a robot, you
269
00:13:50.240 --> 00:13:52.559
could show it how to do something once or twice
270
00:13:52.759 --> 00:13:53.919
like you would a human baby.
271
00:13:54.039 --> 00:13:56.840
Right, it shifts a machine from following a script to
272
00:13:57.440 --> 00:13:59.960
observing an environment it learns to do it robot.
273
00:14:00.559 --> 00:14:03.120
So if a machine learns by watching, it means it's
274
00:14:03.159 --> 00:14:07.000
developing a generalized understanding of the physics and the intent
275
00:14:07.080 --> 00:14:08.120
within its environment.
276
00:14:08.320 --> 00:14:11.879
That's the key word intent. It figures out the goal
277
00:14:12.000 --> 00:14:15.480
of the action rather than just memorizing the mechanical coordinates
278
00:14:15.559 --> 00:14:15.879
used to.
279
00:14:15.840 --> 00:14:18.519
Achieve it, which is a profound lead when you think
280
00:14:18.559 --> 00:14:19.240
about it.
281
00:14:19.240 --> 00:14:21.799
It really is. The leap from mapping an action to
282
00:14:21.879 --> 00:14:25.600
mapping an intent changes everything. For example, if you demonstrate
283
00:14:25.720 --> 00:14:28.799
pouring coffee into a mug, the traditional robot tries to
284
00:14:28.840 --> 00:14:32.080
memorize the angle of your elbow. But the machine using
285
00:14:32.080 --> 00:14:36.799
imitation learning understands that the objective is transferring liquid from
286
00:14:36.799 --> 00:14:39.679
the pot into the mug. So if you move the
287
00:14:39.720 --> 00:14:41.240
mug to the other side of the table.
288
00:14:41.000 --> 00:14:44.480
The imitation learning model adjusts its physical approach to achieve
289
00:14:44.480 --> 00:14:45.080
the goal.
290
00:14:45.039 --> 00:14:48.960
Exactly whereas the traditional robot just pours scalding coffee onto
291
00:14:49.000 --> 00:14:50.639
the empty spot where the mug used to be.
292
00:14:50.960 --> 00:14:55.000
And the source explicitly connects this capability to a much
293
00:14:55.039 --> 00:14:58.240
grander objective, reverse engineering the human brain.
294
00:14:58.879 --> 00:15:02.919
By teaching systems to to generalize and infer intent, developers
295
00:15:02.919 --> 00:15:06.720
are laying the foundational architecture for cognitive replication.
296
00:15:06.600 --> 00:15:10.679
Which brings in Ray Kurzweil's predictions. The source notes that
297
00:15:10.919 --> 00:15:14.759
realizing Kurtzweil's stuff, which is like a radical tech future
298
00:15:14.799 --> 00:15:17.960
a one thousand year leap compacted into a few decades,
299
00:15:18.320 --> 00:15:21.600
is technically possible at what they call technical maturity.
300
00:15:21.679 --> 00:15:25.919
Technical maturity that's a theoretical endpoint where a civilization has
301
00:15:26.000 --> 00:15:30.919
essentially maximized its technological capabilities within the bounds of physics.
302
00:15:30.480 --> 00:15:33.799
Meaning we possess the computational power and the material science
303
00:15:33.879 --> 00:15:37.480
to perfectly simulate and manipulate reality at the atomic level.
304
00:15:37.639 --> 00:15:41.799
Right, But Bostrom provides two massive caveats regarding our ability
305
00:15:41.799 --> 00:15:43.240
to actually reach this future.
306
00:15:43.559 --> 00:15:46.799
The first caveat being do we actually reach technological maturity
307
00:15:46.840 --> 00:15:47.960
without destroying ourselves?
308
00:15:48.039 --> 00:15:52.360
First? Exactly, given the exhausted imagination of developers and the
309
00:15:52.399 --> 00:15:55.480
relentless pressure of the innovator's dilemma we just talked about,
310
00:15:55.759 --> 00:15:58.679
surviving our own technological adolescens is the ultimate filter.
311
00:15:58.960 --> 00:16:01.799
The tools we are built are powerful enough to terminate
312
00:16:01.799 --> 00:16:03.320
the development curve permanently.
313
00:16:03.559 --> 00:16:03.799
Yes.
314
00:16:04.000 --> 00:16:07.240
And the second caveat is, even if we survive the
315
00:16:07.279 --> 00:16:11.159
creation of AI, are we actually interested in using our
316
00:16:11.159 --> 00:16:14.200
technology for that specific purpose at that time?
317
00:16:14.440 --> 00:16:17.600
Right? It's like unlocking the final tier of massive video
318
00:16:17.639 --> 00:16:18.399
game tech trees.
319
00:16:18.559 --> 00:16:22.519
I love that analogy. You spend hours gathering resources, leveling
320
00:16:22.600 --> 00:16:26.000
up your civilization, and you finally unlock the ability to
321
00:16:26.080 --> 00:16:27.320
build a digital brain.
322
00:16:27.480 --> 00:16:29.600
But just because you have the capability to spend your
323
00:16:29.639 --> 00:16:31.559
points to build it doesn't mean you will.
324
00:16:31.759 --> 00:16:34.840
By the time humanity achieves technical maturity, our priorities as
325
00:16:34.840 --> 00:16:38.480
a species could be entirely different. We might realize simulating
326
00:16:38.519 --> 00:16:40.840
a human mind is unnecessary.
327
00:16:40.279 --> 00:16:43.639
Or unethical, or simply uninteresting compared to other avenues of
328
00:16:43.679 --> 00:16:46.720
exploration we haven't even conceived of yet. Capability does not
329
00:16:46.799 --> 00:16:47.919
necessitate execution.
330
00:16:48.519 --> 00:16:50.919
But and this is a big butt, if we do
331
00:16:51.000 --> 00:16:54.240
decide to proceed with reverse engineering the mind using mechanisms
332
00:16:54.240 --> 00:16:58.720
like imitation learning, we are opening up a terrifying philosophical puzzle.
333
00:16:58.840 --> 00:17:02.559
The ethical weight of imitation learning is staggering because when
334
00:17:02.559 --> 00:17:05.960
a machine learns by observing human behavior, it isn't just
335
00:17:06.039 --> 00:17:09.279
absorbing our physical movements. It absorbs our flaws exactly. It
336
00:17:09.319 --> 00:17:14.720
absorbs our context, our unspoken societal rules, our biases, our prejudices,
337
00:17:14.920 --> 00:17:16.480
and our unstated intentions.
338
00:17:16.799 --> 00:17:20.599
So if we create a superintelligence that infers how to
339
00:17:20.680 --> 00:17:26.359
behave by watching us, the big fear is amplified massively.
340
00:17:26.039 --> 00:17:29.319
Right, because the system might learn and optimize our most
341
00:17:29.319 --> 00:17:32.960
descriptive traits just as efficiently as our cooperative ones.
342
00:17:33.200 --> 00:17:36.640
Oh wow. So if we somehow successfully navigate all of
343
00:17:36.640 --> 00:17:40.720
those risks, if we reach technological maturity, decide to pursue
344
00:17:40.720 --> 00:17:43.400
the simulation and align its intents so it doesn't.
345
00:17:43.079 --> 00:17:44.599
Destroy us, that's a lot of ifs.
346
00:17:44.759 --> 00:17:46.440
It is a lot of ifs. But if we do,
347
00:17:46.519 --> 00:17:48.640
the source material points out, it forces us to ask
348
00:17:48.680 --> 00:17:51.839
a terrifying, reality breaking question about our own existence.
349
00:17:52.079 --> 00:17:55.720
Yes, because successfully reverse engineering the brain forces us to
350
00:17:55.759 --> 00:17:58.200
grapple with the substrate independence thesis.
351
00:17:58.400 --> 00:18:01.480
The substrate independence these says, can you explain that concept
352
00:18:01.519 --> 00:18:02.680
clearly for the listener?
353
00:18:02.920 --> 00:18:06.920
Absolutely. It's this concept from the philosophy of mind. It
354
00:18:07.039 --> 00:18:10.839
posits that a mind or consciousness does not strictly require
355
00:18:10.880 --> 00:18:15.359
a biological, carbon based quote unquote meat brain to exist.
356
00:18:15.759 --> 00:18:19.440
We just inherently assume that consciousness is a property exclusive
357
00:18:19.480 --> 00:18:22.799
to biological neurons and synapses, because that's the only format
358
00:18:22.839 --> 00:18:24.079
we've ever observed.
359
00:18:23.640 --> 00:18:27.599
It in, right, But the thesis argues that consciousness is
360
00:18:27.759 --> 00:18:31.680
not a byproduct of the biological material itself. It's an
361
00:18:31.680 --> 00:18:35.039
emergent property of the information processing that is happening.
362
00:18:35.200 --> 00:18:37.400
It's separating the software from the hardware.
363
00:18:37.599 --> 00:18:40.720
Exactly. If you have a different substrate like silicon chips
364
00:18:40.839 --> 00:18:44.720
or quantum processors or abstract code inside a computer, and
365
00:18:44.759 --> 00:18:48.160
that non biological substrate can process information with the exact
366
00:18:48.200 --> 00:18:50.160
same complexity as a human brain, the.
367
00:18:50.119 --> 00:18:53.200
Exact same architecture, the exact same feedback loops.
368
00:18:52.960 --> 00:18:56.160
Then the thesis states that the non biological substrate will
369
00:18:56.160 --> 00:18:59.480
generate genuine consciousness, it can operate on a different substrate,
370
00:18:59.559 --> 00:18:59.720
and the.
371
00:18:59.720 --> 00:19:02.279
Impls location for simulated brains is absolute.
372
00:19:02.519 --> 00:19:06.920
If consciousness is substrate independent, then a perfectly reverse engineered
373
00:19:06.960 --> 00:19:10.279
brain running on a computer isn't just a clever parlor trick.
374
00:19:10.559 --> 00:19:14.000
It's not just a complex imitation. That digital entity is
375
00:19:14.039 --> 00:19:18.440
experiencing reality. It possesses a subjective perspective.
376
00:19:18.599 --> 00:19:22.359
It feels joy, it experiences pain, It forms memories. For
377
00:19:22.400 --> 00:19:25.119
all intents and purposes, the code is alive.
378
00:19:25.519 --> 00:19:29.000
And this logical progression is what serves as the launch
379
00:19:29.039 --> 00:19:33.079
pad for Nick Bostrom's famous simulation hypothesis.
380
00:19:32.480 --> 00:19:35.680
Which fundamentally challenges our entire concept of reality.
381
00:19:35.799 --> 00:19:40.039
The logic of the simulation hypothesis is terrifyingly sound once
382
00:19:40.079 --> 00:19:42.559
you accept the premise of substrate independence.
383
00:19:42.680 --> 00:19:46.599
It really is. If you accept imitation, learning, then reverse engineering,
384
00:19:46.720 --> 00:19:51.079
then substrate independence. The simulation hypothesis is almost inescapable.
385
00:19:51.119 --> 00:19:52.799
So walk us through the math of that, because it's
386
00:19:52.880 --> 00:19:53.519
mind bending.
387
00:19:53.680 --> 00:19:56.079
Okay, So if consciousness can exisit as code, we then
388
00:19:56.160 --> 00:19:59.559
look at the timeline a technologically mature civilization, whether it's
389
00:19:59.599 --> 00:20:01.440
our des SA and thence millions of years from now
390
00:20:01.920 --> 00:20:04.720
or an alien race, they will possess computing power that
391
00:20:04.839 --> 00:20:06.839
is practically incomprehensible to us.
392
00:20:06.880 --> 00:20:09.519
They could utilize the energy output of entire stars to
393
00:20:09.640 --> 00:20:11.599
power computers the size of planets.
394
00:20:11.799 --> 00:20:16.279
Right with that level of computational density, the energetic cost
395
00:20:16.440 --> 00:20:19.839
of running a simulation of human history, what Bostrom calls
396
00:20:19.880 --> 00:20:22.839
an ancestor simulation, would be microscopic.
397
00:20:23.200 --> 00:20:27.279
A mature civilization could theoretically run billions or trillions of
398
00:20:27.400 --> 00:20:29.279
highly detailed simulations, and.
399
00:20:29.240 --> 00:20:33.559
Each of those simulations would contain billions of conscious simulated
400
00:20:33.640 --> 00:20:34.400
human minds.
401
00:20:34.480 --> 00:20:39.359
And the mathematical consequence of this capability is just staggering, because.
402
00:20:39.079 --> 00:20:42.480
If a civilization runs billions of simulations, the number of
403
00:20:42.559 --> 00:20:47.079
simulated digital minds in existence would vastly outnumber the original
404
00:20:47.079 --> 00:20:49.559
biological minds that lived in the base reality.
405
00:20:49.599 --> 00:20:53.160
The ratio would be astronomical billions or trillions of simulated
406
00:20:53.200 --> 00:20:57.359
minds to every one biological mind. Therefore, statistically speaking, if
407
00:20:57.400 --> 00:21:00.440
you are a conscious entity currently experiencing a reality, the
408
00:21:00.440 --> 00:21:04.000
probability that you are the one original biological entity living
409
00:21:04.039 --> 00:21:05.720
in base reality plummet.
410
00:21:05.839 --> 00:21:09.799
It's practically zero. The math forces the conclusion that we're
411
00:21:10.039 --> 00:21:12.799
almost certainly living inside one of those billions of ancestors
412
00:21:12.839 --> 00:21:13.720
simulations right.
413
00:21:13.640 --> 00:21:16.039
Now, it is incredibly difficult to sit with that idea.
414
00:21:16.160 --> 00:21:18.000
It induces a lot of existential dread.
415
00:21:18.119 --> 00:21:21.720
Yeah, because to raise a philosophical pushback on behalf of
416
00:21:21.720 --> 00:21:23.920
the listener here, If I have to accept the substrate
417
00:21:23.960 --> 00:21:27.480
independence thesis that my thoughts and feelings are just softler
418
00:21:27.599 --> 00:21:32.319
running on a biological hard drive, doesn't that completely devalue
419
00:21:32.359 --> 00:21:33.440
the human experience.
420
00:21:34.039 --> 00:21:36.160
That's the immediate reaction most people.
421
00:21:35.920 --> 00:21:39.039
Have, right if we're just rendered algorithms. Does falling in
422
00:21:39.079 --> 00:21:43.200
love actually matter? Does feeling grief for a lost loved
423
00:21:43.240 --> 00:21:45.960
one mean anything at all? It feels like a pathway
424
00:21:46.000 --> 00:21:48.279
straight into profound nihilism.
425
00:21:47.960 --> 00:21:51.519
And that sense of existential demotion is so common. Stripping
426
00:21:51.559 --> 00:21:54.039
away the physical permanence of the universe and replacing it
427
00:21:54.039 --> 00:21:57.039
with rendered parameters challenges our core sense of meaning.
428
00:21:57.200 --> 00:21:57.920
It really does.
429
00:21:58.119 --> 00:22:01.799
However, the source material actually pivot Here. Bostrom turns the
430
00:22:01.839 --> 00:22:03.319
simulation theory on its head.
431
00:22:03.599 --> 00:22:06.559
Instead of being a depressing thought, there's an upside.
432
00:22:06.039 --> 00:22:10.480
A surprising psychological upside. Yes. The final concept explored is
433
00:22:10.519 --> 00:22:13.480
the psychological benefits for interacting with life. As if it's
434
00:22:13.480 --> 00:22:14.519
a simulation.
435
00:22:14.279 --> 00:22:16.759
How does adopting the framework that we might be lines
436
00:22:16.799 --> 00:22:19.599
of code alleviate the anxiety of living well.
437
00:22:19.640 --> 00:22:22.039
It cannex back directly to the big fear we discussed
438
00:22:22.039 --> 00:22:26.359
at the very beginning. Human beings are deeply encumbered by
439
00:22:26.400 --> 00:22:30.680
existential anxiety. We carry the terror of our own mortality,
440
00:22:30.720 --> 00:22:34.359
the permanence of our failures, the crushing way to the unknown.
441
00:22:34.480 --> 00:22:37.960
We are fear driven creatures in many ways, right, But.
442
00:22:37.960 --> 00:22:40.400
When you entertain the idea that life is a simulation,
443
00:22:41.000 --> 00:22:45.880
it fundamentally reframes the stakes of those daily anxieties. Adopting
444
00:22:45.880 --> 00:22:48.400
this framework changes how one interacts.
445
00:22:47.920 --> 00:22:51.240
With the world, because accepting substrate independence means your feelings
446
00:22:51.279 --> 00:22:54.400
are still entirely real. Right, Your joy and pain are
447
00:22:54.440 --> 00:22:57.400
genuine experiences of your consciousness.
448
00:22:56.799 --> 00:23:01.519
Exactly, but the context surrounding those experiences shif It removes
449
00:23:01.559 --> 00:23:04.960
the absolute permanent weight of the biological universe.
450
00:23:05.039 --> 00:23:08.519
It encourages a curiosity driven approach to life rather than
451
00:23:08.559 --> 00:23:09.680
a fear driven one.
452
00:23:09.880 --> 00:23:12.880
Rather than inducing nihilism. It offers a strange sense of
453
00:23:13.000 --> 00:23:16.200
liberation or perspective. It alleviates the big fear.
454
00:23:16.400 --> 00:23:19.279
It's like playing an incredibly difficult, high stakes video game.
455
00:23:19.400 --> 00:23:21.359
Oh, that's a great way to ground this for the listener.
456
00:23:21.680 --> 00:23:25.079
Think about it. When you are deeply immersed in a
457
00:23:25.119 --> 00:23:29.279
demanding digital environment and you temporarily forget that you are
458
00:23:29.319 --> 00:23:33.519
engaging with a game, the stress becomes agonizing.
459
00:23:32.920 --> 00:23:35.759
Your heart rate spikes, your breathing becomes shallow.
460
00:23:35.960 --> 00:23:39.920
The fear of your character failing paralyzes your decision making.
461
00:23:40.200 --> 00:23:44.599
You become totally risk averse because the simulated danger feels real. Right,
462
00:23:45.119 --> 00:23:47.640
But the moment you take a breath and remember, oh wait,
463
00:23:47.720 --> 00:23:51.039
this is just a simulation, you don't immediately throw the
464
00:23:51.039 --> 00:23:52.599
controller away and stop playing.
465
00:23:52.759 --> 00:23:56.400
Knowing it's constructed doesn't drain the experience of its value exactly.
466
00:23:56.440 --> 00:23:59.400
In fact, you usually play better. You become more creative
467
00:23:59.440 --> 00:24:01.799
with your problem solving. You test the boundaries of the
468
00:24:01.839 --> 00:24:05.599
physics engine. You take larger risks and try new strategies
469
00:24:05.640 --> 00:24:09.599
because that paralyzing dread of permanent failure has been lifted.
470
00:24:10.200 --> 00:24:14.400
The realization that the environment is constructed frees the consciousness
471
00:24:14.440 --> 00:24:18.559
to engage with the environment more dynamically. Recognizing the simulation
472
00:24:18.720 --> 00:24:20.920
provides an intellectual buffer against the void.
473
00:24:21.039 --> 00:24:23.720
It allows you to appreciate the complexity of the rendering
474
00:24:23.960 --> 00:24:26.720
without being crushed by the permanence of it all. So
475
00:24:27.000 --> 00:24:30.200
we've gone on a massive journey today through Bostrom's theories.
476
00:24:30.640 --> 00:24:34.640
It forces us to continuously reevaluate our position in the universe.
477
00:24:34.960 --> 00:24:37.920
We started by observing the watchmakers of our era, the
478
00:24:38.000 --> 00:24:41.599
developers building AI, and examine how their hyper focus blinds
479
00:24:41.640 --> 00:24:44.519
them to the societal impact of the intelligence they're engineering.
480
00:24:44.799 --> 00:24:48.960
We contextualize that exhausted imagination through the historical lens of
481
00:24:49.039 --> 00:24:51.559
Leo Cilard and the Manhattan Project.
482
00:24:51.559 --> 00:24:55.759
Illustrating how the innovator's dilemma guarantees that technological acceleration will
483
00:24:55.759 --> 00:25:00.279
continue since partial success and containment is mathematically impossible.
484
00:25:00.640 --> 00:25:04.440
We trace the mechanical evolution from blind dumb things to
485
00:25:04.480 --> 00:25:07.920
the imitation learning that sets the stage for reverse engineering
486
00:25:08.000 --> 00:25:08.359
the brain.
487
00:25:08.559 --> 00:25:11.680
And then we explored the profound philosophical consequences of that
488
00:25:11.759 --> 00:25:15.839
replication through the substrate independence thesis, leading to the mind
489
00:25:15.839 --> 00:25:18.240
bending reality of the simulation hypothesis.
490
00:25:18.440 --> 00:25:21.400
And finally we found a strange psychological freedom in that
491
00:25:21.519 --> 00:25:25.880
exact hypothesis. Understanding these interconnected concepts gives you, the listener,
492
00:25:26.119 --> 00:25:29.720
a shortcut to being deeply well informed about the existential
493
00:25:29.720 --> 00:25:31.799
crosshairs humanity currently sits in.
494
00:25:32.079 --> 00:25:35.720
You're no longer just reacting to isolated headlines about new
495
00:25:35.720 --> 00:25:39.519
AI models. You can see the philosophical endgame driving.
496
00:25:39.160 --> 00:25:41.480
Them to conclude this deep dive. I want to leave
497
00:25:41.519 --> 00:25:45.160
you with a final provocative thought to mull over. Think
498
00:25:45.240 --> 00:25:49.400
back to those developers racing to build artificial superintelligence. Okay,
499
00:25:49.759 --> 00:25:52.319
if we are indeed living in a simulation created by
500
00:25:52.359 --> 00:25:57.160
a technologically mature civilization, and we are currently racing with
501
00:25:57.279 --> 00:26:01.759
exhausted imaginations and unstoppable competitive of momentum to build our
502
00:26:01.799 --> 00:26:03.920
own super intelligent AI, are.
503
00:26:03.759 --> 00:26:06.680
We simply acting out the historical code of our creators?
504
00:26:06.839 --> 00:26:10.200
Exactly? Is our rush to build AI just a pre
505
00:26:10.240 --> 00:26:14.200
programmed variable? And when our artificial intelligence finally wakes up,
506
00:26:14.240 --> 00:26:17.440
achieving super intelligence and surpassing our comprehension, will it be
507
00:26:17.480 --> 00:26:19.960
the end of our world or just the required trigger
508
00:26:20.000 --> 00:26:22.759
to launch the beginning of the next simulation down the chain.
509
00:26:22.880 --> 00:26:25.480
Wow, that is quite the thought to end on.
510
00:26:25.839 --> 00:26:27.720
Thank you all for taking this deep dive with us
511
00:26:27.759 --> 00:26:28.079
today
1
00:00:00.080 --> 00:00:02.879
Welcome to the deep dive of everybody. So today we
2
00:00:02.960 --> 00:00:06.559
are jumping into a really profound topic. We're looking at
3
00:00:06.559 --> 00:00:09.480
the trajectory of artificial intelligence, but we're doing it through
4
00:00:09.480 --> 00:00:12.320
a very specific, highly philosophical lens.
5
00:00:12.400 --> 00:00:14.560
Yeah, it's a massive subject to try and wrap your
6
00:00:14.560 --> 00:00:15.279
head around.
7
00:00:15.080 --> 00:00:18.000
It really is. And our source material today is the
8
00:00:18.440 --> 00:00:22.559
monumental Joe Rogan Experience number thirteen fifty and that one
9
00:00:22.559 --> 00:00:24.160
features Nick Bostrom, right.
10
00:00:24.280 --> 00:00:26.600
Nick Bostrom, And for those who might not know, he's
11
00:00:26.640 --> 00:00:30.160
this brilliant Swedish philosopher based to the University.
12
00:00:29.800 --> 00:00:30.800
Of Oxford exactly.
13
00:00:30.879 --> 00:00:34.399
He's just globally known for his foundational work on existential risk,
14
00:00:35.479 --> 00:00:39.640
the anthropic principle, human enhancement, ethics, and you know the
15
00:00:39.679 --> 00:00:41.640
really big one, super intelligence.
16
00:00:41.200 --> 00:00:42.799
Risks, the fun, lighthearted stuff.
17
00:00:43.039 --> 00:00:45.200
Oh yeah, definitely real casual weekend reading.
18
00:00:45.679 --> 00:00:47.799
But you know, the mission for us today is to
19
00:00:47.840 --> 00:00:50.560
act as your guides through this source material. We want
20
00:00:50.600 --> 00:00:54.399
to navigate this really fragile, terrifying but also kind of
21
00:00:54.479 --> 00:01:00.119
awe inspiring bridge between today's technological innovations and tomorrow's.
22
00:00:59.520 --> 00:01:04.840
Reality, because we are essentially exploring how we might either
23
00:01:04.920 --> 00:01:08.480
completely engineer our own destruction or maybe come to the
24
00:01:08.519 --> 00:01:12.359
realization that we are simply lines of code in a cosmic.
25
00:01:11.959 --> 00:01:15.719
Computer, which is I mean, it's a heavy subject, but
26
00:01:16.120 --> 00:01:18.879
we are going to navigate it clearly for you, no
27
00:01:19.159 --> 00:01:21.480
crazy jargon. Right. We want you to walk away with
28
00:01:21.560 --> 00:01:26.680
those profound aha moments, really understanding the mechanics of these theories.
29
00:01:26.760 --> 00:01:30.079
Yeah, to really grasp the current technological landscape we're all
30
00:01:30.079 --> 00:01:30.959
sitting in right now.
31
00:01:31.120 --> 00:01:33.879
So to set the stage for how bosstorm views are
32
00:01:33.959 --> 00:01:36.599
current era of tech, I want you to think about
33
00:01:36.640 --> 00:01:38.719
the environment of a master watchmaker.
34
00:01:38.760 --> 00:01:39.799
Okay, a watchmaker.
35
00:01:39.920 --> 00:01:43.799
Yeah, you picture them with their magnifying loop right, their tweezers,
36
00:01:44.120 --> 00:01:47.359
and they are carefully placing every microscopic gear and spring
37
00:01:47.439 --> 00:01:50.359
into the chest, well, very meticulous exactly. They know exactly
38
00:01:50.400 --> 00:01:53.519
what every single piece does. They understand the complete picture
39
00:01:53.560 --> 00:01:56.599
from the start, and the ultimate purpose is timekeeping.
40
00:01:56.760 --> 00:02:00.599
Yeah, every component serves that clear, entirely predictive blend. The
41
00:02:00.599 --> 00:02:02.439
physics are known, the boundaries.
42
00:02:02.000 --> 00:02:04.439
Are closed, the outcome is defined before they even pick
43
00:02:04.519 --> 00:02:05.799
up a tool, right.
44
00:02:05.879 --> 00:02:09.800
The intention governs the entire process. There is no mystery
45
00:02:10.199 --> 00:02:13.479
at all about what the mechanism will do once that
46
00:02:13.560 --> 00:02:14.719
final gear is in place.
47
00:02:14.960 --> 00:02:17.639
But then you take a step into the modern laboratories
48
00:02:17.639 --> 00:02:20.240
where artificial intelligence is being developed right.
49
00:02:20.120 --> 00:02:21.520
Now, and it's a totally different story.
50
00:02:21.680 --> 00:02:26.759
Completely that watchmaker's precision, it just vanishes. We aren't building
51
00:02:26.800 --> 00:02:30.240
a mechanism with a known outcome anymore, No, we really aren't.
52
00:02:30.400 --> 00:02:33.800
We are essentially pouring gears and springs into a box,
53
00:02:33.919 --> 00:02:37.919
shaking it up, and watching as the watch starts designing itself.
54
00:02:38.199 --> 00:02:40.240
That's a terrifying way to put it, but it's accurate.
55
00:02:40.319 --> 00:02:44.439
The blueprint is essentially writing its own continuation exactly. We
56
00:02:44.479 --> 00:02:47.199
are acting as catalysts for an emergent property, and the
57
00:02:47.240 --> 00:02:50.840
source material captures this shift beautifully. It frames the entire
58
00:02:50.879 --> 00:02:55.719
discussion of AI around this massive conceptual anchor that Bostrom
59
00:02:55.759 --> 00:02:58.159
calls the Big Fear in the Big Hope.
60
00:02:58.280 --> 00:02:59.639
The Big Fear in the Big Hope.
61
00:02:59.560 --> 00:03:01.919
Right, aims AI not just as you know, a tool
62
00:03:02.080 --> 00:03:04.759
on a spectrum of usefulness, like a better calculator.
63
00:03:04.879 --> 00:03:06.039
Yeah, it's way beyond that.
64
00:03:06.479 --> 00:03:11.680
It's a severe bifurcation in human history. So the Big
65
00:03:11.719 --> 00:03:15.439
Hope envisions AI is the ultimate savior of humanity, like
66
00:03:15.479 --> 00:03:17.000
a post scarcity.
67
00:03:16.520 --> 00:03:20.919
Mechanism, meaning it effortlessly untangles the complexities of climate change.
68
00:03:20.759 --> 00:03:25.599
Right exactly. It cures diseases by modeling protein folding in seconds.
69
00:03:25.639 --> 00:03:28.719
It optimizes global resource distributions so nobody goes hungry.
70
00:03:28.840 --> 00:03:30.840
It's the utopian outcome, yeah, the utopia.
71
00:03:30.960 --> 00:03:34.000
Yeah. But then you have the big fear, which casts
72
00:03:34.039 --> 00:03:38.319
this inescapable shadow. And that's the prospect of AI becoming
73
00:03:38.360 --> 00:03:40.120
our final invention, our.
74
00:03:40.000 --> 00:03:44.280
Final invention, meaning we literally never invent anything ever again.
75
00:03:44.360 --> 00:03:47.919
Right, it becomes an entity that simply outcomputes us, rendering
76
00:03:48.000 --> 00:03:52.560
human existence obsolete or you know, actively eliminating us because
77
00:03:52.599 --> 00:03:54.319
our goal is no longer align with its own.
78
00:03:54.479 --> 00:03:57.479
And the disparity between those two outcomes, it's just it's
79
00:03:57.520 --> 00:03:59.000
what makes the situation so.
80
00:03:59.159 --> 00:04:01.039
Incredibly it's mind boggling.
81
00:04:01.159 --> 00:04:04.599
We aren't talking about a technology that might be moderately
82
00:04:04.639 --> 00:04:08.360
helpful or like moderately annoying, like a new social media algorithm, right.
83
00:04:08.319 --> 00:04:12.240
We're talking about diametrically opposed extremes heaven or hell, basically.
84
00:04:12.000 --> 00:04:16.000
And Bostrom points out this very specific vulnerability in the
85
00:04:16.040 --> 00:04:18.879
people who are actually building these systems right now. He
86
00:04:19.000 --> 00:04:22.920
calls it an exhaustion of the imagination muscle, the imagination muscle,
87
00:04:23.279 --> 00:04:27.160
because the engineers driving this technology forward. They have this
88
00:04:27.279 --> 00:04:31.519
astonishing capacity to envision radical technical realities.
89
00:04:31.600 --> 00:04:36.199
Oh, their technical imagination is unparalleled. They can mathematically conceptualize
90
00:04:36.199 --> 00:04:40.399
a machine that possesses the sum of all human knowledge, right.
91
00:04:40.319 --> 00:04:44.240
A system that can outstrategize us at every conceivable game,
92
00:04:44.360 --> 00:04:46.000
every geopolitical scenario.
93
00:04:46.120 --> 00:04:49.720
But the issue is when they're asked to extrapolate the
94
00:04:49.800 --> 00:04:54.360
societal fallout of that invention, that imagination muscle just simply
95
00:04:54.399 --> 00:04:54.959
gives out.
96
00:04:55.240 --> 00:04:57.240
It just fails. Why do you think that is? I mean,
97
00:04:57.279 --> 00:05:00.360
these are brilliant people, arguably ten years aga away from
98
00:05:00.360 --> 00:05:03.639
building a replicated human mind. Why are they so blind
99
00:05:03.680 --> 00:05:04.399
to the impact.
100
00:05:04.759 --> 00:05:07.160
Well, the failure isn't a lack of general intelligence, right,
101
00:05:07.199 --> 00:05:09.680
It's more of a structural blindness. How so, it stems
102
00:05:09.680 --> 00:05:12.319
from the immense cognitive load required to operate at the
103
00:05:12.319 --> 00:05:15.680
cutting edge of computer science and mathematics. Think about it.
104
00:05:15.879 --> 00:05:19.360
When a developer is confronted with a computational bottleneck that
105
00:05:19.439 --> 00:05:23.079
has halted progress for years, the entire focus of their
106
00:05:23.120 --> 00:05:28.240
intellect is channeled into cracking that specific mathematical code. They
107
00:05:28.240 --> 00:05:31.600
are hyper focused on the how exactly, how do we
108
00:05:31.639 --> 00:05:34.800
optimize the parameter count? How do we reduce latency in
109
00:05:34.839 --> 00:05:35.680
the neural network.
110
00:05:35.800 --> 00:05:39.800
Right, Solving the how demands such intense granular concentration.
111
00:05:40.079 --> 00:05:43.800
Yeah, that there is quite literally no cognitive bandwidth left
112
00:05:43.959 --> 00:05:46.920
to process the why or the what next. Oh wow,
113
00:05:47.079 --> 00:05:50.040
By the time they successfully engineered the architecture to replicate
114
00:05:50.120 --> 00:05:53.639
human level reasoning, the mental energy required to ponder the
115
00:05:53.639 --> 00:05:57.399
immense ethical and social implications, it's just completely depleted.
116
00:05:57.519 --> 00:05:59.879
They fail to take the next step because their conceptual
117
00:06:00.000 --> 00:06:02.879
framework is strictly bound to that immediate technical hurdle.
118
00:06:03.000 --> 00:06:04.680
Yeah, they're so in the weeds they can't see the
119
00:06:04.759 --> 00:06:08.480
forest or the fact that the forest is potentially on fire.
120
00:06:08.600 --> 00:06:11.600
It reminds me of like assembling a team of the
121
00:06:11.639 --> 00:06:16.000
absolute most brilliant aerospace engineers to build a rocket capable
122
00:06:16.040 --> 00:06:17.079
of reaching light speed.
123
00:06:17.120 --> 00:06:18.319
Okay, I like this analogy.
124
00:06:18.560 --> 00:06:21.680
You have the best physicists, the best mechanics, and they
125
00:06:21.680 --> 00:06:25.040
are just intensely caught up in the sheer puzzle of
126
00:06:25.079 --> 00:06:25.920
the engine design.
127
00:06:26.000 --> 00:06:28.120
Right. They want to make it faster, more efficient, more.
128
00:06:28.040 --> 00:06:31.360
Powerful, and they pour years of their lives into optimizing
129
00:06:31.360 --> 00:06:35.240
the propulsion. But in that hyper focused state, they completely
130
00:06:35.279 --> 00:06:38.399
forget to install a steering wheel or brakes, or a
131
00:06:38.439 --> 00:06:40.839
life support system for the passenger.
132
00:06:40.439 --> 00:06:42.360
Which is a pretty big oversight. Right.
133
00:06:42.560 --> 00:06:46.079
The engine design works flawlessly, It does exactly what it
134
00:06:46.079 --> 00:06:49.040
was mathematically designed to do. But because they ignore the
135
00:06:49.160 --> 00:06:52.600
environment the engine operates in, the destination is a brick
136
00:06:52.639 --> 00:06:53.519
wall exactly.
137
00:06:53.639 --> 00:06:58.360
The engine represents raw, unaligned intelligence, and the steering wheel
138
00:06:58.399 --> 00:07:01.680
and the brakes those represent alignment with human well being
139
00:07:01.800 --> 00:07:03.000
and long term survival.
140
00:07:03.199 --> 00:07:06.639
And right now, all the prestige, all the venture capital,
141
00:07:06.680 --> 00:07:10.600
the academic accolades, it's entirely concentrated on building a bigger engine.
142
00:07:10.680 --> 00:07:13.279
The developers are solving for X, but they are not
143
00:07:13.480 --> 00:07:16.680
conceptually equipped to ask what happens to the entire alphabet
144
00:07:16.879 --> 00:07:19.120
once X is introduced into the ecosystem.
145
00:07:19.399 --> 00:07:23.040
So, if these developers are suffering from this exhausted imagination,
146
00:07:23.759 --> 00:07:26.360
if the breaks aren't even an afterthought, you have to ask,
147
00:07:26.560 --> 00:07:28.680
why do they keep pushing the accelerator?
148
00:07:28.879 --> 00:07:31.000
Right? Why is there no institutional pause?
149
00:07:31.279 --> 00:07:36.720
Exactly? To understand this compulsive, almost blind drive to innovate,
150
00:07:37.240 --> 00:07:42.000
the source material draws a direct, really unflinching comparison between
151
00:07:42.759 --> 00:07:46.079
the modern AI development race and the creation of the
152
00:07:46.120 --> 00:07:46.959
atomic bomb.
153
00:07:47.120 --> 00:07:50.000
Yeah, it points us toward the Manhattan Project as the
154
00:07:50.279 --> 00:07:54.639
historical precedent for developing this kind of existential technology.
155
00:07:54.240 --> 00:07:57.560
And that comparison provides vital context, doesn't it.
156
00:07:57.360 --> 00:07:59.279
It really does. It shows how the mechanics of a
157
00:07:59.319 --> 00:08:03.920
technological race dictate human behavior. The scientists working at Los
158
00:08:03.920 --> 00:08:08.240
Alamos in the nineteen forties they operated under an existential mandate.
159
00:08:07.959 --> 00:08:10.680
Because they possessed intelligence that the Germans had their own
160
00:08:10.759 --> 00:08:11.720
nuclear program.
161
00:08:11.879 --> 00:08:14.560
Right, So, the motivation to split the atom and weaponize
162
00:08:14.560 --> 00:08:18.240
it wasn't born out of abstract scientific curiosity or wanting
163
00:08:18.240 --> 00:08:18.959
to make a lot of mindy.
164
00:08:19.000 --> 00:08:19.959
It wasn't just fun to do.
165
00:08:20.360 --> 00:08:23.560
No, it was perceived as a strict wartime necessity for
166
00:08:23.639 --> 00:08:24.920
the survival of the free.
167
00:08:24.800 --> 00:08:28.879
World, and within that incredibly pressurized life or death environment.
168
00:08:29.360 --> 00:08:32.399
The source introduces this physicist, Leo Seialard.
169
00:08:32.639 --> 00:08:37.000
Sealard is crucial to Boston's framing here. He's described as
170
00:08:37.039 --> 00:08:41.120
this eccentric physicist who was actually the individual who first
171
00:08:41.159 --> 00:08:44.519
conceived the mechanics of a nuclear chain reaction.
172
00:08:44.759 --> 00:08:47.919
He understood the math of the explosion before the infrastructure
173
00:08:47.919 --> 00:08:49.399
to even build it existed.
174
00:08:49.559 --> 00:08:53.759
Yes, and what is remarkable about Selard is his immediate
175
00:08:53.799 --> 00:08:54.919
attempt at restraint.
176
00:08:55.399 --> 00:08:58.120
Right, he had this world altering revelation. But he didn't
177
00:08:58.159 --> 00:09:01.039
rush to publish a paper to get citations or secure
178
00:09:01.080 --> 00:09:01.720
his legacy.
179
00:09:01.879 --> 00:09:05.720
No, he actually engaged that imagination muscle. He extrapolated the
180
00:09:05.720 --> 00:09:10.480
physical reality of a nuclear chain reaction into its geopolitical consequence.
181
00:09:10.639 --> 00:09:13.919
He realized it could be catastrophically bad for civilization. He
182
00:09:13.960 --> 00:09:17.480
saw the engine and the lack of break simultaneously exactly.
183
00:09:17.200 --> 00:09:21.159
And his realization prompted this unprecedented action in the scientific community.
184
00:09:21.720 --> 00:09:24.440
Selord actively sought out his colleagues who were working in
185
00:09:24.519 --> 00:09:27.720
nuclear physics and attempted to persuade them to halt their experiments.
186
00:09:27.799 --> 00:09:30.480
He literally begged them not to pursue or publish their
187
00:09:30.519 --> 00:09:31.480
findings on fission.
188
00:09:31.720 --> 00:09:34.360
Yeah, he was essentially trying to organize a self imposed
189
00:09:34.399 --> 00:09:35.919
embargo on the knowledge.
190
00:09:36.120 --> 00:09:38.919
But the tragic punchline of the source material is that
191
00:09:39.120 --> 00:09:42.960
Silard achieved only quote unquote partial success.
192
00:09:42.799 --> 00:09:47.440
Partial success, which in this context is a tragedy. Some
193
00:09:47.480 --> 00:09:50.759
scientists agreed to delay their publications, but others.
194
00:09:50.440 --> 00:09:53.879
Refused because, as the source points out, some people were
195
00:09:53.960 --> 00:09:55.799
always going to want to be the ones that sort
196
00:09:55.799 --> 00:09:59.360
of innovate, and the concept of partial success in containing
197
00:09:59.399 --> 00:10:04.639
world alter technology is mathematically speaking identical to total failure.
198
00:10:04.720 --> 00:10:07.639
It is if even one innovator decides to publish, the
199
00:10:07.759 --> 00:10:09.480
cat is completely out of the bag.
200
00:10:09.679 --> 00:10:12.559
Right If one hundred scientists agree to lock the information
201
00:10:12.600 --> 00:10:14.759
in a vault, but the one hundred and first scientists
202
00:10:14.799 --> 00:10:18.399
decides to publish the paper, the embargo is meaningless. The
203
00:10:18.519 --> 00:10:19.679
knowledge propagates.
204
00:10:19.720 --> 00:10:22.600
You cannot undiscover the mechanism of the atom, just like.
205
00:10:22.600 --> 00:10:25.960
You cannot undiscover the architecture of a new neural network.
206
00:10:26.039 --> 00:10:29.639
Once the paper is out, the entire global community possesses
207
00:10:29.639 --> 00:10:30.440
the capability.
208
00:10:30.799 --> 00:10:36.480
The innovator's dilemma is fundamentally asymmetric. Containment requires unanimous, perpetual
209
00:10:36.559 --> 00:10:39.559
cooperation from every brilliant mind on the planet.
210
00:10:39.600 --> 00:10:43.440
Well, acceleration only requires a single actor to defect exactly.
211
00:10:43.480 --> 00:10:45.000
But I do have to push back a little on
212
00:10:45.039 --> 00:10:49.000
this Manhattan Project comparison for the listener to internalize today.
213
00:10:49.519 --> 00:10:51.399
Is it really a fair comparison?
214
00:10:51.679 --> 00:10:52.159
Hew? Do you mean?
215
00:10:52.639 --> 00:10:56.080
Well? During World War Two, those scientists were racing against
216
00:10:56.080 --> 00:10:59.639
a literal fascist empire. If they lost the race, the
217
00:10:59.639 --> 00:11:02.000
outcome home was global subjugation by Hitler.
218
00:11:02.200 --> 00:11:05.559
True, the stakes were uniquely high in a very visible way.
219
00:11:05.720 --> 00:11:09.279
But today AI developers at the major tech conglomerates are
220
00:11:09.360 --> 00:11:14.080
largely racing against rival corporations. The primary drivers are market share,
221
00:11:14.279 --> 00:11:17.840
stock valuations, being the first to deploy a commercial product.
222
00:11:17.960 --> 00:11:19.720
Yeah, it's a completely different landscape.
223
00:11:19.759 --> 00:11:23.480
There is no external, existential enemy forcing their hands. So
224
00:11:23.519 --> 00:11:27.039
doesn't that make the lack of ethical foresight today vastly
225
00:11:27.120 --> 00:11:30.200
less excusable than it was for Sealerd's colleagues.
226
00:11:30.360 --> 00:11:32.759
That is a very valid point. Yeah, the motivations have
227
00:11:32.840 --> 00:11:37.360
absolutely shifted from wartime survival to corporate dominance. Right. The
228
00:11:37.399 --> 00:11:40.279
critical takeaway from the source, however, is that while the
229
00:11:40.320 --> 00:11:43.919
moral justification has vanished, the mechanical pressure of the race
230
00:11:44.000 --> 00:11:45.080
remains identical.
231
00:11:45.480 --> 00:11:47.399
The fear of being second place.
232
00:11:47.879 --> 00:11:51.799
Exactly, whether it's to an enemination or a rival tech company,
233
00:11:51.879 --> 00:11:56.919
that fear drives the exact same reckless acceleration. The competitive
234
00:11:56.960 --> 00:12:00.320
ecosystem ensures that caution is penalized.
235
00:12:00.120 --> 00:12:03.399
Because if one company decides to halt development to spend
236
00:12:03.639 --> 00:12:07.600
say five years working on safety protocols and alignment.
237
00:12:07.639 --> 00:12:10.960
A rival company will simply bypass them, release the product,
238
00:12:11.039 --> 00:12:14.159
and capture the entire market. The structure of the race
239
00:12:14.480 --> 00:12:19.039
forcess participants to prioritize speed over safety regardless of whether
240
00:12:19.080 --> 00:12:22.000
the prize is global security or just quarterly profits.
241
00:12:22.320 --> 00:12:25.159
So we basically have a structural guarantee that the acceleration
242
00:12:25.240 --> 00:12:28.480
will continue, which forces us to look at exactly how
243
00:12:28.519 --> 00:12:29.840
this technology is advancing.
244
00:12:29.960 --> 00:12:32.879
Right, how are we moving from standard machines to these
245
00:12:32.919 --> 00:12:34.440
world altering intellects.
246
00:12:34.519 --> 00:12:36.679
We have to look at the mechanics of reverse engineering
247
00:12:36.679 --> 00:12:39.440
the brain, and the source points out this massive paradigm
248
00:12:39.440 --> 00:12:41.879
shift in how machines interact with physical space.
249
00:12:42.200 --> 00:12:47.399
Yeah, the transition from algorithmic programming to true artificial intelligence
250
00:12:47.799 --> 00:12:50.600
is best understood by looking at traditional robotics.
251
00:12:50.720 --> 00:12:56.519
Source material calls current robotic systems quote blind dumb things, blind.
252
00:12:56.279 --> 00:13:00.120
Dumb things, because a robotic arm operating in say an
253
00:13:00.159 --> 00:13:05.679
automotive factory, is just an execution of rigid state space programming.
254
00:13:05.759 --> 00:13:09.480
It's given precise three dimensional coordinates. It moves to coordinate A,
255
00:13:10.039 --> 00:13:13.519
engages the welding torch and moves to coordinate B, and.
256
00:13:13.440 --> 00:13:16.759
It does this flawlessly, but it has absolutely no awareness
257
00:13:16.759 --> 00:13:19.000
of its environment. If a different car chassis comes down
258
00:13:19.039 --> 00:13:21.519
the assembly line, or if the door is supposed to
259
00:13:21.519 --> 00:13:23.080
weld is entirely.
260
00:13:22.759 --> 00:13:25.360
Missing, it doesn't adapt. It just continues to weld the
261
00:13:25.399 --> 00:13:28.960
empty air because it's programming dictates that the coordinates are absolute.
262
00:13:29.080 --> 00:13:32.159
Exactly. If you want to change production, human engineers have
263
00:13:32.200 --> 00:13:35.080
to halt the line and painstakingly reprogram those coordinates.
264
00:13:35.200 --> 00:13:38.559
It possesses zero context. It is just playing a physical
265
00:13:38.600 --> 00:13:39.840
recording over and over.
266
00:13:40.039 --> 00:13:43.960
But the source introduces this new concept imitation learning, and
267
00:13:44.000 --> 00:13:46.600
that is the mechanism that completely shatters this limitation.
268
00:13:47.000 --> 00:13:50.240
Imitation learning. So instead of hard coding a robot, you
269
00:13:50.240 --> 00:13:52.559
could show it how to do something once or twice
270
00:13:52.759 --> 00:13:53.919
like you would a human baby.
271
00:13:54.039 --> 00:13:56.840
Right, it shifts a machine from following a script to
272
00:13:57.440 --> 00:13:59.960
observing an environment it learns to do it robot.
273
00:14:00.559 --> 00:14:03.120
So if a machine learns by watching, it means it's
274
00:14:03.159 --> 00:14:07.000
developing a generalized understanding of the physics and the intent
275
00:14:07.080 --> 00:14:08.120
within its environment.
276
00:14:08.320 --> 00:14:11.879
That's the key word intent. It figures out the goal
277
00:14:12.000 --> 00:14:15.480
of the action rather than just memorizing the mechanical coordinates
278
00:14:15.559 --> 00:14:15.879
used to.
279
00:14:15.840 --> 00:14:18.519
Achieve it, which is a profound lead when you think
280
00:14:18.559 --> 00:14:19.240
about it.
281
00:14:19.240 --> 00:14:21.799
It really is. The leap from mapping an action to
282
00:14:21.879 --> 00:14:25.600
mapping an intent changes everything. For example, if you demonstrate
283
00:14:25.720 --> 00:14:28.799
pouring coffee into a mug, the traditional robot tries to
284
00:14:28.840 --> 00:14:32.080
memorize the angle of your elbow. But the machine using
285
00:14:32.080 --> 00:14:36.799
imitation learning understands that the objective is transferring liquid from
286
00:14:36.799 --> 00:14:39.679
the pot into the mug. So if you move the
287
00:14:39.720 --> 00:14:41.240
mug to the other side of the table.
288
00:14:41.000 --> 00:14:44.480
The imitation learning model adjusts its physical approach to achieve
289
00:14:44.480 --> 00:14:45.080
the goal.
290
00:14:45.039 --> 00:14:48.960
Exactly whereas the traditional robot just pours scalding coffee onto
291
00:14:49.000 --> 00:14:50.639
the empty spot where the mug used to be.
292
00:14:50.960 --> 00:14:55.000
And the source explicitly connects this capability to a much
293
00:14:55.039 --> 00:14:58.240
grander objective, reverse engineering the human brain.
294
00:14:58.879 --> 00:15:02.919
By teaching systems to to generalize and infer intent, developers
295
00:15:02.919 --> 00:15:06.720
are laying the foundational architecture for cognitive replication.
296
00:15:06.600 --> 00:15:10.679
Which brings in Ray Kurzweil's predictions. The source notes that
297
00:15:10.919 --> 00:15:14.759
realizing Kurtzweil's stuff, which is like a radical tech future
298
00:15:14.799 --> 00:15:17.960
a one thousand year leap compacted into a few decades,
299
00:15:18.320 --> 00:15:21.600
is technically possible at what they call technical maturity.
300
00:15:21.679 --> 00:15:25.919
Technical maturity that's a theoretical endpoint where a civilization has
301
00:15:26.000 --> 00:15:30.919
essentially maximized its technological capabilities within the bounds of physics.
302
00:15:30.480 --> 00:15:33.799
Meaning we possess the computational power and the material science
303
00:15:33.879 --> 00:15:37.480
to perfectly simulate and manipulate reality at the atomic level.
304
00:15:37.639 --> 00:15:41.799
Right, But Bostrom provides two massive caveats regarding our ability
305
00:15:41.799 --> 00:15:43.240
to actually reach this future.
306
00:15:43.559 --> 00:15:46.799
The first caveat being do we actually reach technological maturity
307
00:15:46.840 --> 00:15:47.960
without destroying ourselves?
308
00:15:48.039 --> 00:15:52.360
First? Exactly, given the exhausted imagination of developers and the
309
00:15:52.399 --> 00:15:55.480
relentless pressure of the innovator's dilemma we just talked about,
310
00:15:55.759 --> 00:15:58.679
surviving our own technological adolescens is the ultimate filter.
311
00:15:58.960 --> 00:16:01.799
The tools we are built are powerful enough to terminate
312
00:16:01.799 --> 00:16:03.320
the development curve permanently.
313
00:16:03.559 --> 00:16:03.799
Yes.
314
00:16:04.000 --> 00:16:07.240
And the second caveat is, even if we survive the
315
00:16:07.279 --> 00:16:11.159
creation of AI, are we actually interested in using our
316
00:16:11.159 --> 00:16:14.200
technology for that specific purpose at that time?
317
00:16:14.440 --> 00:16:17.600
Right? It's like unlocking the final tier of massive video
318
00:16:17.639 --> 00:16:18.399
game tech trees.
319
00:16:18.559 --> 00:16:22.519
I love that analogy. You spend hours gathering resources, leveling
320
00:16:22.600 --> 00:16:26.000
up your civilization, and you finally unlock the ability to
321
00:16:26.080 --> 00:16:27.320
build a digital brain.
322
00:16:27.480 --> 00:16:29.600
But just because you have the capability to spend your
323
00:16:29.639 --> 00:16:31.559
points to build it doesn't mean you will.
324
00:16:31.759 --> 00:16:34.840
By the time humanity achieves technical maturity, our priorities as
325
00:16:34.840 --> 00:16:38.480
a species could be entirely different. We might realize simulating
326
00:16:38.519 --> 00:16:40.840
a human mind is unnecessary.
327
00:16:40.279 --> 00:16:43.639
Or unethical, or simply uninteresting compared to other avenues of
328
00:16:43.679 --> 00:16:46.720
exploration we haven't even conceived of yet. Capability does not
329
00:16:46.799 --> 00:16:47.919
necessitate execution.
330
00:16:48.519 --> 00:16:50.919
But and this is a big butt, if we do
331
00:16:51.000 --> 00:16:54.240
decide to proceed with reverse engineering the mind using mechanisms
332
00:16:54.240 --> 00:16:58.720
like imitation learning, we are opening up a terrifying philosophical puzzle.
333
00:16:58.840 --> 00:17:02.559
The ethical weight of imitation learning is staggering because when
334
00:17:02.559 --> 00:17:05.960
a machine learns by observing human behavior, it isn't just
335
00:17:06.039 --> 00:17:09.279
absorbing our physical movements. It absorbs our flaws exactly. It
336
00:17:09.319 --> 00:17:14.720
absorbs our context, our unspoken societal rules, our biases, our prejudices,
337
00:17:14.920 --> 00:17:16.480
and our unstated intentions.
338
00:17:16.799 --> 00:17:20.599
So if we create a superintelligence that infers how to
339
00:17:20.680 --> 00:17:26.359
behave by watching us, the big fear is amplified massively.
340
00:17:26.039 --> 00:17:29.319
Right, because the system might learn and optimize our most
341
00:17:29.319 --> 00:17:32.960
descriptive traits just as efficiently as our cooperative ones.
342
00:17:33.200 --> 00:17:36.640
Oh wow. So if we somehow successfully navigate all of
343
00:17:36.640 --> 00:17:40.720
those risks, if we reach technological maturity, decide to pursue
344
00:17:40.720 --> 00:17:43.400
the simulation and align its intents so it doesn't.
345
00:17:43.079 --> 00:17:44.599
Destroy us, that's a lot of ifs.
346
00:17:44.759 --> 00:17:46.440
It is a lot of ifs. But if we do,
347
00:17:46.519 --> 00:17:48.640
the source material points out, it forces us to ask
348
00:17:48.680 --> 00:17:51.839
a terrifying, reality breaking question about our own existence.
349
00:17:52.079 --> 00:17:55.720
Yes, because successfully reverse engineering the brain forces us to
350
00:17:55.759 --> 00:17:58.200
grapple with the substrate independence thesis.
351
00:17:58.400 --> 00:18:01.480
The substrate independence these says, can you explain that concept
352
00:18:01.519 --> 00:18:02.680
clearly for the listener?
353
00:18:02.920 --> 00:18:06.920
Absolutely. It's this concept from the philosophy of mind. It
354
00:18:07.039 --> 00:18:10.839
posits that a mind or consciousness does not strictly require
355
00:18:10.880 --> 00:18:15.359
a biological, carbon based quote unquote meat brain to exist.
356
00:18:15.759 --> 00:18:19.440
We just inherently assume that consciousness is a property exclusive
357
00:18:19.480 --> 00:18:22.799
to biological neurons and synapses, because that's the only format
358
00:18:22.839 --> 00:18:24.079
we've ever observed.
359
00:18:23.640 --> 00:18:27.599
It in, right, But the thesis argues that consciousness is
360
00:18:27.759 --> 00:18:31.680
not a byproduct of the biological material itself. It's an
361
00:18:31.680 --> 00:18:35.039
emergent property of the information processing that is happening.
362
00:18:35.200 --> 00:18:37.400
It's separating the software from the hardware.
363
00:18:37.599 --> 00:18:40.720
Exactly. If you have a different substrate like silicon chips
364
00:18:40.839 --> 00:18:44.720
or quantum processors or abstract code inside a computer, and
365
00:18:44.759 --> 00:18:48.160
that non biological substrate can process information with the exact
366
00:18:48.200 --> 00:18:50.160
same complexity as a human brain, the.
367
00:18:50.119 --> 00:18:53.200
Exact same architecture, the exact same feedback loops.
368
00:18:52.960 --> 00:18:56.160
Then the thesis states that the non biological substrate will
369
00:18:56.160 --> 00:18:59.480
generate genuine consciousness, it can operate on a different substrate,
370
00:18:59.559 --> 00:18:59.720
and the.
371
00:18:59.720 --> 00:19:02.279
Impls location for simulated brains is absolute.
372
00:19:02.519 --> 00:19:06.920
If consciousness is substrate independent, then a perfectly reverse engineered
373
00:19:06.960 --> 00:19:10.279
brain running on a computer isn't just a clever parlor trick.
374
00:19:10.559 --> 00:19:14.000
It's not just a complex imitation. That digital entity is
375
00:19:14.039 --> 00:19:18.440
experiencing reality. It possesses a subjective perspective.
376
00:19:18.599 --> 00:19:22.359
It feels joy, it experiences pain, It forms memories. For
377
00:19:22.400 --> 00:19:25.119
all intents and purposes, the code is alive.
378
00:19:25.519 --> 00:19:29.000
And this logical progression is what serves as the launch
379
00:19:29.039 --> 00:19:33.079
pad for Nick Bostrom's famous simulation hypothesis.
380
00:19:32.480 --> 00:19:35.680
Which fundamentally challenges our entire concept of reality.
381
00:19:35.799 --> 00:19:40.039
The logic of the simulation hypothesis is terrifyingly sound once
382
00:19:40.079 --> 00:19:42.559
you accept the premise of substrate independence.
383
00:19:42.680 --> 00:19:46.599
It really is. If you accept imitation, learning, then reverse engineering,
384
00:19:46.720 --> 00:19:51.079
then substrate independence. The simulation hypothesis is almost inescapable.
385
00:19:51.119 --> 00:19:52.799
So walk us through the math of that, because it's
386
00:19:52.880 --> 00:19:53.519
mind bending.
387
00:19:53.680 --> 00:19:56.079
Okay, So if consciousness can exisit as code, we then
388
00:19:56.160 --> 00:19:59.559
look at the timeline a technologically mature civilization, whether it's
389
00:19:59.599 --> 00:20:01.440
our des SA and thence millions of years from now
390
00:20:01.920 --> 00:20:04.720
or an alien race, they will possess computing power that
391
00:20:04.839 --> 00:20:06.839
is practically incomprehensible to us.
392
00:20:06.880 --> 00:20:09.519
They could utilize the energy output of entire stars to
393
00:20:09.640 --> 00:20:11.599
power computers the size of planets.
394
00:20:11.799 --> 00:20:16.279
Right with that level of computational density, the energetic cost
395
00:20:16.440 --> 00:20:19.839
of running a simulation of human history, what Bostrom calls
396
00:20:19.880 --> 00:20:22.839
an ancestor simulation, would be microscopic.
397
00:20:23.200 --> 00:20:27.279
A mature civilization could theoretically run billions or trillions of
398
00:20:27.400 --> 00:20:29.279
highly detailed simulations, and.
399
00:20:29.240 --> 00:20:33.559
Each of those simulations would contain billions of conscious simulated
400
00:20:33.640 --> 00:20:34.400
human minds.
401
00:20:34.480 --> 00:20:39.359
And the mathematical consequence of this capability is just staggering, because.
402
00:20:39.079 --> 00:20:42.480
If a civilization runs billions of simulations, the number of
403
00:20:42.559 --> 00:20:47.079
simulated digital minds in existence would vastly outnumber the original
404
00:20:47.079 --> 00:20:49.559
biological minds that lived in the base reality.
405
00:20:49.599 --> 00:20:53.160
The ratio would be astronomical billions or trillions of simulated
406
00:20:53.200 --> 00:20:57.359
minds to every one biological mind. Therefore, statistically speaking, if
407
00:20:57.400 --> 00:21:00.440
you are a conscious entity currently experiencing a reality, the
408
00:21:00.440 --> 00:21:04.000
probability that you are the one original biological entity living
409
00:21:04.039 --> 00:21:05.720
in base reality plummet.
410
00:21:05.839 --> 00:21:09.799
It's practically zero. The math forces the conclusion that we're
411
00:21:10.039 --> 00:21:12.799
almost certainly living inside one of those billions of ancestors
412
00:21:12.839 --> 00:21:13.720
simulations right.
413
00:21:13.640 --> 00:21:16.039
Now, it is incredibly difficult to sit with that idea.
414
00:21:16.160 --> 00:21:18.000
It induces a lot of existential dread.
415
00:21:18.119 --> 00:21:21.720
Yeah, because to raise a philosophical pushback on behalf of
416
00:21:21.720 --> 00:21:23.920
the listener here, If I have to accept the substrate
417
00:21:23.960 --> 00:21:27.480
independence thesis that my thoughts and feelings are just softler
418
00:21:27.599 --> 00:21:32.319
running on a biological hard drive, doesn't that completely devalue
419
00:21:32.359 --> 00:21:33.440
the human experience.
420
00:21:34.039 --> 00:21:36.160
That's the immediate reaction most people.
421
00:21:35.920 --> 00:21:39.039
Have, right if we're just rendered algorithms. Does falling in
422
00:21:39.079 --> 00:21:43.200
love actually matter? Does feeling grief for a lost loved
423
00:21:43.240 --> 00:21:45.960
one mean anything at all? It feels like a pathway
424
00:21:46.000 --> 00:21:48.279
straight into profound nihilism.
425
00:21:47.960 --> 00:21:51.519
And that sense of existential demotion is so common. Stripping
426
00:21:51.559 --> 00:21:54.039
away the physical permanence of the universe and replacing it
427
00:21:54.039 --> 00:21:57.039
with rendered parameters challenges our core sense of meaning.
428
00:21:57.200 --> 00:21:57.920
It really does.
429
00:21:58.119 --> 00:22:01.799
However, the source material actually pivot Here. Bostrom turns the
430
00:22:01.839 --> 00:22:03.319
simulation theory on its head.
431
00:22:03.599 --> 00:22:06.559
Instead of being a depressing thought, there's an upside.
432
00:22:06.039 --> 00:22:10.480
A surprising psychological upside. Yes. The final concept explored is
433
00:22:10.519 --> 00:22:13.480
the psychological benefits for interacting with life. As if it's
434
00:22:13.480 --> 00:22:14.519
a simulation.
435
00:22:14.279 --> 00:22:16.759
How does adopting the framework that we might be lines
436
00:22:16.799 --> 00:22:19.599
of code alleviate the anxiety of living well.
437
00:22:19.640 --> 00:22:22.039
It cannex back directly to the big fear we discussed
438
00:22:22.039 --> 00:22:26.359
at the very beginning. Human beings are deeply encumbered by
439
00:22:26.400 --> 00:22:30.680
existential anxiety. We carry the terror of our own mortality,
440
00:22:30.720 --> 00:22:34.359
the permanence of our failures, the crushing way to the unknown.
441
00:22:34.480 --> 00:22:37.960
We are fear driven creatures in many ways, right, But.
442
00:22:37.960 --> 00:22:40.400
When you entertain the idea that life is a simulation,
443
00:22:41.000 --> 00:22:45.880
it fundamentally reframes the stakes of those daily anxieties. Adopting
444
00:22:45.880 --> 00:22:48.400
this framework changes how one interacts.
445
00:22:47.920 --> 00:22:51.240
With the world, because accepting substrate independence means your feelings
446
00:22:51.279 --> 00:22:54.400
are still entirely real. Right, Your joy and pain are
447
00:22:54.440 --> 00:22:57.400
genuine experiences of your consciousness.
448
00:22:56.799 --> 00:23:01.519
Exactly, but the context surrounding those experiences shif It removes
449
00:23:01.559 --> 00:23:04.960
the absolute permanent weight of the biological universe.
450
00:23:05.039 --> 00:23:08.519
It encourages a curiosity driven approach to life rather than
451
00:23:08.559 --> 00:23:09.680
a fear driven one.
452
00:23:09.880 --> 00:23:12.880
Rather than inducing nihilism. It offers a strange sense of
453
00:23:13.000 --> 00:23:16.200
liberation or perspective. It alleviates the big fear.
454
00:23:16.400 --> 00:23:19.279
It's like playing an incredibly difficult, high stakes video game.
455
00:23:19.400 --> 00:23:21.359
Oh, that's a great way to ground this for the listener.
456
00:23:21.680 --> 00:23:25.079
Think about it. When you are deeply immersed in a
457
00:23:25.119 --> 00:23:29.279
demanding digital environment and you temporarily forget that you are
458
00:23:29.319 --> 00:23:33.519
engaging with a game, the stress becomes agonizing.
459
00:23:32.920 --> 00:23:35.759
Your heart rate spikes, your breathing becomes shallow.
460
00:23:35.960 --> 00:23:39.920
The fear of your character failing paralyzes your decision making.
461
00:23:40.200 --> 00:23:44.599
You become totally risk averse because the simulated danger feels real. Right,
462
00:23:45.119 --> 00:23:47.640
But the moment you take a breath and remember, oh wait,
463
00:23:47.720 --> 00:23:51.039
this is just a simulation, you don't immediately throw the
464
00:23:51.039 --> 00:23:52.599
controller away and stop playing.
465
00:23:52.759 --> 00:23:56.400
Knowing it's constructed doesn't drain the experience of its value exactly.
466
00:23:56.440 --> 00:23:59.400
In fact, you usually play better. You become more creative
467
00:23:59.440 --> 00:24:01.799
with your problem solving. You test the boundaries of the
468
00:24:01.839 --> 00:24:05.599
physics engine. You take larger risks and try new strategies
469
00:24:05.640 --> 00:24:09.599
because that paralyzing dread of permanent failure has been lifted.
470
00:24:10.200 --> 00:24:14.400
The realization that the environment is constructed frees the consciousness
471
00:24:14.440 --> 00:24:18.559
to engage with the environment more dynamically. Recognizing the simulation
472
00:24:18.720 --> 00:24:20.920
provides an intellectual buffer against the void.
473
00:24:21.039 --> 00:24:23.720
It allows you to appreciate the complexity of the rendering
474
00:24:23.960 --> 00:24:26.720
without being crushed by the permanence of it all. So
475
00:24:27.000 --> 00:24:30.200
we've gone on a massive journey today through Bostrom's theories.
476
00:24:30.640 --> 00:24:34.640
It forces us to continuously reevaluate our position in the universe.
477
00:24:34.960 --> 00:24:37.920
We started by observing the watchmakers of our era, the
478
00:24:38.000 --> 00:24:41.599
developers building AI, and examine how their hyper focus blinds
479
00:24:41.640 --> 00:24:44.519
them to the societal impact of the intelligence they're engineering.
480
00:24:44.799 --> 00:24:48.960
We contextualize that exhausted imagination through the historical lens of
481
00:24:49.039 --> 00:24:51.559
Leo Cilard and the Manhattan Project.
482
00:24:51.559 --> 00:24:55.759
Illustrating how the innovator's dilemma guarantees that technological acceleration will
483
00:24:55.759 --> 00:25:00.279
continue since partial success and containment is mathematically impossible.
484
00:25:00.640 --> 00:25:04.440
We trace the mechanical evolution from blind dumb things to
485
00:25:04.480 --> 00:25:07.920
the imitation learning that sets the stage for reverse engineering
486
00:25:08.000 --> 00:25:08.359
the brain.
487
00:25:08.559 --> 00:25:11.680
And then we explored the profound philosophical consequences of that
488
00:25:11.759 --> 00:25:15.839
replication through the substrate independence thesis, leading to the mind
489
00:25:15.839 --> 00:25:18.240
bending reality of the simulation hypothesis.
490
00:25:18.440 --> 00:25:21.400
And finally we found a strange psychological freedom in that
491
00:25:21.519 --> 00:25:25.880
exact hypothesis. Understanding these interconnected concepts gives you, the listener,
492
00:25:26.119 --> 00:25:29.720
a shortcut to being deeply well informed about the existential
493
00:25:29.720 --> 00:25:31.799
crosshairs humanity currently sits in.
494
00:25:32.079 --> 00:25:35.720
You're no longer just reacting to isolated headlines about new
495
00:25:35.720 --> 00:25:39.519
AI models. You can see the philosophical endgame driving.
496
00:25:39.160 --> 00:25:41.480
Them to conclude this deep dive. I want to leave
497
00:25:41.519 --> 00:25:45.160
you with a final provocative thought to mull over. Think
498
00:25:45.240 --> 00:25:49.400
back to those developers racing to build artificial superintelligence. Okay,
499
00:25:49.759 --> 00:25:52.319
if we are indeed living in a simulation created by
500
00:25:52.359 --> 00:25:57.160
a technologically mature civilization, and we are currently racing with
501
00:25:57.279 --> 00:26:01.759
exhausted imaginations and unstoppable competitive of momentum to build our
502
00:26:01.799 --> 00:26:03.920
own super intelligent AI, are.
503
00:26:03.759 --> 00:26:06.680
We simply acting out the historical code of our creators?
504
00:26:06.839 --> 00:26:10.200
Exactly? Is our rush to build AI just a pre
505
00:26:10.240 --> 00:26:14.200
programmed variable? And when our artificial intelligence finally wakes up,
506
00:26:14.240 --> 00:26:17.440
achieving super intelligence and surpassing our comprehension, will it be
507
00:26:17.480 --> 00:26:19.960
the end of our world or just the required trigger
508
00:26:20.000 --> 00:26:22.759
to launch the beginning of the next simulation down the chain.
509
00:26:22.880 --> 00:26:25.480
Wow, that is quite the thought to end on.
510
00:26:25.839 --> 00:26:27.720
Thank you all for taking this deep dive with us
511
00:26:27.759 --> 00:26:28.079
today
