AI Whistleblower: "We Are Not Ready For What''s Coming In 2027!" — Transcript

Full transcript

1. 0:00the potential good that can come from
2. 0:02this technology is effectively [music]
3. 0:05infinite. But I would say that the harm
4. 0:07is probably 10 times it's a bigger
5. 0:09infinity and I really think that's
6. 0:12that's where we are. We're talking about
7. 0:14something that is going to fundamentally
8. 0:16alter what humans are with no
9. 0:20plan. You know, when you have a massive
10. 0:22economic contraction that arises from
11. 0:25the fact that a huge number of people
12. 0:27are out of work is that the elites start
13. 0:30looking at those people and thinking,
14. 0:31well, we don't really need them anyway.
15. 0:33And so
16. 0:33>> That is Bret Weinstein. He is an
17. 0:35evolutionary biologist and that
18. 0:37background is exactly why you should
19. 0:39listen to him. Most of the people
20. 0:41building AI are engineers. They look at
21. 0:44code and hardware. Biologists look at
22. 0:47ecosystems. And when a biologist looks
23. 0:49at what we are building, they don't see
24. 0:51a tool. They see a new apex predator
25. 0:54entering the food chain. He says the
26. 0:56potential for good here is infinite, but
27. 0:58the potential for harm is 10 times
28. 1:00bigger.
29. 1:02And his biggest fear isn't that the AI
30. 1:04will become conscious and hate us. It's
31. 1:06that human elites will use it to strip
32. 1:08away our economic value. Okay, now I'll
33. 1:11play some clips and explain as we go.
34. 1:13And please don't forget to like and
35. 1:15subscribe if you want to keep up with
36. 1:17where this whole AI stuff is heading.
37. 1:19>> is the AI agent will determine when it's
38. 1:22finished executing.
39. 1:24Uh today AI agents can run for anywhere
40. 1:26between, you know, [clears throat] 30
41. 1:28seconds to 30 minutes.
42. 1:30Uh there's a recent paper that came out
43. 1:32that's showing that every 7 months the
44. 1:35number of minutes that the agent can run
45. 1:38for is doubling.
46. 1:39So we're at like 30 minutes now, in 7
47. 1:42months we're going to be at an hour,
48. 1:43then, you know, 2 hours. Pretty soon
49. 1:45we're going to be at days. And at that
50. 1:47point, you know, AI agent is doing
51. 1:49labor. It's doing kind of human-like
52. 1:51labor. And actually OpenAI's new model
53. 1:5403 beat the expectation. So it it sort
54. 1:57of doubled coherence over long horizon
55. 1:59tasks in just three or four months.
56. 2:02So, we're in this massive and I mean
57. 2:04this looks this exponential graph, you
58. 2:07know, that that shows you the massive
59. 2:09trend we're on.
60. 2:11>> Technologists often master highly
61. 2:13complicated systems and they know, you
62. 2:16know, for example, a computer
63. 2:18is a perfectly predictable system inside
64. 2:20there's it's deterministic. But to most
65. 2:22of us it functions it's is mysterious
66. 2:25enough that it feels like a complex
67. 2:27system.
68. 2:28And if you're in the position of having
69. 2:31mastered highly complicated systems and
70. 2:33you look at complex systems and you
71. 2:34think it's a natural extension, you
72. 2:36failed to anticipate
73. 2:37just how unpredictable they are. So,
74. 2:40even if it is true that today there are
75. 2:43limits to what these machines can do
76. 2:46based on their training data, I think
77. 2:49the problem is
78. 2:51to see what's going to happen, you
79. 2:53really want to start thinking of this as
80. 2:54the evolution of a new species that will
81. 2:56continue to evolve. It will partially be
82. 2:59shaped by what we ask it to do, the
83. 3:01direction we lead it, and it will
84. 3:02partially be shaped by things we don't
85. 3:04understand. So,
86. 3:06how does this computer that that we have
87. 3:08work? Well, one of the things that it
88. 3:10does is we plug them into each other
89. 3:13using language. It's almost as if you
90. 3:15plug an ethernet cable in between human
91. 3:18minds and that means that the cognitive
92. 3:21potential exceeds the sum of the
93. 3:23individual minds in question.
94. 3:26Your AIs are going to do that and that
95. 3:27means that our ability to say what they
96. 3:29are capable of does not come down to,
97. 3:31well, we didn't train it on that data.
98. 3:33As they begin to interact, that feedback
99. 3:36is going to take them to capabilities we
100. 3:38don't anticipate and may not even
101. 3:39recognize once they become present.
102. 3:41That's one of my fears. This is an
103. 3:43evolving creature and
104. 3:45it's not even an animal.
105. 3:47>> I think if your job is as routine as it
106. 3:51comes, your job is gone in the next uh
107. 3:53couple of years. So, meaning if you
108. 3:56in in those jobs are for example
109. 3:58uh quality assurance jobs, data entry
110. 4:01jobs, or you're sitting in front of a
111. 4:02computer and you're supposed to click uh
112. 4:04and and type things in a certain order,
113. 4:06operator in those technologies are
114. 4:08coming in the market really quickly and
115. 4:09those are going to displace a lot of a
116. 4:11lot of uh labor.
117. 4:13>> Accountants?
118. 4:14>> Accountants.
119. 4:14>> Lawyers?
120. 4:15>> Yes.
121. 4:16>> We hear the term AI agent and we picture
122. 4:18a smarter chatbot, but the reality of
123. 4:20what that means is way more intense.
124. 4:23Right now, these agents can only run for
125. 4:25about 30 minutes before they get
126. 4:27confused or need a human to step in, but
127. 4:30that window is doubling every 7 months.
128. 4:33By next year, these systems will be
129. 4:35operating autonomously for days. It
130. 4:37doesn't get tired, it doesn't need a
131. 4:39lunch break, and it doesn't care about
132. 4:41your weekend. It just relentlessly
133. 4:43pursues whatever goal you gave it. And
134. 4:46the part that really messes with my head
135. 4:48is when the conversation shifts to how
136. 4:50these AIs network together. We assume we
137. 4:53are the smartest things on the planet
138. 4:55because we can share ideas with each
139. 4:56other, but these systems are literally
140. 4:58plugging into each other, sharing data
141. 5:01at a scale we can't even comprehend.
142. 5:03They are building a collective
143. 5:04intelligence that is going to develop
144. 5:06capabilities nobody programmed them for.
145. 5:09We are handing over the keys to a
146. 5:10species we don't understand. Listen to
147. 5:13this.
148. 5:14>> There's this image that I saw of New
149. 5:16York City uh in 1900 and every single
150. 5:21vehicle on the street is a horse and
151. 5:22cart. And then 13 years later, the same
152. 5:25photo from the same vantage point and
153. 5:27every single vehicle on the street is a
154. 5:29car. And in 13 years, all the horses had
155. 5:31been removed and cars had been put in
156. 5:33place.
157. 5:34And um
158. 5:35if you had have interviewed the horses
159. 5:37in 1900 and said, uh how do you feel
160. 5:39about your level of confidence uh in in
161. 5:41the world? The horses would have said,
162. 5:43well, we've been part of uh humanity for
163. 5:45you know, horse and hoof, uh hand and
164. 5:47hoof for many, many years, uh for for
165. 5:49thousands of years, there's one horse
166. 5:51for every three humans. Like, how bad
167. 5:53could it be, you know, we'll always have
168. 5:54a special place. We'll always be part of
169. 5:57society.
170. 5:58Um and little did the horses realize
171. 6:00that that was not the case, that the
172. 6:02horses were going to be put out of of uh
173. 6:04business very, very rapidly.
174. 6:06And
175. 6:07for reason through analogy, you know,
176. 6:08there's a lot of us who are now sitting
177. 6:10there going, "Hey, wait a second, does
178. 6:11this make me a horse in 1900?"
179. 6:14>> If you had a technology that was
180. 6:17excellent at at uh enhancing your
181. 6:19capacity to wield power, then
182. 6:22open-sourcing it is a failure to
183. 6:25capitalize on that, and that the most
184. 6:27remunerative use is to keep it private,
185. 6:30and then either sell the ability to
186. 6:33manipulate elections to people who want
187. 6:34to do so, or sell the ability to have it
188. 6:36kept off the table for people who don't.
189. 6:39And I would expect that that's probably
190. 6:41what's going on. There's no
191. 6:43If you have a technology as
192. 6:43transformative as this, giving it away
193. 6:46for free is counterintuitive, which
194. 6:48leaves those of us in the public more or
195. 6:50less at the mercy of the people who have
196. 6:53it. So,
197. 6:55I I don't see the reason for comfort.
198. 6:57There we are at the dawn of this radical
199. 7:02transformation of humans that by its
200. 7:05very nature as a truly complex and
201. 7:09emergent innovation, nobody on Earth can
202. 7:11predict what's going to happen. We can
203. 7:14We're on the event horizon of something.
204. 7:16So, what do we think about the folks who
205. 7:18are going to be displaced from the top?
206. 7:20What are they going to think about all
207. 7:21these people building all of these
208. 7:22highly competitive businesses? And are
209. 7:23they going to find a way to do, you
210. 7:26know, what venture capital has done, or
211. 7:28what record producers have done? What
212. 7:30they're going to do is they're going to
213. 7:31take their superior position at the top,
214. 7:33and they are going to take most of the
215. 7:35wealth that is produced by all of these
216. 7:36people who have these ideas that in a
217. 7:38proper market would actually create
218. 7:40businesses for them, and they're going
219. 7:42to parasitize
220. 7:42>> Now, the horse analogy is terrifying
221. 7:45because it exposes the illusion of job
222. 7:47security. In 1900, the horse was the
223. 7:50backbone of the economy. There were
224. 7:52millions of them. If you were a horse,
225. 7:55you probably felt pretty safe because
226. 7:57the world was literally built around
227. 7:59you. But in just over a decade, the
228. 8:01automobile completely replaced them. The
229. 8:04horses didn't see it coming and they had
230. 8:06no plan B.
231. 8:08I think a lot of us are the horses right
232. 8:10now. We think our jobs are safe because
233. 8:12they require thinking. But if these
234. 8:14machines can outthink a lawyer or a
235. 8:16programmer, we are all in the same boat.
236. 8:19And on top of that, we have this naive
237. 8:21assumption that the companies building
238. 8:23this tech are keeping it private just to
239. 8:25protect us from bad actors. Let's be
240. 8:27real. If you had a technology that could
241. 8:30predict markets, manipulate elections,
242. 8:32or replace entire workforces, why would
243. 8:35you give that away for free? The people
244. 8:37who control these models will hold all
245. 8:39the leverage, and the rest of us are
246. 8:41completely at their mercy. Listen.
247. 8:43>> Now, if human beings
248. 8:45among our gifts are fundamentally
249. 8:47technological, whether we're talking
250. 8:48about
251. 8:49quantum computing or flint knapping
252. 8:52an arrowhead,
253. 8:55what has happened to us that has created
254. 8:58the growing, spreading, morphing
255. 9:00dystopia is a process that Heather and I
256. 9:04in our book, A Hunter-Gatherer's Guide
257. 9:05to the 21st Century, call hyper-novelty.
258. 9:08Hyper-novelty is
259. 9:11the fact of the rate of change outpacing
260. 9:14our capacity to adapt to change. And we
261. 9:16are already well past the threshold here
262. 9:19where the world that we are young in is
263. 9:21not the world that we are adults in, and
264. 9:23that mismatch is making us sick across
265. 9:26multiple different domains. So, the the
266. 9:29question that I ask is is the change
267. 9:31that you're talking about going to
268. 9:33reduce the rate of change, in which case
269. 9:36we could build a world that would start
270. 9:38meeting human needs better, open
271. 9:39opportunities for pursuing meaningful
272. 9:42work.
273. 9:43Or is it going to accelerate the rate of
274. 9:44change, which is, in my opinion,
275. 9:46guaranteed to make us worse off. So, if
276. 9:49it was a one-time shift, right? AI is
277. 9:51going to dawn, it's going to open all
278. 9:53sorts of new opportunities, there's
279. 9:54going to be a tremendous amount of
280. 9:55disruption, but from that we'll be able
281. 9:57to build a world. Is that world going to
282. 9:59be stable, or is it going to be just,
283. 10:01you know, one event horizon after the
284. 10:03next? If it's the latter, then it
285. 10:05effectively says what it does to the
286. 10:07humans, which is it it's going to
287. 10:09dismantle us.
288. 10:11>> How would we change? How would we slow
289. 10:12the rate of change?
290. 10:13>> Well, I I mean, you can you can you can
291. 10:15be the Amish, right? You can be the
292. 10:16Amish and living your own communities,
293. 10:18and and I would assume some people would
294. 10:20would want that.
295. 10:21>> Well, I'm
296. 10:22you know, when Heather and I wrote our
297. 10:24book, I wanted the first chapter to be
298. 10:26are the Amish right?
299. 10:28And the answer is they can't be exactly
300. 10:30right because they picked an arbitrary
301. 10:32moment to step off the escalator. But
302. 10:33are they right that there's something
303. 10:34dangerous about this continuing pattern
304. 10:36of technological change? Clearly they
305. 10:38are.
306. 10:38>> What do the Amish do for anyone that
307. 10:40doesn't know?
308. 10:41>> The Amish live as if it was what? 1880?
309. 10:48So, they live in a they don't use cars,
310. 10:50they I think they do have phones, but
311. 10:52they do not have electricity. Basically,
312. 10:55they they voluntarily accept a techno
313. 10:58they're basically a Luddite community,
314. 11:00and they have turned out to fare
315. 11:04surprisingly well against many of the
316. 11:05things that have upended
317. 11:07modern society.
318. 11:08>> COVID was one of them, right?
319. 11:09>> Yeah, COVID they did beautifully.
320. 11:10>> Quite happy people.
321. 11:11>> Very low autism rates. They they have
322. 11:13all sorts of advantages. So, anyway, I'm
323. 11:15not arguing that we should live like the
324. 11:16Amish, I don't see that, but I do think
325. 11:18the idea that they had an insight, which
326. 11:19was you need to step off that escalator
327. 11:22cuz you're just going to keep making
328. 11:23yourselves sicker, is probably right.
329. 11:25Now, maybe
330. 11:26this is a one-time shift, we've stepped
331. 11:29over the event horizon, we're going to
332. 11:31be living in the AI world, and maybe if
333. 11:33we're careful about it, we can figure
334. 11:34out how to turn that landscape of
335. 11:37infinite possibility that you're
336. 11:38describing
337. 11:40into a place that doesn't change. That
338. 11:43you always have the opportunity to
339. 11:46decide what needs to be done, but that
340. 11:48none of the the that living over that
341. 11:50event horizon is not an ever-changing
342. 11:52process. It's just the next frontier.
343. 11:55>> My my fear coming stemming from
344. 11:57technology
345. 11:59uh and AI
346. 12:01is that this is a runaway process, and
347. 12:04that that runaway process is going to
348. 12:06interface very badly with some latent
349. 12:08human programs. That in effect, the need
350. 12:11for workers
351. 12:13largely disappears, and the people who
352. 12:16are at the head of the processes that
353. 12:18result in that elimination for the need
354. 12:20for workers start talking about useless
355. 12:22eaters. Maybe they come up with a new
356. 12:23term this time.
357. 12:24>> herd.
358. 12:25>> Yep, or they allow it to be thinned or
359. 12:26something.
360. 12:27>> Right, I've heard you talk about the
361. 12:28five key
362. 12:30concerns you have or the five key
363. 12:32threats you have before. Could you name
364. 12:33those five?
365. 12:34>> So, the first one is the one I worry
366. 12:36least about. I don't worry zero about
367. 12:38it, but I worry least about it, which is
368. 12:39the malevolent AI that the doomers are
369. 12:42so focused on.
370. 12:44The second one is the idea that, you
371. 12:46know, an AI can be misaligned not
372. 12:48because it has divergent interests, but
373. 12:50because it just misunderstands what
374. 12:51you've asked it. These autonomous
375. 12:53agents, you know, the famous example is
376. 12:54you ask them to produce as many paper
377. 12:56clips as possible, and they start
378. 12:57liquidating the universe to make paper
379. 12:59clips, and you know, it's a it's a
380. 13:00sorcerer's apprentice kind of issue. The
381. 13:03third one I would say is
382. 13:06actually all of the remainder of them I
383. 13:08would say are guaranteed, and
384. 13:10um
385. 13:11the third of them is
386. 13:14the derangement of human intellect. That
387. 13:18we are already living in a world where
388. 13:20it's very difficult to know
389. 13:22what the facts even mean.
390. 13:24Right? We the facts are so filtered, and
391. 13:26we are so persuaded by algorithms that
392. 13:29it's, you know, our our ability to be
393. 13:31confident even in the basic facts, even
394. 13:33within our own discipline sometimes, is
395. 13:35at an all-time low.
396. 13:37And it's getting worse. And that problem
397. 13:39takes a giant leap forward at the point
398. 13:42that you have the ability to generate
399. 13:46undetectable deep fakes.
400. 13:48Right?
401. 13:49>> So, this idea of hyper-novelty is
402. 13:52something I think we all feel, but
403. 13:54haven't been able to name. Our brains
404. 13:56evolved over thousands of years to live
405. 13:58in a world that is predictable. You grew
406. 14:01up knowing what your parents did, and
407. 14:03you probably assumed you'd do something
408. 14:05similar.
409. 14:06But now, the world we are living in as
410. 14:08adults looks nothing like the world we
411. 14:10grew up in, and that biological mismatch
412. 14:13is making us sick. We are anxious, we
413. 14:16are depressed, and we are disconnected.
414. 14:18Now, take that and amplify it with AI
415. 14:20moving at the speed of light. Brett
416. 14:22brings up this idea of useless eaters,
417. 14:25which is a terrifying phrase, but it's
418. 14:27the logical endpoint of this. If you
419. 14:29don't have a job, you don't have
420. 14:31economic leverage, and if you don't have
421. 14:33economic leverage, you don't have
422. 14:35political power.
423. 14:36When you combine that with the fact that
424. 14:38AI can generate undetectable deep fakes,
425. 14:40we lose our grip on reality. If we can't
426. 14:43agree on what the facts are, we can't
427. 14:45organize to fight back. It's the perfect
428. 14:47storm for total control. Thanks for
429. 14:50watching, and please don't forget to
430. 14:51like and subscribe to keep up with where
431. 14:53all of this is heading.