Elon Musk on AI: humans will no longer be in control in ten years | The Economist — Transcript

Full transcript

1. 0:00I think AI may exceed the sum of human
2. 0:03intelligence
3. 0:05in about in around five years.
4. 0:07>> In five years,
5. 0:08>> roughly five years is my guess. Um,
6. 0:12so there really won't be anything that
7. 0:15AI
8. 0:16can't do better than humans apart from
9. 0:19being human perhaps.
10. 0:20>> In a more prosaic level, what will life
11. 0:22be like?
12. 0:25The most likely outcome
13. 0:28is an age of amazing abundance uh where
14. 0:34anyone can have anything they can think
15. 0:36of.
16. 0:37This is this is this may sound
17. 0:39preposterous but well here we are in
18. 0:412026.
19. 0:43Let's see where we stand in 2036. Now of
20. 0:45course there are things that could
21. 0:47derail this like a a massive global
22. 0:50thermonuclear war or something like
23. 0:51that. Um, but assuming that there's not
24. 0:54World War II,
25. 0:56uh, I think I think we're headed for an
26. 0:59age of amazing abundance. So, this is, I
27. 1:01guess, a message of of optimism
28. 1:05and, uh, excitement about the future.
29. 1:07>> You're sure that humans will no longer
30. 1:09be in control in 10 years time?
31. 1:12>> I think it is unlikely. Um, in the same
32. 1:14way that if you had a vastly more like,
33. 1:20you know,
34. 1:22the
35. 1:24like like if if if the difference in
36. 1:26intelligence between AI and humans is
37. 1:30vastly greater than the difference in
38. 1:31intelligence between AI and chimpanzees,
39. 1:34it's hard to imagine that the
40. 1:36chimpanzees would be in charge.
41. 1:38>> I'm really sorry. We are essentially um
42. 1:42an evolved form of chimpanzee. So it
43. 1:45wasn't that long ago that we're swinging
44. 1:47through the trees eating bananas um in
45. 1:51the jungle and we still eat bananas.
46. 1:53>> You said there was a 10 to 20% chance of
47. 1:55killer robots wiping out humanity.
48. 1:57>> So you sounded worried then. Now you
49. 2:00seem very relaxed about it. Have you
50. 2:02changed your view?
51. 2:03>> I still think there's risk associated
52. 2:06with AI and robots. It's not zero. Um,
53. 2:09I' I guess I've come to
54. 2:12my my sort of philosophical conclusion
55. 2:15is to look on the bright side. Um, I
56. 2:19like I I can't see any way to really
57. 2:21stop this incredible momentum of AI and
58. 2:25robots. Um, and and at times I sort of
59. 2:29think, well, perhaps, you know, even if
60. 2:31there was a stop button, we probably
61. 2:33shouldn't press it. uh because
62. 2:38because the most likely outcome is
63. 2:41incredible abundance for for all.
64. 2:45>> That's what I thought actually. It
65. 2:46struck me that now you seem to have
66. 2:48decided it's inevitable.
67. 2:51Let's hope for the best and go for the
68. 2:53ride.
69. 2:54>> Yeah. I mean, yes, pretty much. Let's uh
70. 2:59let's enjoy the ride is is is I think my
71. 3:02philosophy at this point because I mean
72. 3:03honestly this there's it's it's there
73. 3:05there seems to be just an inexurable
74. 3:08uh progress here in in AI and and in
75. 3:12robots
76. 3:13>> that um
77. 3:15>> but you even if I wanted to stop it I
78. 3:18don't I couldn't um
79. 3:20>> do you think there's still a 10 to 20%
80. 3:21chance that killer robots mean the end
81. 3:24of humanity?
82. 3:27That's a high P doom as they say.
83. 3:30>> Well,
84. 3:32in the long run, we're all dead of
85. 3:33course. Um, you know, um, so I mean, so
86. 3:36I mean like according to the standard
87. 3:40model of physics, the end of the
88. 3:42universe is um a long boring sort of
89. 3:47thin gray soup of uh gradual cooling. um
90. 3:54meaning the sort of the heat death of
91. 3:56the universe it's called. So now if if
92. 3:58the heat death of the universe is in
93. 4:00fact the outcome for our reality, then
94. 4:03it really is all about the journey
95. 4:05because the destination is terrible.
96. 4:07>> But 10 years [laughter] is not very
97. 4:08Yeah, I get that. But that could be a
98. 4:10long way and you're basically very long
99. 4:12way.
100. 4:12>> If it's 10 to 20% in 10 years, you know,
101. 4:15would you go into one of your rockets if
102. 4:16you thought there was a 10 to 20% chance
103. 4:18of it blowing up and killing you?
104. 4:21>> Yes. But [laughter]
105. 4:23but let's say there's you can't do
106. 4:25anything about it. Mhm.
107. 4:27>> that I guess the only thing we can try
108. 4:29to do is minimize the probability of the
109. 4:32bad outcome. Um but trying to stop the
110. 4:36intelligence in robots.
111. 4:38Um
112. 4:40I mean for the longest time I
113. 4:44declined to participate in AI or was you
114. 4:47know you know created open AI as a
115. 4:51essentially a counterweight to to Google
116. 4:52because at the time they were um they
117. 4:55had more or less a monopoly on AI in
118. 4:57creating open AI and then having
119. 4:59anthropics spin out of that and now
120. 5:01currently anthropics leader in AI. Um,
121. 5:06I mean the these actions have actually
122. 5:08resulted in
123. 5:10knock-on effects that accelerated AI,
124. 5:13which wasn't really my intention.
125. 5:16So, it just seems like all roads lead to
126. 5:19acceleration of AI.
127. 5:22So, then I'm like, okay, well, you can
128. 5:24just sort of be sad about it or
129. 5:27join the club, I suppose.
130. 5:29>> Well, you've certainly joined the club.
131. 5:30>> Yeah, [laughter]
132. 5:31>> you definitely joined the club. If you
133. 5:33can't beat him, join them. You know,
134. 5:35>> a number of the other leaders of of
135. 5:38model makers have been pushing for
136. 5:40different ways to try and minimize the
137. 5:42risks. Um, you know, just last week,
138. 5:44Demises came out with a proposal for a
139. 5:47private public regulator. Do you think
140. 5:50you guys should be working more actively
141. 5:53>> to get something in place to control the
142. 5:56risks or do you think that's impossible
143. 5:57too? Yeah, my recommendation to Dennis
144. 6:00because we I talked to him for a few
145. 6:02hours before he published his his piece
146. 6:04was that um the most immediate thing
147. 6:07that we could do is to have the leading
148. 6:08AI companies um at least just meet uh or
149. 6:13have a some sort of call once every few
150. 6:16weeks um and just
151. 6:19uh discuss
152. 6:22any safety and security issues.
153. 6:23>> But why why haven't you done that yet? I
154. 6:25mean you don't all like each other very
155. 6:27much. Yeah. But you're not all you're
156. 6:28not all I mean you're also all competing
157. 6:31right? You all want to be first.
158. 6:32>> Yes. But that's that's where I think
159. 6:35>> um the competitors can keep it keep each
160. 6:38other honest. So meaning meaning like if
161. 6:41um like I think it's quite difficult for
162. 6:43someone in the government who doesn't
163. 6:44have a deep technical understanding and
164. 6:46isn't driving the frontier of of AI to
165. 6:50know whether something should be
166. 6:51released or not. Um however the
167. 6:54competitors I think will can you know if
168. 6:58given sort of a week or two to review a
169. 7:00new model um can highlight issues and uh
170. 7:04and have an incentive to keep all all
171. 7:06the competitors have an incentive to
172. 7:07keep the others honest.
173. 7:09>> So you'd have a system where the model
174. 7:10makers would check each other's work. If
175. 7:12there was something that worried them,
176. 7:13you'd tell the government. Yeah. The
177. 7:15government would then think, okay,
178. 7:16>> if there's something that was worrisome
179. 7:17and that company was not doing anything
180. 7:19to address that risk, then that would be
181. 7:22the the moment for government to step in
182. 7:24and take action, which is literally what
183. 7:26just what
184. 7:28>> does all this have to happen?
185. 7:31>> Well, I think it would be good to do
186. 7:32that I think
187. 7:35immediately really.
188. 7:37>> Yeah, that's what people keep telling
189. 7:38me. we we have to get this some kind of
190. 7:39system in the next 6 months because of
191. 7:42the pace at which the models are
192. 7:43improving.
193. 7:45>> Yeah, 6 months is a long time. Um,
194. 7:49>> so you think shorter than that?
195. 7:52>> Well, I mean there there's so many AI
196. 7:55breakthrough announcements, sometimes
197. 7:57multiple per day. I mean, how many were
198. 8:00there last week?
199. 8:01>> A lot.
200. 8:01>> I lost track, you know.
201. 8:02>> Yeah, a lot.
202. 8:04Um,
203. 8:06>> so, so you you basically, you know,
204. 8:08whether it's Demis' proposal or others,
205. 8:10you need to get together and get this
206. 8:11set up kind of now. It sounds like
207. 8:13you're telling me.
208. 8:14>> Yeah. I I I would recommend that we at
209. 8:17least have some sort of informal g um
210. 8:21you know weekly or bi-weekly um call and
211. 8:26and and that like I said just that that
212. 8:28there's that that
213. 8:30um there's maybe a week or two weeks of
214. 8:33of early access by competitors and this
215. 8:37is why I think the incentives work is
216. 8:38because if if you know competing
217. 8:40companies I think are likely to
218. 8:43understand if something is a security
219. 8:44risk
220. 8:45and they're not going to be shy about
221. 8:47highlighting that that their
222. 8:49competitor's models should be delayed in
223. 8:51release.
224. 8:52>> You guys don't much like each other and
225. 8:54you don't much trust each other. You're
226. 8:55insulting each other on on social media
227. 8:57constantly. So, do you do you think you
228. 9:00can do that with people that you dislike
229. 9:02that that where there seems to be such
230. 9:03kind of personal vitrial between some of
231. 9:05these leaders?
232. 9:07>> Um, yes. Well, I'm I'm not a fan of Sam
233. 9:11album because if you started a nonprofit
234. 9:13that was meant to be an open-source AI
235. 9:16company and um you know owned by the
236. 9:21world uh and it somehow got turned into
237. 9:23an $800 billion for-profit company with
238. 9:26closed source. I think you'd be like
239. 9:28well wait a second that's uh the exact
240. 9:30opposite of what I donated the money
241. 9:32for. That that's my issue. I think it's
242. 9:34a legitimate one. Um, and then Daario I
243. 9:38think also is not a fan of Sam Alman
244. 9:40because he felt uh, in fact the reason
245. 9:42the anthropic team left OpenAI is
246. 9:44because they they didn't trust Sam Alman
247. 9:46otherwise anthropic wouldn't exist. They
248. 9:49would still be at OpenAI. Um,
249. 9:52I I think Dario is a very principled
250. 9:55person. He cares about things a lot. uh
251. 9:58cares about the future of the world. And
252. 10:00I I think everyone I've met at Anthropic
253. 10:02so far has been I'm like no one set off
254. 10:06my evil detector. You know, they always
255. 10:07seem seem like they're well-intentioned.
256. 10:10Um and uh the road to hell is I think
257. 10:13mostly paved with bad intentions. There
258. 10:16are a few well-intentioned paving stones
259. 10:17in there, so we we don't want to be
260. 10:19complacent. Um but uh you at the end of
261. 10:23the day, if we have to talk, we'll talk.
262. 10:24I mean, set aside our personal
263. 10:26differences for the good of the world
264. 10:28type of thing.