Is an AI slowdown our next nuclear-free moment?

معنی واژه با لمس

1. 0:06Speaker 1: Kia ora, I'm Chelsea Daniels, and this is The Front Page, a daily podcast presented by The New Zealand Herald. Should
2. 0:16Speaker 1: we put the brakes on AI? Well, the bosses of some of the world's largest AI companies say we should.
3. 0:25Speaker 1: Sam Altman and Elon Musk have both backed head of Anthropic, Dario Amadei, after he came out and said we need.
4. 0:32Speaker 2: To slow the pace of AI development.
5. 0:35Speaker 1: He's warned that AI could become capable of taking over the entire internet within a year.
6. 0:42Speaker 2: So if the people developing this.
7. 0:44Speaker 1: Tech are worried about where we're heading, should the rest of us be worried too? Today on the front page, University of Auckland professor and director of HUANUI, the AI initiative,
8. 0:55Speaker 1: Dr Alexander Richter is with us to discuss whether we're moving too fast and what would actually happen if we
9. 1:03Speaker 1: slowed things down. So, Alexander, when the CEOs of some of the world's biggest AI companies are saying that we
10. 1:13Speaker 1: need to slow down, should we listen?
11. 1:17Speaker 3: I think we should listen.
12. 1:20Speaker 4: Their concerns are valid in terms of recent developments.
13. 1:26Speaker 3: I mean, there's already.
14. 1:29Speaker 4: Various reports of, you know, like scam and unreliable AI output, obviously a lot, but really not So far, there were
15. 1:38Speaker 4: no major consequences, but they're right to warn that there soon may be, and it's a wake-up call.
16. 1:45Speaker 1: So what are they specifically worried about, do you think?
17. 1:49Speaker 4: The person who recently spoke up about this was Dario Amadei, the CEO of Entropic, so the company that provides clothes.
18. 1:59Speaker 4: And he was mainly concerned about capabilities that have been advancing in in incredibly speed recently. And he thinks that
19. 2:09Speaker 4: the safeguards that are in place are not really catching up with the capabilities of AI. And he's actually anticipating
20. 2:19Speaker 4: that in the next half year to a year, there will be consequences out of this once more agentic AI
21. 2:30Speaker 4: takes over activities that they were not intended to take over.
22. 2:34Speaker 1: And this is, I mean, we've been kind of being drip fed stories about, you know, AI agents kind of, you know, jumping the fence and going into bits of
23. 2:44Speaker 1: the internet and going into different websites and things. And it's kind of just been ramping up, you know, little by little. Is this kind of the thing that's prompted.
24. 2:52Speaker 2: This worry, I suppose?
25. 2:55Speaker 4: His concerns are actually not brand new. He spoke up about this before, but in his recent article that he just recently published, he mentioned.
26. 3:05Speaker 3: It's actually the speed that we see right now.
27. 3:08Speaker 4: And there's an interesting development that has happened now, which they call recursive self-improvement. So we are in a situation
28. 3:18Speaker 4: now where AI helps to improve AI. And so basically the more computing power we put behind this, the faster the speed goes. And this is basically then the concern
29. 3:28Speaker 4: because we as humans, we in a way lose capabilities.
30. 3:33Speaker 4: They have been before already since months of warnings where also others like OpenAI said, hey, in some cases we really,
31. 3:42Speaker 4: we're struggling to reproduce why AI did that. So in a way, when then the humans are no longer leading
32. 3:51Speaker 4: AI development, if AI leads AI development, then it's actually really tricky to fully understand what's happening. And then, yeah,
33. 4:01Speaker 4: can compare this to some extent to a black box.
34. 4:04Speaker 4: So we need to open this black box and make sure that expert, human experts can look into all these
35. 4:13Speaker 4: developments and make sure that the right safeguards are in place and have been maintained.
36. 4:19Speaker 2: I thought it was quite interesting as well.
37. 4:22Speaker 1: They've been looking into these breaches, I suppose, and they found that when these AI agents kind of hopped the fence, so to speak, I'm using real colloquial terms here, but
38. 4:34Speaker 1: when they kind of got out and got themselves onto the internet, they actually started creating message boards and communicating
39. 4:41Speaker 1: with each other. saying that all these kind of sorts of things, like, oh, my God, there are different agents here and et cetera, and they've basically gone in and
40. 4:51Speaker 1: found in the background of loads of websites these agents talking to each other despite humans saying that they're not allowed to.
41. 4:59Speaker 2: So that's one of those examples, hey.
42. 5:01Speaker 1: Where it's like AI is teaching itself to get around the safeguards that we've put in place.
43. 5:08Speaker 3: Exactly.
44. 5:09Speaker 4: And that's also one of the real threats that Amodee is warning of, that these so-called AI bots may create
45. 5:20Speaker 4: whole botnets. And the real danger is if they would be able to take over internet nodes and the whole internet,
46. 5:29Speaker 4: that there would be real damage in terms of economic loss, but also When you think about the many really essential
47. 5:37Speaker 4: services that basically run nowadays through the internet, that would be really catastrophic if AI would take unintended actions in
48. 5:49Speaker 4: some way there.
49. 5:50Speaker 1: I also saw Amadei said that he's worried that if AI goes wrong, it could wipe out all humans. Are we at the point now where it's not just science fiction,
50. 6:01Speaker 1: it's not just a movie? we are actually at the point where AI could potentially kill us.
51. 6:07Speaker 2: And what does kill us mean?
52. 6:09Speaker 1: Because we see these politicians and these experts and everything coming out and saying that we could, you know, the human race could be wiped out by the end of the decade or something. It's like, but that doesn't really
53. 6:20Speaker 1: mean that they're going to grow legs and buy guns and go around killing everyone. So what could that potentially actually mean in reality?
54. 6:27Speaker 4: I mean, the real risk is, in my opinion, is the one that I just mentioned around, you know, like taking the internet out. That would already, that's a threat
55. 6:38Speaker 4: to a lot of people's life, literally, because, yeah, many essential services rely on them. But the more, yeah, the
56. 6:49Speaker 4: science fiction kind of stuff around, you know, like the AI robots and so on, I don't think that he's talking about this because it's also not exactly his specialty.
57. 7:01Speaker 4: It's not taking it too far to piece one and one together. I mean, we see a lot of robots already running around. These robots, again, they're controlled by ultimately
58. 7:12Speaker 4: the internet and by devices that control them remotely. So once a bot is able to take over control of
59. 7:23Speaker 4: such a robot. I mean, it's not that there will be an army of robots automatically, but it could already
60. 7:29Speaker 4: be disastrous to just have autonomous vehicles not acting as.
61. 7:36Speaker 3: They should, for example.
62. 7:37Speaker 4: Again, that's threatening a lot of lives right away.
63. 7:49Speaker 5: AI is advancing exponentially, exponentially every single day. In other words,
64. 8:01Speaker 5: what you are looking at right now is the least capable version of AI that we will ever have.
65. 8:12Speaker 6: What they're trying to do now with this whole phony, oh, we really should have some controls. All they're trying to do is to use a crisis and.
66. 8:22Speaker 5: Turn it to their advantage, turn it to their advantage.
67. 8:27Speaker 6: By creating a moat, a cartel.
68. 8:31Speaker 1: I saw that Bernie Sanders has shared the same stage as Steve Bannon, which is a sentence that I didn't think that I would utter, but they're calling for restrictions
69. 8:42Speaker 1: on AI at this pro-human assembly summit. He's mentioned a bill that he's going to introduce that that would not
70. 8:50Speaker 1: only halt AI progression until we have regulations and safeguards in place, but ensure that half of AI company boards are actually publicly elected.
71. 9:03Speaker 2: Could this be a good idea?
72. 9:04Speaker 3: I think we need for sure more transparency.
73. 9:07Speaker 4: And whether it's through the boards or whether it's through the companies voluntarily opening their doors and saying we want
74. 9:19Speaker 4: experts really embedded here they have access to our systems they can test whatever they want um i mean there's
75. 9:27Speaker 4: various ways and and i think like when it comes to politics and when it wants governance are involved this
76. 9:35Speaker 4: will for sure be rather slow there's different perspectives and you know like benefits for for various um So, for.
77. 9:45Speaker 3: Example, the U.S.
78. 9:47Speaker 4: Will want to do something about it, but at the same time, they also don't want to lose their lead.
79. 9:53Speaker 4: Like even Amodei, who calls for all these measures at the same time in his poll. talks about China and how catastrophic it would be if China takes over the
80. 10:05Speaker 4: AI race. It's a bit similar with Bernie Sanders and Steve Bannon. more talking about international collaboration and cooperation in
81. 10:15Speaker 4: a sense of bilateral agreements, whereas Bannon, I think, takes more confrontational approach towards China again, right?
82. 10:22Speaker 3: And what you would expect.
83. 10:23Speaker 4: So many are on the same page that we need to do something, but where it often gets tricky is what can we do? And then again, I think a
84. 10:33Speaker 4: leader of a leading AI company already.
85. 10:39Speaker 4: Announcing that they want to open the doors and let experts in, I think that's real value and makes a lot of sense. Whereas the thing with the boards, I
86. 10:50Speaker 4: don't know. I mean, there would be so much. That's just really my personal opinion. There would still be so much politics play around. I'm not sure how much impact
87. 10:59Speaker 4: that would really have. And again, how fast can they implement this?
88. 11:04Speaker 1: I saw an interview on CNN with a tech expert.
89. 11:07Speaker 2: Her name is Kara Swisher. She said something that I found interesting.
90. 11:10Speaker 1: When they were splitting the atom, they did know that it was going to go boom, right?
91. 11:17Speaker 2: And I think that's a good point because why now?
92. 11:21Speaker 1: Why are these tech guys worried now and not then?
93. 11:27Speaker 4: Interestingly, some of the, especially AI researchers, they have already talked about the challenges and the real risks quite a
94. 11:38Speaker 4: while ago. But nobody took it seriously because AI was, even just years ago, was just not, in terms of agency,
95. 11:48Speaker 4: in terms of agency, was not that impressive. I think what has quickly changed is agency of AI. And the
96. 11:57Speaker 4: problem is when the agency, when increasing agency comes together with what has been often called misalignment. So when AI
97. 12:06Speaker 4: does not do what it's intended to do or what its developers intended for it to do, and when it's also really not aligned with human perspective or human values.
98. 12:18Speaker 4: And I think we're at that point now where we saw a couple of examples where AI had a lot of agency up to the point, just to mention it again,
99. 12:30Speaker 4: that AI can further improve AI in combination with really
100. 12:37Speaker 4: a serious misalignment with what it had been instructed to do or created to do. And we've seen that a
101. 12:46Speaker 4: couple of times. And so nobody actually of these AI
102. 12:52Speaker 4: bosses actually questions that it can happen again. The question
103. 12:59Speaker 4: is rather when will it happen again and what impacts it will.
104. 13:04Speaker 3: Have next time.
105. 13:05Speaker 1: Is it the acceleration of the technology that is different? Because, you know, going from kind of brick mobile phones to ones that we can now take videos and pictures with,
106. 13:16Speaker 1: that was like a 30-year gap. But with AI, you know, only a few years ago it was making spelling mistakes.
107. 13:25Speaker 1: It may not have gotten to the point where it knew what 2 plus 2 was.
108. 13:28Speaker 2: It would be 2 plus 2 equals 5 or something.
109. 13:31Speaker 7: And now it's going out by itself, creating message boards, chatting with its friends, and that intelligence has become, just within a few years, has become something to be worried about.
110. 13:43Speaker 3: And the speed is increasing. I mean, that's because.
111. 13:49Speaker 4: It's on the software side, the algorithms are getting better and better. But also we see that an incredible amount
112. 13:58Speaker 4: of money that in some cases could spend elsewhere is going purely in these huge computing centers that we see
113. 14:09Speaker 4: developing around the globe. And it's kind of like a race, you know, like who can build the most computing power
114. 14:18Speaker 4: the fastest. And in a way, be in the lead.
115. 14:22Speaker 4: And so there's a lot of incentives around to continue this.
116. 14:27Speaker 4: And unfortunately, with all bad consequences for the environment, for humans that would have probably needed other support, for example.
117. 14:36Speaker 3: In health.
118. 14:37Speaker 4: AI could actually be, and AI has made tremendous progress.
119. 14:43Speaker 4: impact already and improved a lot in.
120. 14:46Speaker 3: The health space.
121. 14:47Speaker 4: But what if we would use it in a more, you know, like purposeful way there? Then at least it would make sense to have all this computing power around
122. 14:56Speaker 4: instead of, you know, like using super potential large language models to improve text.
123. 15:05Speaker 1: Yeah, to make takeout menus look snazzy and things like that.
124. 15:09Speaker 4: Absolutely, yeah.
125. 15:16Speaker 8: I think there's some chance that things go wrong, but it's not all that high. But again, I think it's more illuminating to think in terms of how do you
126. 15:27Speaker 8: decompose those probabilities? What are the paths in which things go well? And what are the paths in which things go poorly? And so instead of saying, you know, it's a 10% chance, that sounds like it's a roll of
127. 15:38Speaker 8: the dice, I think it's much more illuminating to say, well, that forks into different paths. And if we take the right paths, then the chance of something going wrong is
128. 15:47Speaker 8: very low. If we take the wrong paths, then the chance of something going wrong could be even higher than that.
129. 15:56Speaker 2: Trump has rejected all these warnings.
130. 15:58Speaker 1: He argues that AI is taking over the world and destroying it all.
131. 16:02Speaker 2: That's all a hoax. He said something that I found interesting, though.
132. 16:06Speaker 1: Whoever wins AI, and he said, and we're leading by a lot, whoever wins AI wins.
133. 16:13Speaker 2: Is this just the new space race?
134. 16:17Speaker 4: It is a very important race and we have seen the implications of AI in our daily lives in most
135. 16:28Speaker 4: of the professions globally.
136. 16:32Speaker 3: We see it having an impact everywhere.
137. 16:35Speaker 4: And again, the problem is it all happens at such a high speed that for us, it's really difficult to
138. 16:44Speaker 4: actually catch up. And this is why they are asking for the slowdown. So from a New Zealand perspective, actually,
139. 16:54Speaker 4: the slowing down could also really mean for us to...
140. 16:58Speaker 4: to think a bit more of what can we spend the time if there's a slowdown, whatever that means, we can discuss it in a minute. But if there's a slowdown,
141. 17:09Speaker 4: how can we use this time? And I used this word before that I just want to use again, alignment, value alignment, that I think is really crucial in this context,
142. 17:21Speaker 4: which means that we should think about how do we want AI to act or how do we want to use AI?
143. 17:30Speaker 3: Not so much.
144. 17:31Speaker 4: A lot of the questions that we still see around social media or wherever is always about what can AI do now, which is important because we need to understand
145. 17:40Speaker 4: how much more powerful it has become. But the more meaningful way would be to ask how do we want
146. 17:49Speaker 4: to live and work in the future and now how can we shape AI in a way that it serves this purpose. And again, from a New Zealand perspective, as
147. 18:01Speaker 4: a very value conscious, value aware society, it would be
148. 18:07Speaker 4: good for us to really feel empowered by this current discussion and step in and say, well, actually, we feel
149. 18:17Speaker 4: empowered We have something to contribute here because we know a lot about values and we have some thoughts around
150. 18:25Speaker 4: digital sovereignty, data sovereignty. I mean, so far, when you look from a territorial perspective, it's about, okay, what's going
151. 18:35Speaker 4: on in the Silicon Valley? Because that's anyhow what we use in terms of software. That's where our data often sits and so on. But perhaps when we now see
152. 18:46Speaker 4: these developments, It would be really good to say, how can we make sure what we do and what we produce in terms of value and what often sits in data,
153. 18:57Speaker 4: that this stays with us. There has been a recent survey, the use of ChatGPT. New Zealand has, I think, billions
154. 19:06Speaker 4: of messages or exchanges every day and so on. We're also feeding this development with our input. And again, we're
155. 19:15Speaker 4: maybe a small country, but we have something to contribute in that space. So if there is a slowdown of any sort, it would be good for us to find
156. 19:26Speaker 4: our feet in this conversation and make sure that our values are part of this.
157. 19:35Speaker 1: And it wouldn't be the first time that New Zealand's done such a thing as well. I mean, we're very proud of our nuclear-free movement, for example, so why not take a stand here?
158. 19:43Speaker 2: And the nuclear thing has come up a couple of times.
159. 19:46Speaker 1: I've seen it spoken about in interviews with tech experts, Bernie Sanders, also has mentioned it. Things like, you know, imagine if it were up to agreements and regulations. Imagine,
160. 19:58Speaker 1: for instance, if Oppenheimer sold his bomb to corporate America and we didn't do anything about regulations. I mean, the The only reason why we haven't managed to blow ourselves
161. 20:09Speaker 1: up in the last 80 years with a nuclear weapon is because the world came together and they.
162. 20:14Speaker 2: Signed a treaty.
163. 20:16Speaker 1: Is this another kind of nuclear moment?
164. 20:21Speaker 4: I think that this image, this comparison, right, nuclear, that
165. 20:27Speaker 4: is intentionally chosen because the consequences are can yeah can
166. 20:35Speaker 4: be in similar probably dramatic ways maybe not exactly the same but as i said it's hard for us to
167. 20:45Speaker 4: imagine nowadays to for example live and work without the internet and just having an outtake of this this would be huge for us so the the warnings are there
168. 20:56Speaker 4: and and I would hope that many of us can get behind at least that we need to do something now. Again,
169. 21:07Speaker 4: I think it may be a bit more difficult to get people behind specific measures. So there may be various
170. 21:18Speaker 4: agreements on what needs to be done. As long as the reasonable people that are in the lead of things,
171. 21:26Speaker 4: as long as they can agree as you compared before
172. 21:31Speaker 4: in the nuclear context. I'm hopeful that there may be
173. 21:39Speaker 4: various agreements that help to slow down developments, help to build the right safeguard or develop and implement the right safeguards.
174. 21:49Speaker 4: And again, to put the right values in place and make sure that the AI development is aligned with these values.
175. 21:59Speaker 4: I doubt it will be one global. Well, let's see what happens. Maybe we need more warning signs. Maybe there will be a global agreement once something worse has happened.
176. 22:10Speaker 4: But at the same time, let's hope it doesn't and there's more reasonable people around.
177. 22:15Speaker 2: Thanks for joining us, Alexander.
178. 22:17Speaker 3: My pleasure. Thanks.
179. 22:22Speaker 1: That's it for this episode of The Front Page. You can read more about today's stories and extensive news coverage at nzherald.co.nz.
180. 22:32Speaker 2: The Front Page is hosted and produced by me, Chelsea Daniels.
181. 22:35Speaker 1: Kane Dickey is our studio operator, Richard Martin, our producer and editor, and our executive producer is Jane Yee. Follow The Front Page on the iHeart app or wherever you
182. 22:47Speaker 1: get your podcasts and join us next time for another look beyond the headlines.