WEBVTT - Will AI Radically Change the World by 2027?... from Risky Business 0:00:15.356 --> 0:00:15.796 Pushkin. 0:00:20.236 --> 0:00:21.156 Hey, it's Jacob. 0:00:21.236 --> 0:00:25.316 There's another Pushkin show called Risky Business. It's a show 0:00:25.356 --> 0:00:29.276 about decision making and strategic thinking. It's hosted by Nate 0:00:29.356 --> 0:00:32.196 Silver and Riya Kanakova, and they just had a really 0:00:32.276 --> 0:00:36.356 interesting episode about the future of AI. So we're going 0:00:36.436 --> 0:00:39.516 to play that episode for you right now. We'll be 0:00:39.596 --> 0:00:42.156 back later this week with the regular episode of What's 0:00:42.156 --> 0:00:48.836 Your Problem. 0:00:48.956 --> 0:00:52.316 Welcome back to Risky Business, a show about making better decisions. 0:00:52.476 --> 0:00:54.156 I'm Maria Kanakova. 0:00:53.716 --> 0:00:54.516 And I'm Nate Silver. 0:00:54.836 --> 0:00:56.556 Hey. Today on the show is going to be a 0:00:56.596 --> 0:00:57.836 little bit doomtastic. 0:00:58.556 --> 0:01:00.996 Yeah, I mean, I don't know if it's like worse 0:01:01.156 --> 0:01:05.156 and thinking about like the global economy going into a 0:01:05.196 --> 0:01:09.676 recession because of dumb fuck tariff policies about how we're 0:01:09.676 --> 0:01:11.436 all going to die in seven years and staid no, 0:01:11.516 --> 0:01:14.716 I'm just kidding. This is a very very intelligent and 0:01:14.756 --> 0:01:18.196 well written and thoughtful report called AI twenty twenty seven 0:01:18.276 --> 0:01:20.556 that we're to spend the whole show on I because 0:01:20.556 --> 0:01:23.196 think it's such an interesting to talk about, but that 0:01:23.316 --> 0:01:29.116 you know, includes some dystopian possibilities. I would say it does. 0:01:28.956 --> 0:01:32.036 Indeed, So let's get into it and hope that you 0:01:32.076 --> 0:01:34.196 guys are all still here to listen to us in 0:01:34.276 --> 0:01:38.316 seven years. 0:01:38.476 --> 0:01:43.516 The contrast is interesting between like all the chaos we're 0:01:43.556 --> 0:01:47.356 seeing with tariff policy in terms of starting a trade 0:01:47.356 --> 0:01:51.316 war with China and then other types of chaos. It's 0:01:51.316 --> 0:01:53.476 interesting to kind of look at this. I mean, I 0:01:53.476 --> 0:01:58.396 wouldn't call it a more optimistic future exactly, like but 0:01:58.476 --> 0:02:00.476 like on a different trajectory of like a future that's 0:02:00.476 --> 0:02:04.556 going to change very fast, according to these authors, with 0:02:04.716 --> 0:02:11.036 profound implications for you know, everything, the humans species. These 0:02:11.196 --> 0:02:14.036 researchers and authors are saying that everything is going to 0:02:14.156 --> 0:02:16.076 change profoundly. 0:02:15.756 --> 0:02:16.476 And. 0:02:18.156 --> 0:02:22.116 Even though there is some hedging hair, this is kind 0:02:22.156 --> 0:02:26.076 of their base case scenario. And like you know, base 0:02:26.156 --> 0:02:28.836 case number one and base case number two differ. There's 0:02:28.876 --> 0:02:30.476 like kind of a choose your own adventure at some 0:02:30.556 --> 0:02:33.396 point in this report, but they're both very different than 0:02:33.396 --> 0:02:36.236 the status quo, right, and the notionately hear from that, 0:02:36.396 --> 0:02:42.196 like everything becomes different If ais become substantially more intelligent 0:02:42.236 --> 0:02:44.196 than human beings. People can debate and we will debate 0:02:44.236 --> 0:02:46.156 on this program what that means. But yeah, do you 0:02:46.156 --> 0:02:48.036 want to do you want to contextualize this for me? 0:02:48.116 --> 0:02:49.476 Do you want to tell people too. Yeah, there's are 0:02:49.516 --> 0:02:50.396 of this report. 0:02:50.396 --> 0:02:56.196 Absolutely absolutely. So the report is authored officially by five people, 0:02:56.396 --> 0:02:59.516 but I think unofficially there's also a sixth. We've got 0:03:00.476 --> 0:03:05.556 Eli Leiflund, who's a super forecaster and he was ranked 0:03:05.636 --> 0:03:09.916 first on Rand's forecasting initiative, so he is someone who 0:03:10.036 --> 0:03:13.116 is very good at kind of looking at the future 0:03:13.156 --> 0:03:16.556 trying to predict what's going to happen. You have Jonas Walmer, 0:03:16.636 --> 0:03:21.436 who's a VC at Macroscopic Ventures. Thomas Larson was the 0:03:21.516 --> 0:03:26.476 former executive director of the Center for AI Policy, so 0:03:26.596 --> 0:03:30.396 that is a center that advises both sides of the 0:03:30.436 --> 0:03:34.756 aisle on how AI is going to go. And Romeo Dean, 0:03:34.876 --> 0:03:39.676 who is part of Harvard's AI Safety Student Team, so 0:03:39.796 --> 0:03:42.876 someone who is still a student, still learning, but kind 0:03:42.876 --> 0:03:48.196 of the next generation of people looking at AI. And 0:03:48.396 --> 0:03:53.276 finally we have Daniel Cochtaylo, who basically had written a 0:03:53.356 --> 0:03:57.316 report back in twenty twenty one. He was a AI 0:03:57.396 --> 0:04:01.596 researcher and he looked at predictions for AI for twenty 0:04:01.636 --> 0:04:04.916 twenty six and it turns out that his predictions were 0:04:04.956 --> 0:04:09.076 pretty spot on, and so Open AI actually hired Daniel 0:04:09.316 --> 0:04:10.476 as a result of this report. 0:04:10.476 --> 0:04:12.516 Obviously he left, yes, and. 0:04:12.436 --> 0:04:14.596 Then and then he left exactly. 0:04:14.236 --> 0:04:17.996 And importantly, So there's also Scott Alexander exactly. 0:04:17.996 --> 0:04:19.916 So I was saying, he's the he's the person who 0:04:20.356 --> 0:04:23.276 kind of is in the background, and he's you guys 0:04:23.356 --> 0:04:26.556 might know him as the author behind astral Codex. 0:04:27.396 --> 0:04:29.036 And I know Scott. He's one of the kind of 0:04:29.076 --> 0:04:36.236 fathers of what you might call rationalism. I think Scott, 0:04:36.236 --> 0:04:38.516 when I interviewed him from a book, was happy enough 0:04:38.516 --> 0:04:40.916 with that term and accused me or co opted me 0:04:40.956 --> 0:04:45.916 into also being a rationalist. These people are somewhat adjacent 0:04:45.956 --> 0:04:48.516 to the effective altruist, but not quite right. They're just 0:04:48.516 --> 0:04:53.596 trying to apply a sort of thoughtful, rigorous, quantitative lens 0:04:53.796 --> 0:04:57.876 to big picture problems, including existential risk, of which most 0:04:57.916 --> 0:05:00.756 people in this community believe that AI is both an 0:05:00.756 --> 0:05:04.596 existential risk and also kind of an existential opportunity, right, 0:05:04.596 --> 0:05:06.676 that it could transform things. You talk to Sam Alp 0:05:06.676 --> 0:05:09.076 and he'll say, we're going to cure cancer and elimina 0:05:09.196 --> 0:05:13.676 poverty and whatever else. Right. And Scott's also an excellent writer. 0:05:13.916 --> 0:05:17.116 And so let me disclose something which is slightly important here. 0:05:17.196 --> 0:05:20.116 So I actually was approached by some of the authors 0:05:20.196 --> 0:05:22.996 of this report a couple of months ago, I guess 0:05:22.996 --> 0:05:26.276 it was in February ish, just to give feedback and 0:05:26.356 --> 0:05:29.796 chat with them. So I'm working off the draft version right, 0:05:29.836 --> 0:05:32.356 which I do not leave. They changed very much, so 0:05:32.396 --> 0:05:34.636 my notes pertained to an earlier draft. I did not 0:05:34.676 --> 0:05:35.836 have time this morning to go back and re. 0:05:35.836 --> 0:05:39.636 Read it, so I was not on the inside loops. 0:05:39.716 --> 0:05:42.396 So I did not get an earlier draft. And I've 0:05:42.436 --> 0:05:46.276 read this draft and basically just to kind of big 0:05:46.316 --> 0:05:51.876 picture sum it up, it outlines two scenarios, right, two 0:05:51.916 --> 0:05:56.276 major scenarios for how AI might change the world as 0:05:56.556 --> 0:06:01.916 soon as twenty thirty. Now important note like that that 0:06:02.196 --> 0:06:04.996 date is could have hedged. It might be sooner, it 0:06:05.116 --> 0:06:08.476 might be later. There's kind of a there's a confidence 0:06:08.516 --> 0:06:13.556 interval there. But the two different scenarios. One basically we 0:06:13.716 --> 0:06:18.156 do twenty thirty, humanity disappears and is taken over by AI. 0:06:18.716 --> 0:06:22.836 The positive report is in twenty thirty basically we get 0:06:23.636 --> 0:06:26.716 AIS that are aligned to our interests, and we get 0:06:26.796 --> 0:06:30.676 kind of this AI utopia where AIS actually help make 0:06:30.756 --> 0:06:33.556 life much better for everyone and make the standard of 0:06:33.556 --> 0:06:36.436 living much higher. But the crucial turning point is before 0:06:36.516 --> 0:06:39.556 twenty thirty. And the crucial kind of question at the 0:06:39.596 --> 0:06:44.716 center of this is will we be able to design 0:06:45.076 --> 0:06:50.116 AIS that are truly aligned to human interests rather than 0:06:50.196 --> 0:06:52.956 just appear to be aligned and kind of lying to 0:06:53.036 --> 0:06:57.436 us while actually following their own agenda. And how we 0:06:57.556 --> 0:07:01.396 handle that is kind of the lynchpin. And it's actually interesting, 0:07:01.516 --> 0:07:03.876 Nate that you started out with China, because a lot 0:07:03.916 --> 0:07:06.156 of the policy choices and a lot of what they 0:07:06.236 --> 0:07:09.916 see as kind of the decision points will affect the 0:07:09.956 --> 0:07:15.076 future of humanity actually hinge on the US China dynamic, 0:07:15.476 --> 0:07:19.756 how they compete with each other, and how that sometimes 0:07:19.876 --> 0:07:24.596 might basically clash against safety concerns because no one wants 0:07:24.596 --> 0:07:27.876 to be left behind. Can we manage that effectively and 0:07:27.956 --> 0:07:31.436 can kind of that transition work in our favor as 0:07:31.436 --> 0:07:33.636 opposed to against us. I think that this is kind 0:07:33.636 --> 0:07:35.356 of one of the big questions here, And so it's 0:07:35.396 --> 0:07:37.556 funny that we're seeing all of this trade war right 0:07:37.596 --> 0:07:40.156 now as the sideport is coming out. 0:07:41.036 --> 0:07:46.396 Yeah. Look, I think this exercise is partly just a 0:07:46.516 --> 0:07:49.676 forecasting exercise, right, I mean, obviously it's just kind of 0:07:49.716 --> 0:07:53.836 like fork at the bottom where we learn to have 0:07:53.916 --> 0:07:56.996 an AI slow down or we kind of are pressing 0:07:57.036 --> 0:07:59.676 fully on the accelerator, right, Like, in some ways, scenarios 0:07:59.676 --> 0:08:05.676 are like not that different, right, either one assumes remarkable 0:08:05.756 --> 0:08:09.276 rates of technological growth that I think even AI I'm 0:08:09.276 --> 0:08:11.956 never quite sure who to call an optimist or a pessimist, right, 0:08:12.036 --> 0:08:15.916 even AI believers you know, might think is a little 0:08:15.956 --> 0:08:17.796 bit aggressive. Right, But what they want to do is 0:08:17.796 --> 0:08:22.476 they want to have like a specific, fleshed out scenario 0:08:22.596 --> 0:08:24.436 for how the world would look like. It's kind of 0:08:24.476 --> 0:08:28.236 like a modal scenario. And like, I think they'd say that, like, 0:08:29.796 --> 0:08:32.156 we're not totally sure about either of these necessarily, right, 0:08:32.156 --> 0:08:34.516 And I don't think they'd be as like pedantic as 0:08:34.516 --> 0:08:36.756 to say, if you do X, Y and Z, then 0:08:37.476 --> 0:08:39.636 we'll save the world and have utopia, and if you don't, 0:08:39.716 --> 0:08:42.516 then we'll all die, Right. I think they'd probably say 0:08:42.596 --> 0:08:45.516 it's unclear and there's kind of like risk either way. 0:08:45.636 --> 0:08:47.636 We wanted to go through the scenario of like fleshing 0:08:47.636 --> 0:08:51.076 out like what the world might look like, right. I 0:08:51.116 --> 0:08:53.436 do think one thing that's important is that whatever decisions 0:08:54.036 --> 0:08:58.156 are made now could get locked in, right that you 0:08:58.436 --> 0:09:03.236 pass certain points in overturn and it becomes very hard 0:09:03.276 --> 0:09:06.196 to decelerate like an arms race. This is you know 0:09:06.236 --> 0:09:09.236 what we found during the Cold War for example. I 0:09:09.236 --> 0:09:10.636 mean one of the big things I look at is like, 0:09:11.236 --> 0:09:15.116 do we force the AI to be transparent in its 0:09:15.156 --> 0:09:18.156 thinking with humans? Right? Like, now there's been a movement 0:09:18.196 --> 0:09:22.116 toward the ai'll actually explicate it's thinking more. I'll ask 0:09:22.116 --> 0:09:25.556 it a query open AI. The Chinese models to this too, right, 0:09:25.556 --> 0:09:27.836 and I'll say, I am thinking about X, Y and Z, 0:09:27.956 --> 0:09:30.116 and I'm looking up PD and Q and now I'm 0:09:30.116 --> 0:09:33.276 reconsidering this. It's actually has this chain of thought process, right, 0:09:33.676 --> 0:09:36.876 which is explicated in English. You know. One concern is 0:09:36.876 --> 0:09:39.236 that what if the AI just kind of communicates to 0:09:39.276 --> 0:09:44.756 another in these implicit vectors that's inferring from all the 0:09:44.796 --> 0:09:48.236 texts it has. It's kind of unintelligible to human beings, right, 0:09:48.276 --> 0:09:50.196 and maybe kind of quote unquote thinking in that way 0:09:50.796 --> 0:09:52.676 in the first place, and then does us the favor 0:09:52.756 --> 0:09:55.356 of like translating back so it it goes from English 0:09:55.916 --> 0:09:58.676 to kind of this big bag of numbers is one 0:09:58.716 --> 0:10:02.156 a researcher called it, right, and then it translates it 0:10:02.236 --> 0:10:04.396 back into English or whatever land which you want. Really, 0:10:04.876 --> 0:10:07.956 in the end, what if it just skip cuts out 0:10:07.956 --> 0:10:10.276 that last step, right, then we can't kind of like 0:10:11.036 --> 0:10:13.996 check what AI is doing. Then it can behave deceptively 0:10:14.596 --> 0:10:17.596 more easily. So you know, so that part seems to 0:10:17.596 --> 0:10:21.636 be important. I want to hear your your first impressions 0:10:21.676 --> 0:10:23.756 before I kind of poison the well too much. 0:10:24.916 --> 0:10:28.796 Well, my first impressions is that the alignment problem is 0:10:28.836 --> 0:10:32.436 a very real one and an incredibly important one to solve. 0:10:32.756 --> 0:10:35.276 And what I got from this is that actually, the 0:10:35.356 --> 0:10:39.476 problem that I've had with like these initial AI lms 0:10:39.996 --> 0:10:42.596 is the kernel of what they're seeing there. Right, So 0:10:42.716 --> 0:10:44.476 you and I have talked about this on the show 0:10:44.556 --> 0:10:47.596 in the past, and I've said, well, my problem is 0:10:47.596 --> 0:10:50.516 that when I'm a domain expert, right, I start seeing 0:10:50.596 --> 0:10:54.316 some inaccuracies, and I start seeing like places where like 0:10:54.356 --> 0:10:57.876 it either just didn't do well or made shit up 0:10:57.956 --> 0:11:00.956 or whatever it is. Now, I think it's very clear 0:11:00.996 --> 0:11:03.516 that those problems aren't going to go away, right, that 0:11:03.516 --> 0:11:07.036 that is going to get much much better. However, the 0:11:07.196 --> 0:11:10.436 kernel of it's showing me some thing, but that might 0:11:10.516 --> 0:11:13.076 just be you know, I have no way of verifying 0:11:13.116 --> 0:11:15.996 if that's what's going on, what it's reading, like, how 0:11:16.036 --> 0:11:18.396 it's I don't want to say thinking about it, even 0:11:18.436 --> 0:11:23.276 though in the report they do use thinking, but it's normal. 0:11:22.316 --> 0:11:24.636 Vizing too much, I think is correct. I think it's. 0:11:24.756 --> 0:11:28.316 Okay, we'll stick to that language. Yeah, okay, So, so 0:11:28.476 --> 0:11:33.276 how it's thinking about it that those little problems and 0:11:33.316 --> 0:11:36.836 like the little the glitches and the things that it 0:11:36.916 --> 0:11:40.036 might be doing where it starts actually glitching on purpose 0:11:40.436 --> 0:11:43.516 are not going to be visible to the human eye. 0:11:43.716 --> 0:11:45.516 And so one of the main things that they say 0:11:45.556 --> 0:11:50.036 here is that as AI internal R and D gets 0:11:50.316 --> 0:11:55.516 rapidly faster, so that means basically AI's researching AIS, right, 0:11:55.796 --> 0:11:59.316 and so internally they start developing new models, and as 0:11:59.356 --> 0:12:03.516 they kind of surpass human ability to monitor it, it 0:12:03.596 --> 0:12:07.396 becomes progressively more difficult to figure out, Okay, is the 0:12:07.436 --> 0:12:10.636 AI actually doing what I want to do? Is the 0:12:10.676 --> 0:12:15.076 output that it's giving me its actual thought process, and 0:12:15.356 --> 0:12:18.196 is it accurate or is it like trying to deceive me? 0:12:18.276 --> 0:12:21.356 But it's actually kind of inserting certain things on purpose 0:12:21.516 --> 0:12:24.036 because it has different goals, right, because it is actually 0:12:24.076 --> 0:12:27.436 secretly misaligned, but it's very good at persuading me that 0:12:27.476 --> 0:12:29.956 it's aligned. Because one of the things that actually came 0:12:29.996 --> 0:12:31.716 out of this report and I was like, huh, you 0:12:31.756 --> 0:12:37.356 know this is interesting is if we get this remarkable 0:12:37.636 --> 0:12:42.556 improvement in AI, it will also remarkably improve at persuading us, right, 0:12:42.636 --> 0:12:46.876 so as part of making yeah, so this is this 0:12:46.916 --> 0:12:50.116 is but I've never even thought about that. I was like, okay, fine, 0:12:50.276 --> 0:12:52.476 But one of the things that I do buy is 0:12:52.516 --> 0:12:54.996 that it's going to be very difficult for us to 0:12:55.036 --> 0:12:57.476 monitor it and to figure out like is it truly 0:12:57.516 --> 0:13:01.276 aligned with human wants, with human desires, with human goals, 0:13:01.636 --> 0:13:04.676 And the experts who are capable of doing that, I 0:13:04.676 --> 0:13:07.796 think are actually going to dwindle right as AI starts 0:13:07.796 --> 0:13:11.036 proliferating in society. And so to me, that is something 0:13:11.036 --> 0:13:13.196 that is actually quite worrisome, and that is something that 0:13:13.236 --> 0:13:15.916 we really need to be paying attention to now. Just 0:13:15.956 --> 0:13:19.596 to fast forward a little bit, in their doomsday scenario 0:13:20.236 --> 0:13:24.036 in twenty thirty one, AI takes over it basically like 0:13:24.476 --> 0:13:28.476 suddenly releases some chemical agents, right, and humanity dies and 0:13:28.516 --> 0:13:30.996 the rest of the stragglers are taken care of by drones, 0:13:30.996 --> 0:13:34.476 et cetera. I don't even like it's it doesn't even. 0:13:34.396 --> 0:13:35.516 Quick and painless death. 0:13:35.596 --> 0:13:38.876 I will say, let's hope we don't know what the 0:13:38.916 --> 0:13:42.196 chemical agents are might not be quick and painless. Some 0:13:42.276 --> 0:13:45.316 chemical agents are actually a very painful death thing. So 0:13:45.716 --> 0:13:49.116 let's hope let's hope it's quick and painless. Quick, quick, yes, 0:13:49.116 --> 0:13:53.756 stuff quick? Okay, hopefully some chemical agents are not quick. 0:13:54.756 --> 0:13:58.636 Let's hope it's it's quick and painless. But if they're 0:13:58.676 --> 0:14:01.796 actually capable of deception at that high level, then you 0:14:02.036 --> 0:14:05.196 technically don't even need them to do it. If we're 0:14:05.196 --> 0:14:08.196 trusting medicine and all sorts of things to the AIS, 0:14:08.236 --> 0:14:11.556 it's pretty easy for it to actually manipulate something and 0:14:11.716 --> 0:14:16.836 actually insert something into codes, et cetera that will fuck 0:14:16.916 --> 0:14:19.636 up humanity in a way that we can't that we 0:14:19.676 --> 0:14:22.876 can't actually figure out at the moment, right, Like the 0:14:22.916 --> 0:14:24.556 way I think of it, And this is not from 0:14:24.716 --> 0:14:26.556 the this is not from the paper, but this is 0:14:26.636 --> 0:14:29.276 just the way that like my mind processed. It is 0:14:29.316 --> 0:14:33.876 like think about DNA, right, Like you have these remarkably complex, 0:14:34.436 --> 0:14:38.796 huge strands of data, and as we've found out, but 0:14:38.876 --> 0:14:43.516 it's taken forever, one tiny mutation can actually be fatal, right, 0:14:43.556 --> 0:14:47.156 but you can't spot that mutation. Sometimes that mutation isn't 0:14:47.196 --> 0:14:50.156 fatal immediately but will only manifest at a certain point 0:14:50.196 --> 0:14:52.396 in time. That's the way that my mind tried to 0:14:52.556 --> 0:14:56.836 kind of try to conceptualize what this actually means. And 0:14:56.876 --> 0:14:59.396 so I think that, you know, that would be easy 0:14:59.476 --> 0:15:03.556 for a deceptive AI to do and to me like that, 0:15:03.556 --> 0:15:06.076 that's kind of the big takeaway from this report is 0:15:06.116 --> 0:15:08.996 that we need to make sure that we are building 0:15:09.316 --> 0:15:15.196 AIS that will not deceive right that they're capabilities, they 0:15:15.556 --> 0:15:18.756 explain them in an honest way, and that honesty and 0:15:18.796 --> 0:15:22.196 trust is actually prioritized over other things, even though it 0:15:22.276 --> 0:15:24.916 might slow down research, it might slow down other things, 0:15:24.956 --> 0:15:29.236 but that that kind of alignment step is absolutely crucial 0:15:29.396 --> 0:15:32.956 at the beginning, because otherwise humans are human, right, they're 0:15:32.956 --> 0:15:37.956 easily manipulated, And we often trust that computers are quote 0:15:38.036 --> 0:15:41.116 unquote rational because they're computers, but they're not. They have 0:15:41.196 --> 0:15:43.276 their own inputs, they have their own weights, they have 0:15:43.316 --> 0:15:47.516 their own values, and that could just lead us down 0:15:47.756 --> 0:15:48.476 a dark path. 0:15:48.556 --> 0:15:51.876 Yeah, so let me follow up with this pushback. I 0:15:51.876 --> 0:15:54.636 guess right, Like, first of all, I don't know that 0:15:54.716 --> 0:15:58.556 humans are so easily persuaded. This is my big critique 0:15:58.636 --> 0:16:03.636 with like all the misinformation people who say, well, misinformation 0:16:03.716 --> 0:16:06.196 is the biggest problem that society basis. It's like people 0:16:06.236 --> 0:16:10.476 are actually pretty stubborn and they're kind of sound pretentious. 0:16:10.916 --> 0:16:13.676 They're kind of basian and how they formulate their beliefs, right, 0:16:13.716 --> 0:16:15.796 they have some notion of reality. They're looking at the 0:16:15.836 --> 0:16:20.356 credibility of the person who is telling them these remarks. 0:16:20.676 --> 0:16:22.956 If it's an unpersuasive source, it might make them less 0:16:22.996 --> 0:16:25.516 likely to believe they're balancing what other information with their 0:16:25.556 --> 0:16:29.716 lived experience so called Right. You know, part of the 0:16:29.756 --> 0:16:33.116 reason that like I am skeptical of AIS being super 0:16:33.116 --> 0:16:36.316 persuasive is like you know that it's an AI. You 0:16:36.316 --> 0:16:37.796 know it's trying to persuade you. You know what I mean. 0:16:38.036 --> 0:16:40.156 Like if you go and play poker against like a 0:16:40.196 --> 0:16:42.636 really chatty player, if Phil Hemmi with or Scott Sieber 0:16:42.956 --> 0:16:45.396 or someone like that, Right, you know, on some level 0:16:45.476 --> 0:16:49.636 of the best play is just to totally ignore it. Right, 0:16:49.756 --> 0:16:51.716 You know that they are trying to sweet talk you 0:16:51.916 --> 0:16:55.436 into doing exactly what they want you to do, and 0:16:55.516 --> 0:16:59.196 so the best play is to disengage or literally you 0:16:59.196 --> 0:17:01.036 can randomize your movis videos no notion of what the 0:17:01.076 --> 0:17:06.236 game theoretical optimal play might be, right, or salesmen or 0:17:06.396 --> 0:17:09.836 politicians have reputations for being Oh he's a little too smooth. 0:17:09.876 --> 0:17:12.916 Gavin newsome a little too fucking smooth. Right, I don't 0:17:12.956 --> 0:17:15.076 find Gavin news some persuasive at all. Right, use a 0:17:15.076 --> 0:17:19.076 little too from the hair gel to the constantly shifting vibes. 0:17:19.116 --> 0:17:20.956 I mean, I don't really find Gavin news some persuasive 0:17:20.956 --> 0:17:22.596 at all, even though like an AI, I'd say, boy, 0:17:22.596 --> 0:17:25.796 Gavin new some good looking guys look gravelly throated. But 0:17:25.956 --> 0:17:28.276 you know whatever, I mean. Look, the big critique I 0:17:28.276 --> 0:17:31.636 have of this project, and by the way, I think 0:17:31.636 --> 0:17:35.396 this is an amazing project in addition to like wonderful 0:17:35.436 --> 0:17:37.956 writing if you viewed on the web not your phone. 0:17:38.036 --> 0:17:41.516 So these very cool, like little infographic that updates everything 0:17:41.556 --> 0:17:43.516 from like the market value. If they don't call it 0:17:43.516 --> 0:17:45.596 open aye, they call it open brain. I guess is 0:17:45.596 --> 0:17:46.796 what they settle on for a substitute. 0:17:47.036 --> 0:17:50.836 Yeah, they call everything something else, just to make sure 0:17:50.876 --> 0:17:53.476 that they're not stepping on any toes. So they have 0:17:53.636 --> 0:17:57.476 open Brain, and then they have Deep Scent from China. 0:17:57.756 --> 0:18:02.036 I wonder which one that could be. I wonder. But 0:18:02.116 --> 0:18:06.196 it's beautifully presented and written, and like I appreciate they're 0:18:06.196 --> 0:18:09.436 going out on a limb here, you know, I mean, 0:18:09.476 --> 0:18:11.676 I think they have. It's been fairly well received. They've 0:18:11.676 --> 0:18:15.156 gotten some pushback both from inside and I think outside 0:18:15.236 --> 0:18:18.276 the AI safety community. Right, but they're putting their necks 0:18:18.316 --> 0:18:21.076 in the line, hear. They will look if things look 0:18:21.076 --> 0:18:23.716 pretty normal, if looks pretty normal in twenty thirty two 0:18:23.956 --> 0:18:27.676 or whatever, right, then they will look dumb for having 0:18:27.676 --> 0:18:28.876 published this. 0:18:29.196 --> 0:18:32.676 Well, and they actually have that right as a scenario 0:18:32.796 --> 0:18:35.516 that you end up looking stupid if everything goes well. 0:18:35.796 --> 0:18:38.516 But that's okay. Now, can I push back on the 0:18:38.556 --> 0:18:42.676 persuasion thing a little bit, just on two things. So 0:18:42.836 --> 0:18:45.676 first of all, the poker example is not actually a 0:18:45.676 --> 0:18:49.956 particularly applicable one here, because you know that you're playing 0:18:50.036 --> 0:18:52.196 poker and you know that someone is trying to get 0:18:52.196 --> 0:18:55.796 information and deceive you the tricky thing. So this is 0:18:55.876 --> 0:18:57.996 kind of when I spend time with con artists. The 0:18:57.996 --> 0:19:01.356 best con artists aren't Gavin Newsome, Like, they're not car salesmen. 0:19:01.676 --> 0:19:04.036 You have no idea they're trying to persuade you to 0:19:04.076 --> 0:19:07.316 do something. They are just like nice, affable people who 0:19:07.396 --> 0:19:10.756 are incredibly charismatic, and even in the poker community, by 0:19:10.756 --> 0:19:12.796 the way, like some of the biggest grifters who like 0:19:12.836 --> 0:19:16.276 it comes out later on, we're just like stealing money 0:19:16.276 --> 0:19:19.076 and doing all of these things. Are charming, right, They're 0:19:19.116 --> 0:19:22.236 not sleazy looking like they have no signs of, oh, 0:19:22.276 --> 0:19:24.516 I'm a salesman. I'm trying to sell you something. The 0:19:24.556 --> 0:19:28.116 people who are actually good at persuasion you do not 0:19:28.316 --> 0:19:31.316 realize you are being persuaded. And I think people are 0:19:31.356 --> 0:19:35.556 incredibly easy to kind of to subtly lead in a 0:19:35.596 --> 0:19:38.756 certain direction if you know how to do it, and 0:19:38.836 --> 0:19:42.236 I think ais could do that, and they might persuade 0:19:42.236 --> 0:19:44.116 you when you don't even think they're trying to persuade you. 0:19:44.116 --> 0:19:46.436 You might just ask like, can you please summarize this 0:19:46.516 --> 0:19:49.196 research report and the way that it frames it right, 0:19:49.236 --> 0:19:53.116 the way that it summarizes it just subtly changes the 0:19:53.156 --> 0:19:55.516 way that you think of this issue. We see that 0:19:55.556 --> 0:19:57.596 in psych studies all the time, by the way, where 0:19:57.636 --> 0:20:00.956 you have different articles presented in slightly different orders, slightly 0:20:00.956 --> 0:20:04.316 different ways, and people from the same political beliefs, you know, 0:20:04.356 --> 0:20:09.436 same starting point, come away with different impressions of what 0:20:09.556 --> 0:20:11.716 kind of the right course of action is or what 0:20:11.756 --> 0:20:13.956 this is actually trying to tell you. Because the way 0:20:13.996 --> 0:20:18.236 the information is presented actually influences how you think about it. 0:20:18.236 --> 0:20:22.156 It's very very easy to do subtle manipulations like that. 0:20:22.516 --> 0:20:25.236 And if we're relying on AI on a large scale 0:20:25.276 --> 0:20:27.676 for a lot of our lives, I think that if 0:20:27.676 --> 0:20:30.396 it has like a quote unquote master plan, you know, 0:20:30.476 --> 0:20:33.236 the way that they present in this report, then persuasion 0:20:33.276 --> 0:20:34.756 in that sense is actually going to be. 0:20:35.356 --> 0:20:37.796 You'll know you're being manipulated, right, that's the issue. 0:20:37.876 --> 0:20:39.076 No, you don't know. That's the thing. 0:20:39.196 --> 0:20:42.036 People will be manipulated because it's AI and that I 0:20:42.076 --> 0:20:43.836 don't but I don't know. 0:20:43.956 --> 0:20:47.796 I honestly, like Nate, I applaud your belief in human's 0:20:47.796 --> 0:20:50.916 ability to adjust to this, but I don't know that 0:20:50.956 --> 0:20:53.916 they will because I've just seen enough people who are 0:20:53.916 --> 0:21:00.116 incredibly intelligent fall for cons and then be very unpersuadable 0:21:00.276 --> 0:21:02.996 that they have been conned, right, instead doubling down and 0:21:03.036 --> 0:21:06.196 saying no, I have not. So humans are stubborn, but 0:21:06.196 --> 0:21:08.676 they're also stubborn and saying I have not been deceived, 0:21:08.676 --> 0:21:10.996 I have not been manipulated, when in fact, they have 0:21:11.436 --> 0:21:14.036 to protect their ego and to protect their view of 0:21:14.076 --> 0:21:17.076 themselves as people who are not capable of being manipulated 0:21:17.196 --> 0:21:20.156 or deceived, and I think that that is incredibly powerful, 0:21:20.476 --> 0:21:23.596 and I think that that's going to push against your optimism. 0:21:23.636 --> 0:21:25.836 I hope you're right, but from what I know, I 0:21:25.836 --> 0:21:26.516 don't think you are. 0:21:27.436 --> 0:21:29.436 I'm not quite sure you call it optimism some I 0:21:29.436 --> 0:21:31.476 guess maybe we do like different views of human nature. 0:21:31.516 --> 0:21:33.876 But like there's not yet a substantial market for like 0:21:33.996 --> 0:21:38.836 AI driven art we're writing, and I'm sure there will 0:21:38.876 --> 0:21:44.236 be one eventually, right, But like people understand that context matters, right, 0:21:44.236 --> 0:21:45.756 that you could heavy I create a rip off the 0:21:45.796 --> 0:21:47.316 Mona Lisa, but you can also buy rip off the 0:21:47.356 --> 0:21:49.436 Mona Lisa and Canal Street for five bucks, right, and 0:21:49.516 --> 0:21:53.316 like it. You know, So it's it's the intentionality of 0:21:53.916 --> 0:21:57.556 the act and the context of the speaker. Now, sound 0:21:57.636 --> 0:22:01.836 like super woke, I guess right where you're coming from, 0:22:01.836 --> 0:22:05.556 A I think that actually is how humans communicate. Like 0:22:06.076 --> 0:22:08.596 art that might be pointless drivel coming from somebody can 0:22:08.596 --> 0:22:10.876 be something to coming from a Jackson Ballock or whatever. 0:22:10.916 --> 0:22:11.076 You know. 0:22:11.156 --> 0:22:13.436 Absolutely, I think that that's a really important point. By 0:22:13.476 --> 0:22:14.956