WEBVTT - How Scared Should You Be About AI? 0:00:05.000 --> 0:00:09.039 Hey, welcome to It could happen here a podcast about 0:00:09.119 --> 0:00:13.000 things falling apart, and today it's kind of going to 0:00:13.000 --> 0:00:16.320 be a conversation about is shit falling apart? Are we 0:00:16.360 --> 0:00:19.799 all about to be devoured by a rogue AI? Is 0:00:19.840 --> 0:00:23.120 your job about to be devoured by a rogue AI? 0:00:23.560 --> 0:00:25.919 These are the questions that we're going to, you know, 0:00:26.079 --> 0:00:29.160 talk around and about and stuff today and with us 0:00:29.160 --> 0:00:34.000 today is Noah John Syracusa, a math professor at Bentley University. Noah, 0:00:34.040 --> 0:00:37.640 welcome to the show. Thanks for having me, and I'm 0:00:37.680 --> 0:00:41.120 reaching out. We're talking right now because there's an article 0:00:41.200 --> 0:00:44.080 that was put up in The New York Times on 0:00:44.159 --> 0:00:46.919 March twenty four, twenty twenty three, titled you can have 0:00:46.960 --> 0:00:49.000 the Blue Pill or the Red Pill and We're out 0:00:49.000 --> 0:00:53.000 of Blue Pills, which is a fun title by Yuval Harari, 0:00:54.000 --> 0:00:58.080 Tristan Harris and as a Raskin. And it's an article 0:00:58.160 --> 0:01:01.240 that is kind of about the pit falls and dangers 0:01:01.280 --> 0:01:05.040 of AI research, of which there definitely are some. I 0:01:05.160 --> 0:01:07.000 enjoyed your thread on the matter. I thought it was 0:01:07.040 --> 0:01:09.640 a lucid breakdown of the things the article gets right 0:01:09.720 --> 0:01:11.360 and the areas in which I think they're a bit 0:01:11.440 --> 0:01:14.160 fear mongery. So yeah, I think that's probably a good 0:01:14.160 --> 0:01:16.760 place to start, unless you wanted to start by just 0:01:16.800 --> 0:01:20.160 kind of generally talking about where you kind of are 0:01:20.160 --> 0:01:24.399 on AI and what you kind of think, you know, 0:01:24.560 --> 0:01:29.200 the technology is advancing towards right now. Yeah, I mean, 0:01:29.240 --> 0:01:31.400 I think I can probably answer both those questions and 0:01:31.440 --> 0:01:33.840 the same because part of why I enjoyed writing that 0:01:33.880 --> 0:01:37.200 threat dissecting the article is I just had the strangest 0:01:37.200 --> 0:01:39.480 feeling reading it that I agreed with it so much 0:01:39.480 --> 0:01:42.640 in principle and yet somehow objected it to so much 0:01:42.720 --> 0:01:45.920 in detail. Yeah, and it thinking about that article helped 0:01:45.920 --> 0:01:48.160 me think about my own feelings on AI, which you know, 0:01:48.240 --> 0:01:50.360 every day of the week is slightly different because so 0:01:50.440 --> 0:01:55.320 much news happens. Yeah, I found myself overall deeply frustrated 0:01:55.360 --> 0:01:57.560 that I agree with the central conclusion, which is that 0:01:58.000 --> 0:02:00.840 maybe we shouldn't be just like plowing headlong into this 0:02:00.880 --> 0:02:04.320 and should be more careful when we we we screw 0:02:04.320 --> 0:02:07.640 around with technology like this, which I agree with and 0:02:07.680 --> 0:02:09.639 I feel like should have been the thing we did 0:02:09.680 --> 0:02:12.600 with like I don't know, Facebook, Twitter, like all of 0:02:12.639 --> 0:02:16.200 these things, Like it's less my obsession is less with 0:02:16.280 --> 0:02:19.200 like the specific dangers of AI, and more with what 0:02:19.360 --> 0:02:22.880 we keep letting these guys who are fundamentally like gamblers 0:02:22.880 --> 0:02:26.440 within your capital money really put our society through the 0:02:26.520 --> 0:02:29.280 ringer without ever asking should we like do any research 0:02:29.320 --> 0:02:31.639 on maybe how social media affects children and like how 0:02:31.720 --> 0:02:35.919 all of these different things. And it's it's right that, like, yeah, 0:02:35.960 --> 0:02:38.600 we should be concerned about what these people are going 0:02:38.639 --> 0:02:41.959 to do with AI, but also why why now? Why 0:02:42.040 --> 0:02:46.040 just now? Yeah? And that raises a really good point, 0:02:46.120 --> 0:02:49.919 which is what's different now versus what we've been experiencing 0:02:49.919 --> 0:02:52.600 with social media? And just to give your listeners some context, 0:02:53.280 --> 0:02:55.560 one of the three authors on this Sneer Times article 0:02:56.160 --> 0:02:58.960 is famous for writing this book Sapiens that's a sweeping 0:02:58.960 --> 0:03:01.639 history of humanity, and the other two are actually most 0:03:01.639 --> 0:03:05.560 famous for the Netflix documentary The Social Dilemma. So they 0:03:05.720 --> 0:03:08.079 really are in this camp of warning people about social 0:03:08.080 --> 0:03:11.040 media algorithms. And as exactly as you're saying, that's sort 0:03:11.080 --> 0:03:14.520 of this thing that we've been dealing with, probably quite poorly, 0:03:14.639 --> 0:03:16.359 and now we're kind of moving on to the next 0:03:16.440 --> 0:03:19.560 societal risk, which is AI. So that as a really 0:03:19.600 --> 0:03:21.680 important question of what's different now? And I think That's 0:03:21.720 --> 0:03:23.680 one of the things the articles trying to address, which 0:03:23.760 --> 0:03:27.200 is many of the problems that we already have with algorithms, 0:03:27.280 --> 0:03:30.480 data driven algorithms, and even AI as it's used in 0:03:30.560 --> 0:03:34.200 social media is still happening now, but somehow things feel 0:03:34.240 --> 0:03:37.920 like they're aspiring out of control. Yeah, and I think, 0:03:38.520 --> 0:03:40.120 I mean, honestly, I think a lot of this just 0:03:40.360 --> 0:03:44.040 has to do with culturally, what are touchstones for AI 0:03:44.160 --> 0:03:47.800 we're going into this, you know, which are Skynet? You know, 0:03:48.120 --> 0:03:50.600 like it's that sort of thing, and you do see. 0:03:50.840 --> 0:03:54.880 I feel like the uncredited fourth author on this particular 0:03:54.960 --> 0:03:58.640 article is James Cameron, because there's pieces of it throughout 0:03:58.680 --> 0:04:03.240 this There's some it opens actually pretty provocatively. Imagine that 0:04:03.280 --> 0:04:06.160 you are boarding an airplane. Half the engineers who built 0:04:06.160 --> 0:04:07.800 it tell you there is a ten percent chance the 0:04:07.840 --> 0:04:10.320 plane will crash, killing you and everyone else on it. 0:04:10.520 --> 0:04:13.480 Would you still board? In twenty twenty two, over seven 0:04:13.560 --> 0:04:17.320 hundred top academics and researchers behind the leading artificial intelligence companies, 0:04:17.480 --> 0:04:20.200 we're asking a survey about future AI risk. Half of 0:04:20.200 --> 0:04:22.400 those surveys stated there was a ten percent or greater 0:04:22.480 --> 0:04:26.599 chance of human extinction from future AI systems. Which yeah, 0:04:27.240 --> 0:04:29.600 let's zoom in on. Yeah, yeah, let's talk about that, 0:04:31.240 --> 0:04:33.080 because what I tried to do in my thread was 0:04:33.120 --> 0:04:35.320 go through all the claims and assertions and really pause 0:04:35.400 --> 0:04:38.520 and say hold on. But that's a great one to start, 0:04:38.520 --> 0:04:41.240 because there's a lot to dig in right there. Yeah. So, 0:04:41.320 --> 0:04:44.920 first of all, there's a huge difference in that airplanes 0:04:44.920 --> 0:04:47.520 are based on science and physics and things that we 0:04:47.640 --> 0:04:51.000 understand pretty well. There's a lot to it, and there's 0:04:51.040 --> 0:04:53.080 been millions of flights, so you have a lot of data. 0:04:53.120 --> 0:04:55.039 You know, how many planes crash and how many don't. 0:04:55.480 --> 0:04:58.320 Maybe one engine goes out, you can do the statistics 0:04:58.320 --> 0:05:01.080 and CEO you know whatever, percent of planes without the 0:05:01.200 --> 0:05:04.920 engine still land safely. The problem with AI is we're 0:05:05.000 --> 0:05:08.839 just guessing, right There's no way to know one hundred 0:05:08.880 --> 0:05:10.640 years from now or ten years from now what it's 0:05:10.640 --> 0:05:13.400 going to do, what the real risks are, so we speculate, 0:05:14.080 --> 0:05:17.719 and that's not uncharted territory, right let nuclear weapons were 0:05:17.720 --> 0:05:22.200 first introduced, people had to guess and speculate. But the danger, 0:05:22.240 --> 0:05:25.360 I think is putting it in that same category as 0:05:25.400 --> 0:05:28.080 things like airplanes or climate change. I'll like to think 0:05:28.080 --> 0:05:31.480 about climate change when you see these you know, what's 0:05:31.480 --> 0:05:34.919 the IPCC if forget the acronym in these reports, that's 0:05:34.920 --> 0:05:38.240 based on thousands of scientists digging into thousands of published 0:05:38.240 --> 0:05:41.520 papers and all this data really modeling the environment. There's 0:05:41.560 --> 0:05:44.000 a lot of meat and substance to it. The problem 0:05:44.000 --> 0:05:46.280 with the AI is it's mostly people I hate to 0:05:46.320 --> 0:05:48.040 say it, but like me or like you, just kind 0:05:48.080 --> 0:05:51.719 of guessing and thinking, maybe this will happen, maybe that'll happen. 0:06:02.120 --> 0:06:04.880 The reasonable thing to say if you're in the airs, 0:06:04.880 --> 0:06:07.760 which is like, yeah, I have concerns that AI could 0:06:07.800 --> 0:06:11.720 cause serious negative externalities for the human race, perfectly reasonable statement. 0:06:11.880 --> 0:06:14.799 It is physically impossible to say there's a ten percent chance, 0:06:14.960 --> 0:06:18.720 exactly because it's never done that before. You know, I'm 0:06:18.760 --> 0:06:21.839 a math professor, and I'm the first to say numbers 0:06:21.920 --> 0:06:24.479 don't have some intrinsic meaning. Right. If I just say 0:06:24.600 --> 0:06:27.240 something has maybe a fifteen percent, I'm just making up 0:06:27.240 --> 0:06:29.160 I'm pulling out of my ass. Yeah, it doesn't make 0:06:29.200 --> 0:06:33.880 it true. So it's this, it's a general pet thieve 0:06:33.880 --> 0:06:36.600 I have of sort of giving a false sense of 0:06:36.720 --> 0:06:40.000 precision by using numbers that you don't really know where 0:06:40.000 --> 0:06:42.200 they came from or they're just made up. So that's 0:06:42.240 --> 0:06:46.159 one issue is these numbers are made up, and asking 0:06:46.200 --> 0:06:48.840 a thousand people to make up numbers isn't necessarily any 0:06:48.880 --> 0:06:51.400 better than asking one or two. You know, then if 0:06:51.400 --> 0:06:54.920 the numbers made up, it's made up. So that's one issue. Yeah, 0:06:54.960 --> 0:06:57.560 I also do think and I'm not the I saw 0:06:57.640 --> 0:07:00.120 someone making no I think it was Ben Collins, who 0:07:00.120 --> 0:07:02.839 writes for NBC. On Twitter made a note that like, well, 0:07:02.839 --> 0:07:05.239 the fact that all of these statements about like how 0:07:05.360 --> 0:07:07.760 dangerous they are about human extinction are coming out of 0:07:07.760 --> 0:07:10.400 people in the AI industry has started to kind of 0:07:10.400 --> 0:07:13.720 feel like marketing. That's right, Yeah, exactly, it's a little 0:07:13.720 --> 0:07:15.920 bit of buzz marketing going on here. And I think 0:07:15.960 --> 0:07:18.119 you mentioned social media, and the authors of this article 0:07:18.160 --> 0:07:20.720 mentioned social media, and we have to look to the 0:07:20.760 --> 0:07:22.840 past right to understand the future. I think that's the 0:07:22.840 --> 0:07:25.000 only way to do it. So, what was one of 0:07:25.040 --> 0:07:28.800 the biggest scandals in social media was Cambridge Analytica And 0:07:28.920 --> 0:07:31.480 as you know, we probably remember, this was this data 0:07:31.520 --> 0:07:34.280 privacy scandal where a bunch of data was collected from 0:07:34.320 --> 0:07:37.800 Facebook users that shouldn't have been you know, people didn't 0:07:37.800 --> 0:07:40.000 realize that the data has been collected, they didn't approve it, 0:07:40.440 --> 0:07:43.480 and it was used for this election company or this 0:07:43.880 --> 0:07:48.880 political company that was trying to profile people and influenced 0:07:48.920 --> 0:07:54.400 campaigns towards Donald Trump towards Brexit. So this was a 0:07:54.480 --> 0:07:57.360 huge scandal, and you know, Facebook was fine five billion 0:07:57.400 --> 0:08:01.080 dollars or something, very justifiably. But I would say what 0:08:01.120 --> 0:08:04.480 it was in retrospect was a data privacy issue. People's 0:08:04.520 --> 0:08:07.120 personal data was leaked when it shouldn't have been. The 0:08:07.240 --> 0:08:10.040 problem was there was so much fear and fear mongering 0:08:10.080 --> 0:08:14.720 over it that people felt this data was used by 0:08:14.720 --> 0:08:17.880 these sort of algorithmic mind lasers to kind of know 0:08:18.080 --> 0:08:20.520 us in such great detail and get us, trick us 0:08:20.520 --> 0:08:23.680 into voting for Donald Trump and targeting us. And the 0:08:24.160 --> 0:08:25.960 journey is still kind of out, but most of the 0:08:25.960 --> 0:08:29.880 evidence looks like Cambridge Analytica it wasn't that effective. They 0:08:29.920 --> 0:08:32.760 just couldn't do it. And it turns out you can 0:08:32.760 --> 0:08:34.800 know a lot about a person, a lot about their data, 0:08:34.840 --> 0:08:37.320 and it's really hard to influence them to change them. 0:08:38.040 --> 0:08:40.480 So what happened I think was there was a lot 0:08:40.520 --> 0:08:44.160 of alarm set spread rightly so about the tech companies. 0:08:44.160 --> 0:08:46.000 They have too much power, too much data, they know 0:08:46.040 --> 0:08:48.480 too much about us, and this horrible thing happened. The 0:08:48.559 --> 0:08:52.520 problem was a lot of the alarmism then actually reinforced 0:08:52.559 --> 0:08:55.280 this aura of power, of godlike power that the tech 0:08:55.280 --> 0:09:00.640 companies have. People criticizing them actually gave them more potency 0:09:00.679 --> 0:09:03.719 than they deserved. And then suddenly Google and Facebook and 0:09:03.800 --> 0:09:05.680 all they had. It wasn't sudden, but it kind of 0:09:05.720 --> 0:09:08.880 built it up. They had this aura that our algorithms 0:09:08.880 --> 0:09:11.120 are it's so insanely powerful, and we have to make 0:09:11.120 --> 0:09:12.880 sure they stay in the right hands, and we can 0:09:13.320 --> 0:09:16.360 do so much. And that's unfortunately what I see happening 0:09:16.360 --> 0:09:18.760 now a lot, and that is kind of the setting 0:09:18.800 --> 0:09:24.360 for critiquing this article. Yeah, absolutely agree that this stuff 0:09:24.480 --> 0:09:26.760 is risky. AI. I absolutely agree that we could go 0:09:26.800 --> 0:09:29.959 down to dangerous path. But once we start leaving firm 0:09:30.000 --> 0:09:32.599 ground and speculating wildly and using the terminator stuff that 0:09:32.600 --> 0:09:35.560 you described. Yeah, even if you think you're criticizing the 0:09:35.600 --> 0:09:37.520 tech companies, you know what you're doing giving them the 0:09:37.559 --> 0:09:40.400 biggest compliment in the world, saying that you guys have 0:09:40.520 --> 0:09:44.000 created are godlike and you've created these mighty machines, created 0:09:44.040 --> 0:09:46.760 a deity which is very similar to the language this 0:09:47.000 --> 0:09:49.520 argue article has at the end, and I think it's 0:09:49.559 --> 0:09:53.280 kind of worth like, as you're bringing up there are 0:09:53.559 --> 0:09:56.240 real threats. There are real threats that are immediately obvious. 0:09:56.320 --> 0:09:58.120 The threat that a lot of writers are going to 0:09:58.160 --> 0:10:01.640 lose their jobs because companies like BuzzFeed decide to replace 0:10:01.679 --> 0:10:04.480 them with you know, chat, GPT or whatever. The fact 0:10:04.480 --> 0:10:06.280 that a lot of artists are going to lose out 0:10:06.320 --> 0:10:08.120 on work because their work has been hoovered up and 0:10:08.160 --> 0:10:10.760 it's being used to generate Like these are very real 0:10:10.800 --> 0:10:13.440 and very immediate concerns that we don't have to They're 0:10:13.440 --> 0:10:16.160 not hypothetical. We don't have to theorize about the AI 0:10:16.240 --> 0:10:19.280 becoming intelligent for this to be a problem. These are 0:10:19.320 --> 0:10:23.360 things we have to immediately deal with because it puts 0:10:23.360 --> 0:10:27.520 people at risk. It's the same thing with like, you know, 0:10:27.559 --> 0:10:30.360 there's a lot that gets talked about with Cambridge Analytica, 0:10:30.440 --> 0:10:33.680 with kind of like the different Russian disinformation efforts. But 0:10:33.800 --> 0:10:35.920 when I think about the stuff that was happening in 0:10:35.960 --> 0:10:38.960 the same period that worries me more. One of the 0:10:39.000 --> 0:10:42.240 things that occurred is because there was so much money 0:10:42.240 --> 0:10:43.960 to be made if you could get certain things to 0:10:43.960 --> 0:10:47.319 go viral on YouTube, companies that use tools that weren't 0:10:47.320 --> 0:10:52.199 wildly dissimilar from some of these basically generated CGI videos 0:10:52.280 --> 0:10:55.240 based on kind of random terms that they knew were 0:10:55.320 --> 0:10:58.160 likely to trick the algorithm into trending. And god knows 0:10:58.160 --> 0:11:00.280 how many children were parked in front of these like 0:11:00.400 --> 0:11:03.040 very unhinged videos for hours at a time that like 0:11:03.320 --> 0:11:06.320 they would start watching some normal kid musical video or something, 0:11:06.320 --> 0:11:09.080 and then they're watching like the disembodied head of Krusty 0:11:09.120 --> 0:11:11.760 the Crown bounce around while like some sort of nonsense 0:11:11.800 --> 0:11:13.520 song gets sung, and it's like, well, what is that 0:11:13.559 --> 0:11:15.440 actually going to do with kids? Like, we don't know. 0:11:15.520 --> 0:11:20.720 That's unsettling, thoughttling. Yeah, And that's the kind of thing, 0:11:20.960 --> 0:11:24.040 you know, And I'm sure there will be obviously, Like 0:11:24.080 --> 0:11:25.960 one of the things that this article is not wrong 0:11:26.000 --> 0:11:29.360 about is that if we kind of leap forward into 0:11:29.400 --> 0:11:33.040 this technology with the kind of abandon that we're used 0:11:33.080 --> 0:11:36.520 to giving the tech company, there will be unforeseen externalities 0:11:36.559 --> 0:11:39.359 that we can't predict right now that will be very concerning. 0:11:39.679 --> 0:11:41.800 I just don't think it's sky in it. Yeah, And 0:11:41.840 --> 0:11:44.720 that's what was so challenging, not just with that article, 0:11:44.760 --> 0:11:47.720 but with I think the movement we're having is I 0:11:47.800 --> 0:11:51.920 do agree very much in spirit. I agree with the recommendations. 0:11:52.040 --> 0:11:54.400 We need to slow down, we need to be more 0:11:54.480 --> 0:11:59.560 judicious and cautious, we need to really consider these. But again, 0:11:59.600 --> 0:12:03.120 if we overhyped the technology, we may be doing ourselves 0:12:03.120 --> 0:12:06.760 a disservice by empowering the very entities that we're trying 0:12:06.760 --> 0:12:10.679 to take power from. And a sample like that, can 0:12:10.720 --> 0:12:13.080 I read a quick quote from the article, do you 0:12:13.640 --> 0:12:16.320 AI's new mastery of language means it can now hack 0:12:16.320 --> 0:12:21.240 and manipulate the operating system of civilization. By'm gaining mastery 0:12:21.280 --> 0:12:24.439 of language, AI is seizing the master key to civilization 0:12:24.760 --> 0:12:28.480 from bank vaults to holy sepulchers. That's right, and that 0:12:28.800 --> 0:12:30.880 I mean, that is funny, and you're right to laugh. 0:12:31.440 --> 0:12:33.480 Let's actually zoom in a second, and I think this 0:12:33.559 --> 0:12:37.480 is such a tempting trap that AI is super intelligent 0:12:37.520 --> 0:12:40.600 in some respects, right, you can. It's done amazing at 0:12:40.679 --> 0:12:43.839 chasss amazing, It's jeopard be amazing at various things. Chat 0:12:43.960 --> 0:12:47.680 Gypt is amazing at these conversations. So what happens is 0:12:47.760 --> 0:12:51.360 it's so tempting to think AI just equal super smart 0:12:51.880 --> 0:12:54.120 and because it can do those things, and now look, 0:12:54.160 --> 0:12:56.839 it can converse, that it must be the super intelligent 0:12:57.240 --> 0:13:01.400 conversational entity. And it's really good at, you know, taking 0:13:01.559 --> 0:13:03.679 text that's on the web that it's already looked at 0:13:03.720 --> 0:13:05.960 and kind of spinning it around and processing. It can 0:13:06.000 --> 0:13:09.040 come up with poems and weird forms. But that doesn't 0:13:09.120 --> 0:13:13.440 mean it is super intelligent in all respects. For instance, 0:13:13.559 --> 0:13:16.840 one of the main issues is to hack civilization. To 0:13:16.920 --> 0:13:19.760 manipulate us with language, it has to kind of know 0:13:19.840 --> 0:13:22.600 what impact its words have on us, and it doesn't 0:13:22.760 --> 0:13:25.480 really have that. It just has a little conversation a 0:13:25.559 --> 0:13:27.800 textbox and I can give it a thumbs up or 0:13:27.840 --> 0:13:30.760 thumbs down. So the only data that it's collecting for 0:13:30.840 --> 0:13:32.760 me when it talks to me any of these chatbots 0:13:33.120 --> 0:13:36.880 is did I like the response or not. That's pretty 0:13:36.960 --> 0:13:39.319 weak data to try to manipulate me, you know, it's 0:13:39.320 --> 0:13:41.960 so basic. That's not that different than when I watch 0:13:42.000 --> 0:13:44.280 YouTube videos. YouTube knows what videos I like and what 0:13:44.360 --> 0:13:48.160 I don't like. Would you say that YouTube is hacked civilization. No, 0:13:48.280 --> 0:13:51.840 it's addicted a lot of us, but it's not hacked us. Yeah. 0:13:51.880 --> 0:13:55.000 We people have hacked YouTube and that has done some 0:13:55.080 --> 0:13:59.040 damage to other people, like but it's like the thing is. 0:13:59.160 --> 0:14:02.760 And that's part of why while I have many concerns 0:14:02.880 --> 0:14:05.199 about this technology, it's not that it's going to hack 0:14:05.280 --> 0:14:08.560 civilization because like, we're really good at doing that to 0:14:08.720 --> 0:14:11.840 each other. Like there's always huge numbers of people hacking 0:14:12.320 --> 0:14:15.160 bits of the populace and manipulating each other, and there 0:14:15.200 --> 0:14:17.360 always have been. That's why we figured out how to 0:14:17.400 --> 0:14:23.000 paint like it's I do think that there's there's an 0:14:23.000 --> 0:14:28.480 interesting conversation to be had about the part of why 0:14:28.560 --> 0:14:32.680 people are kind of willing to believe anything as possible 0:14:32.680 --> 0:14:34.960 with this stuff is that for folks who were just 0:14:35.040 --> 0:14:38.240 kind of living their lives with a normal amount of 0:14:38.240 --> 0:14:41.160 attention paid to the tech industry, it seems like these 0:14:41.160 --> 0:14:43.640 tools popped out of nowhere a couple of months ago, right. 0:14:43.680 --> 0:14:46.200 It feels like, oh, there has just suddenly been this 0:14:46.280 --> 0:14:49.120 massive breakthrough. And the reality is that all of the 0:14:49.200 --> 0:14:52.080 stuff that people you know, chat gpt, these different ais 0:14:52.080 --> 0:14:54.680 that everybody's talking about, this is technology that people have 0:14:54.760 --> 0:14:57.600 been pouring resources into for years and years and years 0:14:57.600 --> 0:15:00.080 and years and years, and that's why it's able to 0:15:00.240 --> 0:15:02.200 do some of these amazing things that we've seen. But 0:15:02.280 --> 0:15:04.280 it's not I don't think it means that in a 0:15:04.360 --> 0:15:07.359 month it's going to be a thousand times smarter. It's 0:15:07.120 --> 0:15:10.440 it's it's a process of labor, and it was finally 0:15:10.480 --> 0:15:12.880 ready to be unveiled to the extent that it has been. 0:15:13.160 --> 0:15:16.920 Maybe that's right. And a good example is a GPT 0:15:17.080 --> 0:15:19.880 four which recently came out. There was GPT three before 0:15:20.400 --> 0:15:22.880 and chat Gypt, and there was so much speculation that 0:15:23.000 --> 0:15:26.600 GPT four is going to be again this godlike thing 0:15:26.640 --> 0:15:29.479 that just you know, that brings us to the singularity. 0:15:29.520 --> 0:15:32.760 And honestly, it's done better at tests. You know, I 0:15:32.800 --> 0:15:34.520 forget the numbers, but maybe one of them got a 0:15:34.560 --> 0:15:36.920 twenty percent grade on some tests and this one got 0:15:36.960 --> 0:15:40.000 an eighty percent. So that is a significant improvement. Right. 0:15:40.000 --> 0:15:42.400 If you're a teacher and your students improve that much, 0:15:42.720 --> 0:15:45.280 you should be happy, right, But as you said, is 0:15:45.320 --> 0:15:48.360 that a thousand times No, even though the machine is 0:15:48.440 --> 0:15:51.520 much bigger, much more data, and it just shows that. Yeah, Like, 0:15:51.680 --> 0:15:54.760 the reality is this is incremental progress going at a 0:15:54.840 --> 0:15:57.680 very fast rate, very unsettling even for those of us 0:15:58.000 --> 0:16:00.880 following the field closely, where it experiencing that kind of 0:16:00.920 --> 0:16:03.160 vertigo that you're saying that whoa where did this come from? 0:16:03.600 --> 0:16:05.760 So even within the field, and you're absolutely right, if 0:16:05.800 --> 0:16:08.040 you're just at home, you know, not paying attention for 0:16:08.280 --> 0:16:10.480 a week or a month or a year, suddenly the 0:16:10.520 --> 0:16:13.600 stuff pops up. It is disorienting. But one thing I 0:16:13.640 --> 0:16:16.000 think that's helped me at least kind of clarify what 0:16:17.320 --> 0:16:20.200 not even answering what the risks are, but just understanding 0:16:20.200 --> 0:16:23.360 the different camps of why certain people are reacting differently, 0:16:23.800 --> 0:16:26.040 and why even the people afraid of AI seem to 0:16:26.040 --> 0:16:28.800 be now fighting amongst each other and why it's getting fractured. 0:16:29.440 --> 0:16:33.920 Is are you more afraid of this AI used as 0:16:33.920 --> 0:16:37.040 a tool by people or are you more afraid of 0:16:37.080 --> 0:16:39.800 it kind of taking on its own autonomy and kind 0:16:39.800 --> 0:16:42.240 of going rogue and doing its own things. And I'm 0:16:42.360 --> 0:16:45.520 very much afraid of people using it. I think big 0:16:45.560 --> 0:16:47.560 companies are going to use it and there's going to 0:16:47.640 --> 0:16:49.240 be a lot of problems, just like we saw with 0:16:49.320 --> 0:16:53.040 social media. People will get addicted, democracies will be flooded 0:16:53.080 --> 0:16:57.520 with misinformation, It'll be weaponized by various actors, will be 0:16:57.560 --> 0:17:01.720 bought accounts. So I am very concerned about it being used. Basically, 0:17:01.760 --> 0:17:04.600 it performing the job it was told to do. But 0:17:04.680 --> 0:17:07.280 it'll be told to do dangerous jobs, either making money 0:17:07.359 --> 0:17:10.399 or making discord. There's another group of people that are 0:17:10.440 --> 0:17:13.520 more worried about the AI somehow deciding on its own 0:17:13.600 --> 0:17:17.320 to do things to take over. And that's where you know, 0:17:17.600 --> 0:17:19.800 I can't roll it out, But that's where I kind 0:17:19.840 --> 0:17:22.840 of am skeptical. Let's focus on how people are using 0:17:22.920 --> 0:17:26.160 it for now, for the foreseeable future. I don't think 0:17:26.160 --> 0:17:29.080 we need to worry yet, at least about the AI 0:17:29.240 --> 0:17:32.040 somehow having a life of its own and stabbing us 0:17:32.040 --> 0:17:35.160 in the back and enslaving us, because there's just so 0:17:35.240 --> 0:17:37.159 much that can go wrong before you even get to 0:17:37.200 --> 0:17:41.640 that point. Yeah, and it's it's not that's exactly like 0:17:41.720 --> 0:17:44.040 it's a threat triage kind of thing, where like, is 0:17:44.080 --> 0:17:47.080 it theoretically possible that one day human beings could create 0:17:47.119 --> 0:17:50.760 an artificial intelligence that is capable of having its own 0:17:50.880 --> 0:17:54.240 agency that is malicious? Yeah, sure, I guess, Like I 0:17:54.320 --> 0:17:57.720 mean maybe, but man, we're there's a lot of us 0:17:57.720 --> 0:18:00.960 that are very malicious right now that are actively trying 0:18:01.000 --> 0:18:04.080 to harm other people at scale. I'm concerned about how 0:18:04.080 --> 0:18:06.600 they will use AI to do that. I think botonets 0:18:06.600 --> 0:18:08.680 are a really good example. One of the things that 0:18:08.680 --> 0:18:11.879 that these these new this newest generation of AI tools 0:18:11.880 --> 0:18:15.800 allows is more realistic and intelligent bots than I think 0:18:15.840 --> 0:18:18.640 have been accessible at scale before. And that's a very 0:18:18.720 --> 0:18:23.560 real concern. Um. I will say when I kind of sorry, 0:18:23.560 --> 0:18:25.800 when I kind of wargame this back and forth with myself. 0:18:26.080 --> 0:18:30.600 One thing that is oddly comforting is like, well, the 0:18:30.760 --> 0:18:35.760 shared comments that we all inhabit of, like ontological truth 0:18:36.119 --> 0:18:39.719 is already so shattered that like there's there's only so 0:18:39.840 --> 0:18:43.680 much damage. I feel like adding additional bots and additional 0:18:43.720 --> 0:18:48.440 disinformation can really do um. Like I done one thought 0:18:48.480 --> 0:18:50.840 on that though, because I've been digging into that too. 0:18:51.160 --> 0:18:53.040 I've been, you know, trying to ponder how to feel 0:18:53.040 --> 0:18:54.719 about that, because a lot of this I don't know, 0:18:54.800 --> 0:18:58.480 you know, I'm trying to make is. I do think 0:18:58.680 --> 0:19:02.800 if you go back to like two sixteen earlier versions 0:19:02.800 --> 0:19:05.840 of the Internet before leading up to Donald Trump's election, 0:19:06.520 --> 0:19:10.200 I think there was a lot of wild West to Google, 0:19:10.240 --> 0:19:12.359 to social media, to all these things. Right, fake news 0:19:12.440 --> 0:19:14.720 was just like piling up to the top of Google 0:19:14.760 --> 0:19:18.800 search results. That election was so monumental and such a 0:19:19.200 --> 0:19:23.080 seismic shockwave through tech that fake news and misinformation might 0:19:23.080 --> 0:19:25.960 have played a role that they really had to do something, 0:19:26.119 --> 0:19:28.120 and I think some companies are more effective than others. 0:19:28.119 --> 0:19:30.680 I think Google put a lot of effort into making 0:19:30.680 --> 0:19:33.800 sure authoritative sources rise to the top. So what that 0:19:33.880 --> 0:19:36.399 means is when now you go online and you google 0:19:36.440 --> 0:19:40.040 for medical information, the top results you get are WebMD 0:19:40.359 --> 0:19:44.160 or some official CDC, your government thing. They're pretty decent reliable. 0:19:44.720 --> 0:19:46.480 It's not to say there's an all that crap on 0:19:46.520 --> 0:19:48.800 the Internet, but Google has done a pretty good job 0:19:48.880 --> 0:19:51.280 of having the good stuff float to the top, and 0:19:51.320 --> 0:19:54.720 that's the information that people see. So what I'm worried 0:19:54.880 --> 0:19:57.440 is now we might be kind of resetting ourselves back 0:19:57.480 --> 0:19:59.719 to the twenty sixteen where when you're talking to these 0:19:59.760 --> 0:20:02.679 chat bots that are trained on all the internets. Yeah, 0:20:03.280 --> 0:20:06.000 I don't know if the web mds and the CDC 0:20:06.640 --> 0:20:09.120 type of information is necessarily going to float to the top. 0:20:09.240 --> 0:20:12.080 Maybe they'll work that out. But I'm also worried that 0:20:13.240 --> 0:20:15.879 open Ai or Google or Microsoft for wherever, they'll have 0:20:15.920 --> 0:20:18.320 ones that are pretty reasonable and kind of you know, 0:20:18.520 --> 0:20:20.800 tuned to appeal to a lot of people. But Elon 0:20:20.920 --> 0:20:23.679 must might build his own competitor, one that might be 0:20:23.760 --> 0:20:27.120 really tuned to elevate the right wing site your car. 0:20:38.320 --> 0:20:40.679 So I have been messing around, as I mean, and 0:20:40.760 --> 0:20:42.720 you have been doing so in a much more rigorous manner, 0:20:42.760 --> 0:20:44.639 I'm sure. But I've screw around with a couple of 0:20:44.680 --> 0:20:48.640 different AI chat and search engines. I use find PHI 0:20:48.760 --> 0:20:52.119 and D sometimes I've been playing around with BING and 0:20:52.359 --> 0:20:54.280 one of the things I've noticed is that you know, 0:20:54.280 --> 0:20:56.800 if you ask it like, hey, summarize for me, like 0:20:56.840 --> 0:20:59.000 why the Battle of Hastings mattered, You'll get a reasonably 0:20:59.040 --> 0:21:02.240 decent answer. But if I ask it like I don't know, 0:21:02.359 --> 0:21:05.959 specific questions about myself, I've come to I noticed at 0:21:05.960 --> 0:21:07.480