WEBVTT - TechSupport: 22 Jobs For An AI Future 0:00:12.960 --> 0:00:16.080 Welcome to Tech Stuff by'm moz Veloschen here with Cara Price. 0:00:16.360 --> 0:00:19.600 Hello. Hello, So I just read this piece about how 0:00:19.720 --> 0:00:23.000 AI has already started displacing jobs, but it's actually not 0:00:23.040 --> 0:00:24.040 the jobs you would think. 0:00:24.120 --> 0:00:26.640 So do we have something to celebrate this upcoming Labor Day? 0:00:26.720 --> 0:00:28.120 AI not stealing our jobs? 0:00:29.040 --> 0:00:31.680 Well, notebook LM, as you know, is trying to make 0:00:31.760 --> 0:00:36.280 podcast hosts obsolete. But no. According to MIT's State of 0:00:36.320 --> 0:00:39.279 AI and Business Report, AI is replacing jobs that are 0:00:39.360 --> 0:00:43.400 usually outsourced to other countries. That's the short term. Long term, 0:00:43.640 --> 0:00:47.320 around twenty seven percent of white collar jobs will be eliminated. Yeah. 0:00:47.360 --> 0:00:49.440 I know a number of people who are worried about 0:00:49.440 --> 0:00:52.640 the security and longevity of their jobs. But I also 0:00:52.720 --> 0:00:54.920 know people who are thinking about the kind of glass 0:00:54.960 --> 0:00:58.240 half full perspective about new jobs that AI might create. 0:00:58.800 --> 0:01:01.520 And we're sharing a conversation with one of those people today. 0:01:01.800 --> 0:01:04.720 Yeah. You and I recently spoke with journalist Robert Kapps, 0:01:04.720 --> 0:01:07.039 who has been reporting on what jobs could exist in 0:01:07.080 --> 0:01:09.640 an AI driven future, and he actually came up with 0:01:09.680 --> 0:01:12.200 a list of twenty two jobs that don't necessarily exist 0:01:12.319 --> 0:01:14.919 yet but are likely to exist when humans in AI 0:01:15.480 --> 0:01:17.720 work sort of in a hybrid capacity. 0:01:17.920 --> 0:01:18.120 Yeah. 0:01:18.160 --> 0:01:21.280 He lists a few jobs that sounded pretty interesting, and 0:01:21.280 --> 0:01:24.040 then wrote up this article for the New York Times magazine. 0:01:24.240 --> 0:01:27.720 I think my personal favorite was probably AI plumber. That's 0:01:27.760 --> 0:01:30.480 the person who will figure out why an AI system 0:01:30.600 --> 0:01:32.920 might not be working the way it should be, and, 0:01:33.080 --> 0:01:35.160 as Rob puts it, we'll snake the pipes. 0:01:35.720 --> 0:01:35.920 Yeah. 0:01:35.959 --> 0:01:38.120 And I was happy to hear him say that tastemaker 0:01:38.200 --> 0:01:39.240 jobs will last a while. 0:01:39.480 --> 0:01:41.759 Is that because you consider yourself a test maker? 0:01:41.959 --> 0:01:46.520 Yes, a nondescript creative professional. And he says that those 0:01:46.520 --> 0:01:49.720 will continue to exist, So I'm very excited. He mentioned 0:01:49.760 --> 0:01:55.040 one job called a world designer, where a person fabricates 0:01:55.040 --> 0:01:59.360 an entire universe, complete with fictional characters and locations, which 0:01:59.400 --> 0:02:02.720 could apply to everything from marketing campaigns to video games. 0:02:02.880 --> 0:02:04.800 Yeah, I remember him talking about that. He also said 0:02:04.800 --> 0:02:07.040 that he wrote the first draft of the piece using 0:02:07.120 --> 0:02:10.600 AI to see what would happen and his Robert describing 0:02:10.680 --> 0:02:12.560 how that went down with his editor, I. 0:02:12.520 --> 0:02:14.680 Thought it would be a funny joke to play on 0:02:14.720 --> 0:02:17.480 my editor, Bill Wassick that you know, he assigned me 0:02:17.560 --> 0:02:19.400 the article and then like two hours later, I'm like, 0:02:19.440 --> 0:02:24.040 here you go, here's my invoice. I love this new future. No, 0:02:25.440 --> 0:02:26.480 it was a little bit of a luck, but it 0:02:26.560 --> 0:02:29.200 was also you know, you're thinking about AI and jobs, 0:02:29.200 --> 0:02:31.880 and you're thinking about how what are the future jobs? 0:02:31.880 --> 0:02:33.640 So I just thought it was a logic place to 0:02:33.680 --> 0:02:36.640 start with. Okay, well what will my role be as writer? 0:02:37.320 --> 0:02:39.440 And you know, freelance journalism is a hard place to 0:02:39.440 --> 0:02:42.640 be I and we're not extremely well paid people anymore. 0:02:42.960 --> 0:02:44.880 So it's if I can write more, I can make 0:02:44.919 --> 0:02:47.560 more money, you know, just as a purely capitalistic play, 0:02:47.639 --> 0:02:50.880 like that's the dream, right, But of course that wasn't possible. 0:02:51.080 --> 0:02:53.880 It wasn't nearly good enough. And I would suggest this 0:02:54.000 --> 0:02:56.240 like to anybody out there who's very worried about it. 0:02:56.240 --> 0:02:57.960 It's like, just go have it, do your job. Just 0:02:57.960 --> 0:03:00.440 try it, because like it'll teach you a lot about 0:03:00.560 --> 0:03:02.960 how far it still has to go, which can be 0:03:03.000 --> 0:03:04.800 surprising in our current type cycle. 0:03:04.880 --> 0:03:07.760 Right to absolutely no one's surprised. The AI version of 0:03:07.840 --> 0:03:11.240 Robert's story was not published because he couldn't risk his 0:03:11.280 --> 0:03:13.919 reputation on an article where AI may or may not 0:03:13.960 --> 0:03:17.000 be hallucinating, or may or may not be exaggerating things 0:03:17.080 --> 0:03:18.200 or understating things. 0:03:18.440 --> 0:03:21.200 You know, We've talked about cognitive offloading a bunch on 0:03:21.240 --> 0:03:25.960 this show, basically atrophying your own skills by using AI 0:03:26.080 --> 0:03:28.639 too much, and that's something that Robert said he's very 0:03:28.639 --> 0:03:29.200 conscious of. 0:03:29.840 --> 0:03:32.400 When you start to really work with the AI, at 0:03:32.400 --> 0:03:34.600 some point you start to hand over your sense of 0:03:35.360 --> 0:03:38.200 taste and your sense of uh, like is this good. 0:03:38.240 --> 0:03:41.360 You sort of hand over the authority to the AI, 0:03:41.800 --> 0:03:44.920 and the AI doesn't have any capability to accept that authority. 0:03:45.320 --> 0:03:47.320 It will tell you that like, oh, yeah, this is great, 0:03:47.360 --> 0:03:49.560 this is true, this is but like it doesn't It's 0:03:49.640 --> 0:03:52.240 just a machine, right, so it doesn't. It doesn't really 0:03:52.280 --> 0:03:55.840 have the ability to accept that moral responsibility. Like at 0:03:55.840 --> 0:03:58.560 some level there are just these elements that like have 0:03:58.640 --> 0:04:00.840 to come from the human because the AI is just 0:04:00.920 --> 0:04:02.720 not capable of providing them. 0:04:02.760 --> 0:04:04.280 You know what I love the most is that after 0:04:04.400 --> 0:04:06.760 using AI, his next step was to reach out to 0:04:06.800 --> 0:04:09.160 a bunch of people, which I know is basic reporting, 0:04:09.200 --> 0:04:11.200 but I think it really speaks to the core of 0:04:11.240 --> 0:04:14.280 his argument that AI will need the human touch for 0:04:14.360 --> 0:04:16.080 quite some time to come right. 0:04:16.200 --> 0:04:19.159 Human work will continue, but the nature of the work 0:04:19.279 --> 0:04:22.359 may adapt to the AI Revolution and Robert have some 0:04:22.400 --> 0:04:25.480 interesting speculations about what the jobs of the future might 0:04:25.520 --> 0:04:30.960 look like and how they fit into three distinct buckets, trust, integration, 0:04:31.520 --> 0:04:32.280 and taste. 0:04:32.680 --> 0:04:34.840 So we started out by asking how he came up 0:04:34.880 --> 0:04:37.440 with these three buckets, And here's the rest of our conversation. 0:04:38.240 --> 0:04:40.359 The first person I called for the article was the 0:04:40.400 --> 0:04:45.120 Ethan Molloch, who wrote the book Cointelligence, who's Professor Wharton 0:04:45.240 --> 0:04:48.159 and who thinks and writes a ton about AI, and 0:04:48.200 --> 0:04:50.160 I just wanted somebody to talk with about like, Hey, 0:04:50.160 --> 0:04:53.279 how should I even be approaching this? And like right away, 0:04:53.360 --> 0:04:56.159 even from that first conversation, I knew that this was 0:04:56.200 --> 0:05:01.440 going to be way more philosophical than anticipated, because he wasn't. 0:05:01.680 --> 0:05:03.640 He was basically like, there's no way I can tell 0:05:03.680 --> 0:05:05.520 you specific jobs. 0:05:05.839 --> 0:05:08.360 Why did he say that he couldn't tell you specific jobs? 0:05:08.440 --> 0:05:11.400 What was the kind of intellectual exercise or missing step 0:05:11.440 --> 0:05:13.520 from where we are today to knowing what those jobs 0:05:13.520 --> 0:05:14.520 will be part of? 0:05:14.520 --> 0:05:17.440 It is just too fast moving, you know. One of 0:05:17.480 --> 0:05:20.080 his big phrases they told me a couple of times, 0:05:20.240 --> 0:05:22.360 was like it depends on how good the AI gets 0:05:22.440 --> 0:05:25.880 and how fast. Right. But then there are some things, 0:05:26.040 --> 0:05:27.799 and you know, he talks a lot about the jagged 0:05:27.839 --> 0:05:30.640 frontier of AI, which is sort of like the AI 0:05:30.680 --> 0:05:32.440 can be really good at some things and then just 0:05:32.480 --> 0:05:35.320 like really horrible at some things that you would expect 0:05:35.360 --> 0:05:39.440 even the base like estimating the word count of an article, right, 0:05:39.440 --> 0:05:42.719 like you expect any human to nail that pretty easily. 0:05:43.320 --> 0:05:45.120 But when you think about all the different jobs, or 0:05:45.160 --> 0:05:47.719 even your job, or any job, what level are you 0:05:47.800 --> 0:05:52.400 bringing some sort of moral or technical or whatever responsibility 0:05:52.440 --> 0:05:55.279 to that job, Like you are signing off on something, 0:05:55.320 --> 0:05:58.480 you are the person saying like, yes, this is good 0:05:58.520 --> 0:06:00.440 and right and best and you know whatever it be. 0:06:00.480 --> 0:06:01.840 And that can exist in a lot of different things. 0:06:02.040 --> 0:06:05.440 In writing this article, it's I'm accepting responsibility for the 0:06:05.440 --> 0:06:08.680 truthfulness and accuracy of this article, and the AI can't. 0:06:09.080 --> 0:06:11.839 It's just not part of our moral world. Right. You 0:06:11.880 --> 0:06:14.559 can't turn to the AI and blame it when something 0:06:14.600 --> 0:06:16.479 goes wrong. I mean you can, but it doesn't care. 0:06:16.560 --> 0:06:18.279 And so you know, you can think about the like 0:06:18.680 --> 0:06:22.159 extreme far end of that, in like autonomous warfare or 0:06:22.160 --> 0:06:26.200 something where like AI robots kill somebody, they can't accept 0:06:26.360 --> 0:06:30.239 moral responsibility. So the first category was trust and where 0:06:30.320 --> 0:06:33.680 are humans gonna still be very necessary? Whereas AI need 0:06:33.720 --> 0:06:38.279 them to authenticate and to provide that sort of trust. 0:06:38.839 --> 0:06:40.800 I mean, to your point. This piece appeared in the 0:06:40.800 --> 0:06:43.600 New York Times magazine. And when we think about like 0:06:44.000 --> 0:06:46.240 the history of newspapers, or the history of publishing, or 0:06:46.240 --> 0:06:49.560 the history of publishing houses, like as this concept of impremoto, 0:06:49.680 --> 0:06:53.320 like I know if this article or this book or 0:06:53.320 --> 0:06:56.680 this movie comes from this studio or this producer or whatever, 0:06:57.080 --> 0:06:59.120 that I can trust it to a certain extent. The 0:06:59.160 --> 0:07:03.480 output of AI is often entirely divorced from its creator. 0:07:03.600 --> 0:07:05.960 But you know, you mentioned in the piece some of 0:07:06.000 --> 0:07:08.719 the roles that may emerge in this trust bucket, like 0:07:09.040 --> 0:07:14.520 AI auditors, trust authenticators, AI ethicists, And I wanted to 0:07:14.560 --> 0:07:16.320 ask you, obviously, one of the things you have to 0:07:16.320 --> 0:07:18.480 think about is how much cultural demand will there be 0:07:18.560 --> 0:07:20.680 for these types of things. I mean, we're living in 0:07:20.720 --> 0:07:24.840 a world now of alternative facts and conspiracies, and I'm 0:07:24.880 --> 0:07:28.120 just wondering, do you think the demand signal will be 0:07:28.240 --> 0:07:31.000 there from the world, even if it should be morally? 0:07:31.880 --> 0:07:34.800 I think that there. I think that there are a 0:07:34.920 --> 0:07:38.360 whole lot of tasks and a whole lot of things 0:07:38.400 --> 0:07:41.880 that we really would love AI to do, right like 0:07:41.960 --> 0:07:44.160 that that we would be just fine with AI doing 0:07:44.200 --> 0:07:47.200 and in fact, we probably already do a lot of them, right, 0:07:47.240 --> 0:07:50.920 Like I have AI transcribe my interviews, and I don't 0:07:50.960 --> 0:07:53.400 think twice about the moral implications of doing that. And 0:07:53.480 --> 0:07:55.800 I think that like, as humans, we sometimes jump to 0:07:55.880 --> 0:07:59.320 these very extreme cases, right, Like I just used a warfare, 0:07:59.360 --> 0:08:02.800 but like you know, replacing human creativity and things like that, 0:08:02.840 --> 0:08:05.120 and yeah, we're gonna have to think very carefully about 0:08:05.560 --> 0:08:07.880 the lines there. But I think that there's a whole 0:08:07.880 --> 0:08:10.880 bunch of tasks that you're that you're just fine with. 0:08:10.920 --> 0:08:13.280 And so you know, one example that I kind of 0:08:13.280 --> 0:08:16.560 reference in the piece in Trust is that, like, oh, 0:08:16.600 --> 0:08:19.320 I thought that like HVAC repairmen might be some of 0:08:19.320 --> 0:08:21.800 the last people to be affected by AI, But in fact, 0:08:22.520 --> 0:08:25.880 HVAC repairmen have to do a lot of paperwork, right, right, 0:08:26.040 --> 0:08:28.360 and a lot of things that's not really core to 0:08:28.440 --> 0:08:32.160 what they enjoy about the job. But if they're using 0:08:32.200 --> 0:08:34.959 AI to do their contracts and to do their paperwork, 0:08:36.280 --> 0:08:39.320 at some point someone needs to validate that, like those 0:08:39.360 --> 0:08:42.600 contracts are accurate, that they're legally fair, because you can't 0:08:42.679 --> 0:08:46.360 trust the AI, the AI is not worthy entity. So 0:08:46.440 --> 0:08:48.960 like even trust comes in there, like somebody needs to 0:08:49.040 --> 0:08:52.480 validate that, and it becomes a little bit different when 0:08:52.520 --> 0:08:57.040 you haven't created the thing, the contract or the whatever yourself, 0:08:57.120 --> 0:08:58.920 like the same problem I was having with the article. 0:08:58.920 --> 0:09:01.080 It's like because you can't. You have to have a 0:09:01.160 --> 0:09:03.320 sort of slightly different set of skills to be able 0:09:03.360 --> 0:09:05.480 to be like I'm very familiar with where the jagged 0:09:05.559 --> 0:09:07.360 edge of AI what kind of mistakes it makes, and 0:09:07.640 --> 0:09:10.640 there really can be very weird mistakes, right, like unexpected 0:09:10.679 --> 0:09:13.960 not like it's not like backreading something a human rope, 0:09:14.040 --> 0:09:15.839 you know, and you can scale it up where it's 0:09:15.840 --> 0:09:18.560 more complicated than HVAC contracts, and it's something in an 0:09:18.640 --> 0:09:22.200 organization where they're where they have a whole chain of 0:09:22.240 --> 0:09:25.360 systems and somebody has to really know the AI well 0:09:25.480 --> 0:09:26.800 enough to be able to go through that chain and 0:09:26.840 --> 0:09:29.760 give it the like human approval of like, yes, this 0:09:29.840 --> 0:09:33.319 is this is trustworthy and accurate. And I think when 0:09:33.360 --> 0:09:36.720 I talk about ethicists, you know, a big organization might 0:09:36.760 --> 0:09:40.920 be integrating AI enough into the organization that they sort 0:09:40.920 --> 0:09:44.280 of have to have some level of explainability, right and 0:09:44.320 --> 0:09:46.600 some level of justification of like we let the AI 0:09:46.640 --> 0:09:48.960 make these decisions and not these decisions, and here's why 0:09:48.960 --> 0:09:51.400 we do it, and those things need to be explainable 0:09:51.559 --> 0:09:54.800 in a way that is satisfying to all kinds of constituents, right, 0:09:54.880 --> 0:09:58.240 so investors, customers, you know. But it can be like 0:09:58.320 --> 0:10:00.920 if the organization ends up in court, right, they have 0:10:01.000 --> 0:10:03.040 to explain it to a judge and a jury in 0:10:03.040 --> 0:10:07.360 a way that they're like, okay, that's rational and ethically sustainable. So, 0:10:07.520 --> 0:10:08.719 you know, one of the things I like to say 0:10:08.800 --> 0:10:11.360 is that like the AI boon might actually be a 0:10:12.160 --> 0:10:15.320 big boost for philosophy majors who can think through the 0:10:15.720 --> 0:10:19.000 sort of philosophical implications of how the AI is used 0:10:19.000 --> 0:10:24.239 in an organization and create a rational chain of ethical 0:10:24.640 --> 0:10:28.040 explainability for why it's done that way. Because so much 0:10:28.040 --> 0:10:31.600 of large corporations comes down to as they get bigger, 0:10:31.640 --> 0:10:34.199 they're just there's just more and more liability everywhere. 0:10:34.679 --> 0:10:37.720 Maybe one day there will be an AI hallucination interpreter, 0:10:38.559 --> 0:10:41.240 which would be very interesting. I think, I don't know. 0:10:42.120 --> 0:10:45.320 The second bucket is integration, and these jobs seem to 0:10:45.360 --> 0:10:48.040 be more technical in nature. Can you talk a little 0:10:48.040 --> 0:10:51.600 bit more about the integration category and what those jobs 0:10:51.640 --> 0:10:52.080 will be like? 0:10:52.640 --> 0:10:55.360 Sure? So another expert I talked to was a fellow 0:10:55.400 --> 0:10:58.640 named Robert Siemens, who's a professor at New York University 0:10:58.679 --> 0:11:00.920 who studies a lot of this, and he's like, well, 0:11:01.000 --> 0:11:03.040 certainly there's going to be some technical rules, right, like 0:11:03.160 --> 0:11:06.280 people need to understand the AI, and not just like 0:11:06.320 --> 0:11:08.040 we know how to build models. We know how to 0:11:08.040 --> 0:11:11.280 build AI. But these are sort of the most immediate 0:11:11.440 --> 0:11:14.600 jobs that you can see coming up that'll be very 0:11:14.920 --> 0:11:17.680 big for the next some amount of years, you know, 0:11:17.720 --> 0:11:20.920 and they may shrink as the AI changes, right, we 0:11:21.000 --> 0:11:23.679 may need less integrators over time, but like right from 0:11:23.720 --> 0:11:26.040 the jump, you need someone at your organization who really 0:11:26.120 --> 0:11:28.640 understands the models, can really dig down, can really understand 0:11:28.679 --> 0:11:30.720 what the I is doing and why, and can help 0:11:30.800 --> 0:11:35.040 map it to you know, the specific company's peculiarities and 0:11:35.160 --> 0:11:38.880 needs and wants and desires, right, And so one of 0:11:38.920 --> 0:11:41.200 the first ones he talked about was just the AI auditor, 0:11:42.000 --> 0:11:47.240 someone who can go and just really create some sort 0:11:47.280 --> 0:11:50.240 of understanding of the AI for people in the business. 0:11:51.280 --> 0:11:53.120 So you know, you can almost think of these as 0:11:53.120 --> 0:11:56.760 like a sort of twist on or an addition to 0:11:56.840 --> 0:12:01.080 your typical IT manager, but isn't quite so. Just technically focused, right, 0:12:01.160 --> 0:12:03.480 They're just like, Okay, our sales teams and models are 0:12:03.520 --> 0:12:06.200 not showing this, We're not hitting the optimization right, So 0:12:06.360 --> 0:12:08.400 just keeping up with the models and like which which 0:12:08.440 --> 0:12:11.400 one is actually best at algebra right now? Right? Like 0:12:11.480 --> 0:12:14.559 which one is actually best at writing right now? Right? 0:12:14.640 --> 0:12:17.120 It changes every few months. Just even keeping up with 0:12:17.200 --> 0:12:19.520 that if it changes at this pace, can be a 0:12:19.559 --> 0:12:22.600 pretty substantial job. So you can see AI being a 0:12:22.640 --> 0:12:25.160 great optimizer and a great tool, but someone needs to 0:12:25.240 --> 0:12:28.360 understand the system enough to work with the ad to 0:12:28.440 --> 0:12:30.720 make sure that like, oh, we're having this problem, like 0:12:30.920 --> 0:12:33.400 people aren't washing their hands enough. How can AI help 0:12:33.480 --> 0:12:37.120 us in a hospital setting? Someone needs to be thinking about, 0:12:37.160 --> 0:12:40.600 like how to make the AI get to the outcomes 0:12:40.640 --> 0:12:43.040 that they want to see. Because it's a very powerful 0:12:43.120 --> 0:12:45.040 tool for some of these things, but you know, it 0:12:45.080 --> 0:12:47.600 needs someone kind of prompting it and helping and integrating 0:12:47.640 --> 0:12:50.880 it to these very very complex, big organizations. 0:13:00.840 --> 0:13:03.480 After the break, why taste making will be the industry 0:13:03.480 --> 0:13:05.240 of the future, stay with us. 0:13:20.320 --> 0:13:24.120 So the first bucket was trust essentially human in the loop, 0:13:24.679 --> 0:13:30.959 a category of jobs around ultimately like taking responsibility right 0:13:31.320 --> 0:13:34.920 or being the final arbiter of like what is just 0:13:35.000 --> 0:13:38.839 what is legal, what is correct. The second bucket is integration, 0:13:39.000 --> 0:13:43.480 which is essentially like, how do we know enough about 0:13:43.480 --> 0:13:46.800 these tools to harness them, effectively. 0:13:46.640 --> 0:13:48.160 Optimize them and harness. 0:13:48.320 --> 0:13:50.600 The third bucket is taste, and this one I think 0:13:50.880 --> 0:13:53.720 relates most directly to what me, you and Kara do 0:13:53.840 --> 0:13:56.360 for a living. And I was curious, how did you 0:13:56.480 --> 0:13:58.520 choose that word and what does it mean in the 0:13:58.559 --> 0:14:02.319 context of your pend the jobs that may emerge. 0:14:02.920 --> 0:14:04.600 Yeah, I mean, I think taste is going to be 0:14:04.760 --> 0:14:07.360 very core to a lot of things, and not just 0:14:07.440 --> 0:14:10.000 creative jobs. It's also the one that I feel like 0:14:10.000 --> 0:14:12.080 people kind of perk up because it's it's also sort 0:14:12.120 --> 0:14:16.200 of humanly appealing. But you know, as I was just 0:14:16.240 --> 0:14:18.280 thinking about it and talking to people, and I use 0:14:18.360 --> 0:14:20.760 this in the piece, I just had this viral e 0:14:20.760 --> 0:14:24.000 clip of Rick Rubin in my head right of him 0:14:24.040 --> 0:14:27.040 on sixty Minutes and Anderson Cooper talking to him and 0:14:27.080 --> 0:14:29.200 saying like, do you know how to work a soundboard? 0:14:29.240 --> 0:14:31.080 And he's like, no, you know, want to play instruments? 0:14:31.200 --> 0:14:31.240 No? 0:14:32.040 --> 0:14:33.400 Do you know anything about music? 0:14:33.520 --> 0:14:33.560 No? 0:14:33.760 --> 0:14:36.680 And he's like, well, what what do you do? And 0:14:37.120 --> 0:14:39.360 his answer, I'll paraphrase, I don't have in front of me, 0:14:39.440 --> 0:14:44.280 is is you know, I am very confident in my taste, right, 0:14:44.320 --> 0:14:46.720 and so like I had this in my head because 0:14:47.880 --> 0:14:49.640 you know, just thinking about what do we still need 0:14:49.680 --> 0:14:51.360 humans for it? And at a base level, like the 0:14:51.680 --> 0:14:55.000 AI doesn't want anything on its own, right, Like, unprompted, 0:14:55.040 --> 0:14:58.760 it will just sit there idle. Right. So at some point, 0:14:58.880 --> 0:15:02.600 the basic human AI interaction is the human asking the 0:15:02.640 --> 0:15:06.400 AI for something. And to me, one logical route you 0:15:06.400 --> 0:15:08.640 go down if you explore that enough, is that the 0:15:08.760 --> 0:15:12.960 human is providing the taste for what it wants created 0:15:13.080 --> 0:15:15.400 and by I mean created. It could be a creative 0:15:15.440 --> 0:15:17.800 thing like a piece of music, or it could be 0:15:18.000 --> 0:15:22.160 a non creative thing like a business process. Right, But 0:15:22.240 --> 0:15:24.680 at some point somebody is looking at it and saying like, hey, 0:15:24.840 --> 0:15:27.840 I have the vision for what I want. But what 0:15:27.880 --> 0:15:30.000 they are really doing is they're making creative choices. And 0:15:30.000 --> 0:15:34.240 they're not wrong or right choices. They're choices of aesthetic 0:15:34.800 --> 0:15:37.720 or their choices of function, right, Like, so there's multiple 0:15:37.760 --> 0:15:40.320 ways to find a solution, but they're sort of using 0:15:40.320 --> 0:15:42.840 their taste and their judgment to make these creative decisions 0:15:43.160 --> 0:15:44.120 to get to their outcome. 0:15:44.840 --> 0:15:46.960 Cara I'm curious for your take here because you know, 0:15:47.040 --> 0:15:50.800 wearing your other hat, you're a TV producer, successful TV producer. 0:15:51.240 --> 0:15:53.640 How do you think about this idea of taste being 0:15:53.760 --> 0:15:58.400 a key place of human irreplaceability. 0:15:59.240 --> 0:16:04.520 Taste is certainly I think the final frontier to be 0:16:06.080 --> 0:16:08.520 messed with. I try to think of it. Does AI 0:16:08.680 --> 0:16:12.400 have taste? AI has taste insofar as what is fed 0:16:12.440 --> 0:16:14.840 to a large language model. If you feed a lot 0:16:14.880 --> 0:16:20.200 of Beethoven to a large language model and it's spinning 0:16:20.240 --> 0:16:24.760 out music that is supposed to be modeled after Beethoven 0:16:24.880 --> 0:16:28.560 or Bach or whoever, you still need a human being 0:16:29.240 --> 0:16:32.400 who's going to decide is this music good. I think 0:16:32.520 --> 0:16:38.680 things that are synthetically made by large language models can 0:16:38.800 --> 0:16:42.520 be good. But I still think there's someone who's deciding 0:16:42.800 --> 0:16:44.920 ultimately if that thing is good or not. And that's 0:16:44.920 --> 0:16:45.440 a person. 0:16:46.280 --> 0:16:48.280 And that's what I mean, at some level of someone's 0:16:48.360 --> 0:16:51.080 making those decisions. And I think, like, again, there's these 0:16:51.120 --> 0:16:54.400 far out examples of like, oh, I use the AI 0:16:54.480 --> 0:16:57.000 to do everything, and that's going to have a certain quality. 0:16:57.080 --> 0:16:59.880 But there's like, oh, I use the AI to do something, 0:17:00.080 --> 0:17:03.000 and I do some things myself, right, I do something's 0:17:03.040 --> 0:17:05.200 analog and that's going to have a different quality, right, 0:17:05.280 --> 0:17:07.199 Like those can both exist? 0:17:07.480 --> 0:17:12.480 Do you find that quality that qceing has become a 0:17:12.520 --> 0:17:15.359 harder job? Like I do notice that just even in 0:17:15.400 --> 0:17:19.160 LinkedIn posts or in Instagram posts, people are relying more 0:17:19.520 --> 0:17:23.399 on chatch ept to create the language that they're using. 0:17:23.760 --> 0:17:27.600 And there is a sort of en shitification of things 0:17:27.840 --> 0:17:31.359 because we have become used to just accepting that things 0:17:31.400 --> 0:17:34.240 aren't as good anymore because people are using chat ept 0:17:34.920 --> 0:17:37.600 to produce content. I know, is there a job that 0:17:37.680 --> 0:17:40.160 exists to push back on in shitification? 0:17:40.280 --> 0:17:43.800 I guess. I mean, like I've done enough writing experiments 0:17:43.880 --> 0:17:47.480 now that I can tell, especially like LinkedIn is just 0:17:47.560 --> 0:17:50.440 rife with it. Like there's just certain constructions M dash shit. 0:17:51.040 --> 0:17:53.320 The m dash is one but like I love m 0:17:53.400 --> 0:17:56.159 dasher is the one that gets me is the like 0:17:56.280 --> 0:17:58.840 it's not X, it's y. So like when you and 0:17:58.880 --> 0:18:01.160 you'll see these once I tell you this, you'll you'll 0:18:01.160 --> 0:18:04.080 see this on LinkedIn like every single post. It's like 0:18:04.760 --> 0:18:08.520 the future is not you know, apples and oranges, it's bananas. 0:18:08.720 --> 0:18:11.760 It's like that that construction just standing on its own 0:18:11.880 --> 0:18:13.920 is like when the AI is trying to be like 0:18:13.960 --> 0:18:17.199 a muscular copywriter, especially Claude. It just it just it 0:18:17.400 --> 0:18:20.760 just loves that construction, right, And yeah, I think that 0:18:21.320 --> 0:18:25.280 it's funny because when I think about AI in creative fields, 0:18:25.320 --> 0:18:27.320 and I try to think about it sort of more 0:18:27.359 --> 0:18:32.760 optimistic in writing, Like there's this extreme example of like 0:18:33.440 --> 0:18:35.320 I just use it to just write the whole thing, 0:18:35.400 --> 0:18:37.240 or I use it to do whatever. Like there's a 0:18:37.280 --> 0:18:40.440 lot of room between where we are and that being 0:18:40.520 --> 0:18:42.159 the outcome, and there's a lot of like sort of 0:18:42.280 --> 0:18:45.520 positive room between there. And so like one of the 0:18:45.560 --> 0:18:47.639 things that I also do is I work on documentary 0:18:47.680 --> 0:18:50.879 films and I'm making a documentary film right now about 0:18:51.000 --> 0:18:53.640 AI weapons. And it's going to cost a little over 0:18:53.640 --> 0:18:56.400 a million dollars to make that film, right, And that's 0:18:56.520 --> 0:19:01.040 that's relatively inexpensive for a documentary film. And would I 0:19:01.119 --> 0:19:03.080 love it if AI could help me make that film 0:19:03.119 --> 0:19:07.720 for five hundred thousand dollars. Like there's a lot it 0:19:07.760 --> 0:19:10.280 can't do. It's not gonna like DP my shoots and stuff. 0:19:10.320 --> 0:19:13.919 But like, you know, I have an editor, and you 0:19:13.960 --> 0:19:16.679 know it's gonna take months and months to get the 0:19:16.800 --> 0:19:19.639 edit together, and I wouldn't want to replace that editor 0:19:19.640 --> 0:19:23.760 because that editor is a valuable story collaborator who's taste 0:19:23.800 --> 0:19:26.080 I love. But what I love the tools to help 0:19:26.160 --> 0:19:28.720 him be able to do in ten weeks what takes 0:19:28.760 --> 0:19:31.359 him twenty because it can like do sort of first 0:19:31.359 --> 0:19:33.119 cuts quickly for him, and it can sort of like 0:19:33.200 --> 0:19:36.119 let him try things much faster, like so therefore we 0:19:36.200 --> 0:19:38.639 can make that documentary faster and we can move on 0:19:38.680 --> 0:19:42.280 to another one. I would absolutely love that, right, like, 0:19:42.280 --> 0:19:44.720 and I think that he would too, but like to 0:19:44.760 --> 0:19:46.359 go right to like, well, just let's just have the 0:19:46.400 --> 0:19:48.840 AI edit the whole thing and it'll be super cheap. Yeah, 0:19:48.840 --> 0:19:50.280 but it it'll suck. 0:19:51.240 --> 0:19:54.440 Now, But what about what about with this piece though? 0:19:54.480 --> 0:19:58.080 Because V one you knocked up in two hours using 0:19:58.119 --> 0:20:00.560 AI a little bit maybe came in a little bit 0:20:00.560 --> 0:20:02.720 of a strung man in the piece, right, But on 0:20:02.880 --> 0:20:05.560 V two that you wrote yourself, like, were there places 0:20:05.560 --> 0:20:08.600 where you helpfully leveraged AI to make it better? And 0:20:08.640 --> 0:20:10.639 if not, like, when do you imagine starting to do 0:20:10.680 --> 0:20:11.480 that and in what way? 0:20:12.160 --> 0:20:17.000 So I didn't use AI in this piece for that, 0:20:17.080 --> 0:20:18.639 Like I didn't, And I would say one of the 0:20:18.640 --> 0:20:21.080 reasons is I have editors at the New York Times 0:20:21.160 --> 0:20:22.320 magazine who are very good. 0:20:23.200 --> 0:20:24.880