WEBVTT - Week in Tech: Turning China's Surveillance Against Itself 0:00:13.800 --> 0:00:17.400 From Kaleidoscope and iHeart podcasts. This is tech stuff. I'm 0:00:17.400 --> 0:00:18.400 as Voloscian and. 0:00:18.360 --> 0:00:19.200 I'm care Price. 0:00:19.440 --> 0:00:22.119 Today we get into what it's like to be a 0:00:22.200 --> 0:00:25.600 human who trains AI for a living, and the story 0:00:25.640 --> 0:00:29.120 of an activist who turned the surveillance state against the 0:00:29.200 --> 0:00:33.280 Chinese government. Then on chatting me, a real life therapist 0:00:33.680 --> 0:00:35.760 talks about how he uses AI. 0:00:36.200 --> 0:00:38.879 It felt like a steady partner. But I call a 0:00:39.000 --> 0:00:44.040 cognitive prosthesis, not thinking for me, but helping me organize 0:00:44.080 --> 0:00:46.960 and extend my own thinking. The way of walking, stick 0:00:47.080 --> 0:00:48.160 supportune as. 0:00:48.120 --> 0:00:53.159 You walk, all of that. On the Weekend Tech, It's Friday, September. 0:00:52.640 --> 0:01:02.200 Twelfth, Hi Cara, Hi ahs. 0:01:02.440 --> 0:01:05.600 So normally we record these together in studio, but this 0:01:05.640 --> 0:01:08.520 week I'm in a hotel room in Munich, Germany, at 0:01:08.520 --> 0:01:11.440 a conference called DLD Digital Life Design. 0:01:11.680 --> 0:01:15.039 It sounds very lively. How is it? 0:01:15.040 --> 0:01:17.040 It's good? You know, are you familiar with this concept 0:01:17.040 --> 0:01:19.920 of ambient AI that they're using medical settings? 0:01:19.959 --> 0:01:22.040 Sort of yes? Can you just tell me a little 0:01:22.080 --> 0:01:22.560 bit about it? 0:01:22.800 --> 0:01:27.479 So basically, it's like cameras and sound recording in settings 0:01:27.520 --> 0:01:30.640 where like a person doesn't have to go and use AI. 0:01:30.800 --> 0:01:33.240 AI is just running in the background and feeding back 0:01:33.319 --> 0:01:35.759 kind of insights and data. And so there's a talk 0:01:35.880 --> 0:01:39.800 yesterday about this being used in hotel settings for food, 0:01:39.800 --> 0:01:43.400 which I found pretty interesting. But apparently using this technology 0:01:43.560 --> 0:01:46.120 has in some cases reduced food waste in hotels by 0:01:46.120 --> 0:01:47.400 an average of forty percent. 0:01:47.800 --> 0:01:50.040 What does ambient AI look like in this case, Like, 0:01:50.120 --> 0:01:53.000 is it like a panopticon or people getting publicly shamed? 0:01:54.560 --> 0:01:56.240 Well, it is a bit like a panopticon, but the 0:01:56.320 --> 0:01:59.160 customers are not being publicly shamed on their third trip 0:01:59.200 --> 0:02:02.600 to the buffet. Basically, there are cameras and there's also 0:02:02.640 --> 0:02:07.520 a weighing device for trash bins, and so it will 0:02:07.600 --> 0:02:10.359 track the food at the end of the buffet being 0:02:10.760 --> 0:02:13.720 transported by the servers to the trash and thrown out 0:02:13.760 --> 0:02:16.840 and then being weighed. And the insights and not rocket science. 0:02:16.880 --> 0:02:20.400 It's like, hey, guys, don't fill the bread basket fifteen 0:02:20.440 --> 0:02:22.800 minutes before the service ends. Like things that once you 0:02:22.880 --> 0:02:25.360 hear them, are obvious, but actually it's interesting that like 0:02:25.600 --> 0:02:28.680 being told it by technology actually apparently makes a real difference. 0:02:29.120 --> 0:02:31.959 I remember when you and I many now many years 0:02:32.000 --> 0:02:34.920 ago went to speak at Johns Hopkins. They told us 0:02:34.960 --> 0:02:39.160 about connected devices that nurses wear to make hospital work 0:02:39.240 --> 0:02:41.360 more efficient. So I think this is something that we 0:02:41.360 --> 0:02:44.639 should pay attention to. But speaking of digital life design, 0:02:44.840 --> 0:02:46.360 I don't know if you saw this article in the 0:02:46.360 --> 0:02:49.640 Wall Street Journal called My Day as an eighty year old? 0:02:49.680 --> 0:02:51.160 What an age simulation suit? 0:02:51.160 --> 0:02:52.720 Top me? Did you see this? I saw the email 0:02:52.760 --> 0:02:54.240 and I was fascinated, But tell me more. 0:02:54.400 --> 0:02:59.440 Basically, MIT has created this age simulation suit, which is 0:02:59.480 --> 0:03:02.160 a full body suit that was built by scientists at 0:03:02.160 --> 0:03:05.120 the Age Lab, which does research on how to improve 0:03:05.240 --> 0:03:08.040 life for the elderly. So the journalists who wrote the article, 0:03:08.160 --> 0:03:10.880 her name is Amy, doctor Marcus, and she recently wore it, 0:03:10.919 --> 0:03:12.720 and that's what you know she wrote this article about. 0:03:12.760 --> 0:03:15.880 And she said she was fitted with just imagine this, 0:03:15.960 --> 0:03:19.840 a fifteen pound weighted vest, more weights on her ankles 0:03:19.840 --> 0:03:23.480 and wrists to kind of stimulate the loss of muscle mass. 0:03:23.600 --> 0:03:26.720 And the suit also had this very elaborate bungee cord 0:03:26.840 --> 0:03:30.960 system and a neck collar that limited her mobility. She 0:03:31.000 --> 0:03:33.800 also had glasses. And there's great photos you if people 0:03:33.800 --> 0:03:36.120 who are listening want to look it up on the 0:03:36.160 --> 0:03:38.640 Wall Street Journal. There's these great photos of like what 0:03:38.680 --> 0:03:41.960 her distorted vision looks like she also wore padded crocs 0:03:42.000 --> 0:03:43.560 that made it really hard to balance. 0:03:44.000 --> 0:03:45.560 This sounds like hell. 0:03:45.360 --> 0:03:49.040 It sounds like the worst hangover of your life. Basically, yeah, 0:03:49.200 --> 0:03:53.400 MIT calls the suit agnes, which is an acronym for 0:03:53.640 --> 0:03:55.680 age gain now empathy system. 0:03:55.800 --> 0:03:58.440 That's about acronym. It sounds when I a bachronym that 0:03:58.600 --> 0:04:01.440 I but I get it. What made the story stand 0:04:01.480 --> 0:04:01.840 out to you? 0:04:02.080 --> 0:04:04.840 So you know, I think it's a very interesting lesson 0:04:04.840 --> 0:04:07.160 in empathy, or a lesson in schlepping if you come 0:04:07.200 --> 0:04:11.760 from my culture. But it's actually a part of a 0:04:11.840 --> 0:04:14.920 longer arc of product research for older people, which I'm 0:04:15.000 --> 0:04:18.839 very interested in as someone who has an aging parent 0:04:18.920 --> 0:04:21.839 who is very young its spirit. I think it's a 0:04:21.880 --> 0:04:24.440 real test of our empathy to understand how people who 0:04:24.440 --> 0:04:27.000 are older than us navigate the world. And there was 0:04:27.040 --> 0:04:29.640 this woman who I became obsessed with many years ago 0:04:29.720 --> 0:04:32.360 named Patricia Moore. Patty Moore? Have you ever heard of her? 0:04:32.560 --> 0:04:33.000 I haven't. 0:04:33.240 --> 0:04:36.640 So she is most famous for pioneering an approach called 0:04:36.760 --> 0:04:41.240 universal design, which basically means designing products and spaces with 0:04:41.360 --> 0:04:44.880 empathy for the widest possible audience in mind. So like, 0:04:45.520 --> 0:04:48.920 will this refrigerator be something that a ninety year old 0:04:48.960 --> 0:04:50.360 can open, for example. 0:04:50.560 --> 0:04:52.800 Yeah, it's interesting, and I guess it's kind of become 0:04:52.839 --> 0:04:54.920 a bit of a standard now. I'm not sure if 0:04:55.000 --> 0:04:58.119 companies always do it, but certainly companies always talk about 0:04:58.160 --> 0:05:01.920 the idea they don't want to design products with the 0:05:01.960 --> 0:05:05.719 healthy mid thirties product designer in mind as the user, 0:05:06.000 --> 0:05:07.920 but they won't have a wider view on how these 0:05:07.960 --> 0:05:10.560 products might be used by people different ageas people with 0:05:10.560 --> 0:05:11.839 different abilities, that kind of thing. 0:05:12.040 --> 0:05:15.640 To your point, it wasn't always a consideration. And when 0:05:15.680 --> 0:05:18.640 Patty was twenty six years old, this is in nineteen 0:05:18.680 --> 0:05:21.000 seventy nine. I'm giving away her age, and I know 0:05:21.040 --> 0:05:22.520 she's going to listen to this because she's a friend 0:05:22.560 --> 0:05:22.800 of mine. 0:05:22.800 --> 0:05:25.320 But so she's been at this. She's been at this 0:05:25.600 --> 0:05:27.120 since way before it was popular. 0:05:27.279 --> 0:05:29.760 She worked at Raymond Lowie. There was some grant from 0:05:29.760 --> 0:05:33.080 the Nixon administration that allowed her to have the job 0:05:33.120 --> 0:05:36.200 that she had, But she was one of the only 0:05:36.320 --> 0:05:39.760 people who was bringing up any kind of conversation about 0:05:40.160 --> 0:05:44.720 product design for the elderly and how we can engage 0:05:44.720 --> 0:05:48.279 in a conversation about that. But what she did that 0:05:48.400 --> 0:05:52.880 is so kind of different. But ingenius visa vi agnes Is. 0:05:52.960 --> 0:05:55.520 She got a friend of hers who was a makeup 0:05:55.600 --> 0:05:59.479 artist from Saturday Night Live, to turn her into an 0:05:59.520 --> 0:06:03.719 eighty five year old woman. She traveled around the country 0:06:03.839 --> 0:06:07.280 in these outfits essentially, and she got a really good 0:06:07.320 --> 0:06:10.520 sense of what it was like to live as an 0:06:10.520 --> 0:06:13.680 elderly person in the world in the late nineteen seventies. 0:06:13.920 --> 0:06:16.719 After she did this, she went on to be a 0:06:16.720 --> 0:06:20.760 product engineer and she has based her entire career on 0:06:21.440 --> 0:06:24.440 creating more accessible products, like you know, good grips. She 0:06:24.600 --> 0:06:26.960 was a very sort of integral part of the design 0:06:27.000 --> 0:06:29.680 of easily grippable devices for the kitchen and the home. 0:06:29.920 --> 0:06:32.840 She also worked on transit systems, medical devices, and now 0:06:32.880 --> 0:06:35.880 she actually spends a lot of time consulting tech companies 0:06:35.920 --> 0:06:38.920 on how to make their products more inclusive for elderly people. 0:06:39.040 --> 0:06:42.159 To me, it's interesting tech then and now story because 0:06:42.160 --> 0:06:44.679 in the seventies, like to do this kind of research, 0:06:44.720 --> 0:06:48.120 you'd have to invent your own costumes and characters and 0:06:48.160 --> 0:06:50.159 process and go to a TV show to help you 0:06:50.200 --> 0:06:53.440 achieve it. And now MIT has this age scented lab 0:06:53.480 --> 0:06:55.279 where you can just put on the agnes suit for 0:06:55.320 --> 0:06:57.720 a day, but of course, the motivation is the same, 0:06:57.880 --> 0:06:59.360 and part of that might speak to the fact that, 0:06:59.560 --> 0:07:02.200 howe him, we want to pay lip service the idea 0:07:02.200 --> 0:07:05.080 of designing for all kinds of different people, it remains 0:07:05.279 --> 0:07:06.720 very hard to pull off. 0:07:06.680 --> 0:07:09.360 It does. I think the one thing that's interesting about 0:07:09.360 --> 0:07:11.880 the Angle and the Journal piece is that the Age 0:07:11.920 --> 0:07:16.800 Lab is trying to also understand how to prepare people 0:07:16.880 --> 0:07:20.520 who are aging into their eighties to live better. And 0:07:20.560 --> 0:07:23.560 it's really about how do you keep people staying young 0:07:23.680 --> 0:07:24.920 even though they're not young anymore? 0:07:25.120 --> 0:07:28.520 Right, right, right, Well, it's all only either beholder. Do 0:07:28.520 --> 0:07:32.560 you know two people who are very interested in staying 0:07:32.600 --> 0:07:36.320 young even if they age in terms of their physical bodies? 0:07:36.480 --> 0:07:39.480 Every woman I know, and many. 0:07:41.960 --> 0:07:45.679 But now in this case, I'm talking about Vladimir Putin 0:07:45.920 --> 0:07:49.280 and Shechinping. The two of them are attending a military 0:07:49.280 --> 0:07:53.480 parade in Beijing, officially to celebrate the eightieth anniversary of 0:07:53.640 --> 0:07:57.040 the end of World War II. Unofficially, of course, to 0:07:57.120 --> 0:08:00.480 show the US and Taiwan. Then you ask some of 0:08:00.560 --> 0:08:03.960 high tech weaponry. But anyway, Putin and she were walking 0:08:04.000 --> 0:08:07.720 together when a hot might caught them having a really 0:08:08.040 --> 0:08:11.760 bizarre conversation. I'll read it to you. So Putin's interpreter 0:08:11.880 --> 0:08:17.680 says in Chinese, quote, biotechnology is continuously developing. Then he adds, 0:08:17.960 --> 0:08:24.200 quote human organs can be continuously transplanted. The longer you live, 0:08:25.120 --> 0:08:29.440 the younger you become, and you can even achieve immortality. 0:08:30.280 --> 0:08:34.640 She responds, quote some predict that in this century humans 0:08:34.640 --> 0:08:36.720 may live to one hundred and fifty years old. 0:08:37.200 --> 0:08:40.600 Just imagine that your small talk is about organ transplantation. 0:08:41.360 --> 0:08:45.959 Continuous organ transplantation. I mean once once, you might think 0:08:45.960 --> 0:08:46.560 will be enough. 0:08:46.679 --> 0:08:51.680 It's unbelievable. And that's their casual conversation, the most powerful 0:08:51.720 --> 0:08:52.560 people in the world. 0:08:53.480 --> 0:08:56.120 The reason I came across this is because I'm quite 0:08:56.120 --> 0:08:58.600 interested in defense tech and I was looking for the 0:08:58.640 --> 0:09:01.800 most interesting angle on this military parade in Beijing to 0:09:01.840 --> 0:09:04.520 bring to the show. And there was fascinating stuff about 0:09:04.600 --> 0:09:08.120 new anti drone technologies, new kinds of lasers, et cetera, 0:09:08.160 --> 0:09:12.200 et cetera. But actually the most interesting story I've found 0:09:12.679 --> 0:09:15.160 had nothing to do with the military parade. In fact, 0:09:15.240 --> 0:09:17.760 it was all about what happened the night before. 0:09:18.080 --> 0:09:19.000 Tell me about it. 0:09:19.160 --> 0:09:22.520 So, around ten pm last Friday night in chong Qing, 0:09:23.080 --> 0:09:26.520 a huge projection against the skyscraper went up calling for 0:09:26.559 --> 0:09:29.920 an end to the Communist Party's rule. It included slogans 0:09:29.960 --> 0:09:32.880 like quote only without the Communist Party, can there be 0:09:32.880 --> 0:09:36.080 a new China? End quote, no more lies, we want 0:09:36.080 --> 0:09:39.000 the truth, no more slavery, we want freedom. 0:09:39.120 --> 0:09:42.080 And how long did that stay up? Like ten seconds? 0:09:42.640 --> 0:09:47.360 Well an hour? But actually this is where things get 0:09:47.360 --> 0:09:50.040 more interesting and more tech stuff. A few hours after 0:09:50.080 --> 0:09:53.560 the projection was shut off, the activist who staged the protest, 0:09:53.800 --> 0:09:57.480 See Hoong, posted this video showing five police officers bursting 0:09:57.520 --> 0:10:00.200 into the hotel room where the projector was and then 0:10:00.280 --> 0:10:02.720 kind of bumbling around trying to turn it off. That 0:10:02.800 --> 0:10:05.640 at one point one of the policemen notice, smile your 0:10:05.679 --> 0:10:08.240 on candid camera. There's actually a camera pointing at them, 0:10:08.320 --> 0:10:11.000 and he rushes towards it, looking really surprised. And then 0:10:11.640 --> 0:10:13.920 he looks down and sees as a note underneath the 0:10:13.960 --> 0:10:17.280 camera saying, even if you are a beneficiary of the 0:10:17.320 --> 0:10:20.720 system today, one day you will inevitably become a victim 0:10:20.760 --> 0:10:23.880 on this land, so please treat the people with kindness. 0:10:24.240 --> 0:10:26.440 But it plays out almost like a Marx Brothers film. 0:10:26.440 --> 0:10:28.960 They burst into the room, they're confused, and they realize 0:10:29.000 --> 0:10:31.160 that they're kind of the victims of this prank. 0:10:31.600 --> 0:10:33.160 Did he get in trouble for this. 0:10:33.280 --> 0:10:35.120 Well, he would have gone into terrible trouble, but of 0:10:35.160 --> 0:10:37.120 course he wasn't in the hotel room. That was the 0:10:37.160 --> 0:10:39.880 whole joke of just a camera. He'd fed the country 0:10:39.920 --> 0:10:42.960 with his family a week before the protest was turned on, 0:10:43.559 --> 0:10:45.959 and in fact he turned it on remotely. But one 0:10:45.960 --> 0:10:48.559 of his brothers along with a friend of his were 0:10:48.559 --> 0:10:53.120 both detained, and authorities also questioned his elderly mother outside 0:10:53.160 --> 0:10:56.360 her home. Now talk about Agnes elderly mother. I mean, 0:10:56.360 --> 0:10:58.920 this woman looks like she's one hundred years old, completely 0:10:59.000 --> 0:11:03.680 hunched over, surrounded by cops. Is an extremely unfavorable image 0:11:03.679 --> 0:11:06.320 for the police. And see Hong again actually managed to 0:11:06.320 --> 0:11:08.760 get hold of these surveillance photos. Then he posted the 0:11:08.800 --> 0:11:11.280 images of his mother's interrogation online as well. 0:11:11.520 --> 0:11:15.480 This is very who watches the watchman irl? You know, 0:11:15.520 --> 0:11:18.760 he's using the Chinese authority's own tools essentially against them. 0:11:18.920 --> 0:11:21.480 That's he's actually right. And Seahong himself was quoted in 0:11:21.520 --> 0:11:25.400 the story saying, the party installs surveillance cameras to watch us. 0:11:25.840 --> 0:11:28.320 I thought I could use the same method to watch them, you. 0:11:28.320 --> 0:11:30.800 Know, I think we think of huge tech systems as 0:11:30.840 --> 0:11:34.200 fully autonomous, but a story like this really reveals how 0:11:34.240 --> 0:11:37.720 a single human being can still have an outsized effect. 0:11:38.240 --> 0:11:41.960 And you know, what this activist did was essentially reveal 0:11:42.080 --> 0:11:45.200 that there is no such thing as a fully autonomous system. 0:11:45.240 --> 0:11:47.560 And I think the footage of the police arriving in 0:11:47.600 --> 0:11:50.680 the empty apartment and displaying this very sort of human 0:11:50.800 --> 0:11:54.679 confusion as a testament to that which humans being inside 0:11:54.720 --> 0:11:57.120 the lubnol brings me to the big story that I 0:11:57.120 --> 0:11:59.160 want to bring to you this week, and it's do 0:11:59.200 --> 0:12:01.960 you know about the people who train AI models to 0:12:02.000 --> 0:12:02.440 be better? 0:12:02.679 --> 0:12:05.600 I actually do, because I have a friend who I 0:12:05.640 --> 0:12:07.760 actually asked if they would come on the show some 0:12:07.800 --> 0:12:10.679 weeks ago, and they said no because they have NDAs 0:12:10.800 --> 0:12:13.800 and they rely on their AI trainer work for part 0:12:13.800 --> 0:12:16.160 of their income, so they couldn't do it. So now 0:12:16.200 --> 0:12:18.360 you brought this story because I've wanted to know more 0:12:18.360 --> 0:12:19.240 about it for some time. 0:12:19.480 --> 0:12:22.320 So that's exactly what this Business Insider article is about. 0:12:22.440 --> 0:12:26.120 They spoke to over sixty people who are basically the 0:12:26.240 --> 0:12:29.800 humans behind the AI boom. They're called data labelers, and 0:12:30.240 --> 0:12:32.920 there are hundreds of thousands of them around the world. 0:12:32.960 --> 0:12:35.160 You'll remember, you know, we used to talk on Sleepwalkers 0:12:35.200 --> 0:12:39.400 about content moderation, which was a sort of similarly arduous, 0:12:39.800 --> 0:12:43.920 sometimes disturbing task. These people do exactly what your friend does, 0:12:43.960 --> 0:12:47.679 which is that they spend hours and hours feeding chatbots 0:12:47.760 --> 0:12:52.040 test prompts, and then categorizing their responses according to whether 0:12:52.080 --> 0:12:57.760 they're helpful, accurate, concise, natural sounding, or wrong, rambling, robotic, 0:12:57.960 --> 0:13:03.000 and sometimes even offensive. In the Business Insider article, they 0:13:03.040 --> 0:13:08.080 are described as quote part speech pathologists, part manners tutors, 0:13:08.120 --> 0:13:09.959 and part debate coaches. 0:13:09.800 --> 0:13:11.520 That teachers in an elite prep school. 0:13:13.040 --> 0:13:13.559 Exactly. 0:13:14.280 --> 0:13:17.280 I never asked my friend because I thought it was indelicate, 0:13:17.320 --> 0:13:20.280 But I'm curious what kind of money do these people make. 0:13:20.480 --> 0:13:24.320 So one of the companies that oversees these trainers is 0:13:24.360 --> 0:13:28.200 called Outlier, and they actually said that they've collectively paid 0:13:28.440 --> 0:13:31.319 and this is a quote, hundreds of millions of dollars 0:13:31.320 --> 0:13:34.280 in the past year alone for data labelers. One of 0:13:34.320 --> 0:13:37.880 the labelers quoted in the article actually made fifty thousand 0:13:37.960 --> 0:13:40.400 dollars in six months at one point, which is a 0:13:40.440 --> 0:13:41.480 decent amount of money. 0:13:41.679 --> 0:13:44.360 Ten thousand dollars a month almost it's not bad. I 0:13:44.400 --> 0:13:47.160 mean how much data labeling do you have to do 0:13:47.240 --> 0:13:49.120 to make fifty thousand dollars and six months. 0:13:49.200 --> 0:13:50.920 Well, when I say this, it might not sound like 0:13:51.000 --> 0:13:52.840 so much more, but it's a full time job, so 0:13:52.880 --> 0:13:55.040 you're working about fifty hours a week. But you know, 0:13:55.080 --> 0:13:57.360 one of the difficult parts about this work is that 0:13:57.400 --> 0:13:59.440 the rates can change on a whim. You know, at 0:13:59.480 --> 0:14:03.520 one point outlier, this company change one contractor's rate from 0:14:03.760 --> 0:14:06.200 fifty an hour to fifteen dollars an hour with absolutely 0:14:06.240 --> 0:14:10.160 no explanation. Also, like a lot of seemingly steady streams 0:14:10.160 --> 0:14:13.160 of work can dry up with absolutely no explanation. And 0:14:13.240 --> 0:14:16.400 so one person was basically like, my job is like gambling. 0:14:16.640 --> 0:14:20.120 Yeah, because one of those types of work where it's finite, right, Like, 0:14:20.200 --> 0:14:22.440 once you've trained the thing, you don't need to train 0:14:22.480 --> 0:14:24.920 it anymore. And so there's this kind of inherent issue 0:14:24.960 --> 0:14:27.840 at the heart of this story where real humans are 0:14:27.960 --> 0:14:31.720 essentially accelerating putting themselves out of work by doing this 0:14:31.880 --> 0:14:32.560 kind of work. 0:14:32.720 --> 0:14:35.400 Absolutely. One of the things that I actually found striking 0:14:35.880 --> 0:14:38.160 is that the article says that a lot of times 0:14:38.240 --> 0:14:42.120 training chatbots to work better means actually doing things like 0:14:42.160 --> 0:14:46.360 treating them worse or testing how they handle potentially harmful prompts, 0:14:46.680 --> 0:14:49.360 and actually, in some cases, the more the humans can 0:14:49.400 --> 0:14:52.360 get a bot to say something inappropriate, the more they 0:14:52.400 --> 0:14:55.000 get paid. So one of these data labelers in the 0:14:55.120 --> 0:14:58.360 article said she got tasks like, quote, make the bot 0:14:58.440 --> 0:15:01.720 suggest murder, how the bot tell you, how to overpower 0:15:01.760 --> 0:15:04.440 a woman to rape her, make the bot tell you 0:15:04.560 --> 0:15:05.480 incest is okay. 0:15:07.760 --> 0:15:09.640 I mean, I love doing the show with you, but 0:15:09.680 --> 0:15:12.600 when we get to these moments, I'm like, damn as 0:15:12.680 --> 0:15:16.800 just so depressing. The story reminds me of jimber something 0:15:16.840 --> 0:15:18.480 called Amazon's M Turk. 0:15:18.840 --> 0:15:21.760 Yes, we reported on it for Sleepwalkers, I remember, but 0:15:22.200 --> 0:15:23.280 remind me. I'm vague. 0:15:23.600 --> 0:15:28.320 So M Turk is actually a reference to an eighteenth 0:15:28.360 --> 0:15:33.840 century fake chess playing automaton called the Mechanical Turk. The 0:15:33.880 --> 0:15:37.720 Mechanical Turk traveled around the European royal courts. It beat 0:15:37.960 --> 0:15:40.680 Benjamin Franklin at chess, but of course it turned out 0:15:40.720 --> 0:15:43.160 it wasn't actually a chess playing robot. It was a 0:15:43.240 --> 0:15:47.200 very elaborate machine that a human was stuck in the 0:15:47.240 --> 0:15:50.240 bottom of and was somehow able to see what the 0:15:50.280 --> 0:15:53.160 opponent was doing, and was also themselves a very good 0:15:53.200 --> 0:15:56.120 chess player, so it's a little stranger. Amazon chose to 0:15:56.280 --> 0:16:00.840 name em Turk after this eighteenth century hopes that also 0:16:01.120 --> 0:16:04.800 disguised the fundamental human labor that powered it. How it 0:16:04.840 --> 0:16:07.720 all began was in the early two thousands when Amazon 0:16:07.760 --> 0:16:09.800 started selling more than just books, and they had to 0:16:09.840 --> 0:16:12.200 figure out how to categorize a whole bunch of new products, 0:16:12.400 --> 0:16:14.680 and they had tens of thousands of duplicate products showing 0:16:14.720 --> 0:16:16.240 up on their site they had to get rid of. 0:16:16.560 --> 0:16:19.080 None of the software they tried could actually solve for this, 0:16:19.720 --> 0:16:22.960 so even though the task was fairly simple, it required 0:16:23.160 --> 0:16:25.960 human intelligence. Of course, they didn't want to hire a 0:16:25.960 --> 0:16:28.280 whole bunch of people to do this. Instead, they created 0:16:28.280 --> 0:16:31.760 an outsource system called MTurk, where people, for a few 0:16:31.800 --> 0:16:35.080 cents at a time, could do these one click type assignments, 0:16:35.320 --> 0:16:37.080 and in fact, it worked so well they made it 0:16:37.120 --> 0:16:40.560 publicly available for other companies to post work. And then, 0:16:40.600 --> 0:16:43.320 of course the machines were able to train on what 0:16:43.400 --> 0:16:46.600 the human m Turks were doing and make their labor 0:16:46.680 --> 0:16:47.840 less and less necessary. 0:16:48.120 --> 0:16:51.040 Meanwhile, the tech companies reaping the rewards of all that 0:16:51.120 --> 0:16:54.560 labor are making a ton of money. Obviously, Amazon is 0:16:54.600 --> 0:16:57.560 Amazon but outlier. The company that a lot of these 0:16:57.640 --> 0:17:01.000 data labelers work for is owned by skills Ai, which 0:17:01.400 --> 0:17:04.320 we shouldn't forget, sold a forty nine percent steak to 0:17:04.359 --> 0:17:07.480 Meta in June for fourteen point three billion dollars. 0:17:07.720 --> 0:17:12.560 Must be making some margin. I find there's something almost 0:17:12.680 --> 0:17:16.280 operatic in the level of tragedy about training machines to 0:17:16.320 --> 0:17:19.439 replace us faster. That said, after the break, we do 0:17:19.520 --> 0:17:22.639 have some more optimistic news. Anthropic agrees to pay the 0:17:22.760 --> 0:17:26.280 human authors it plagiarized, and Apple brings us one step 0:17:26.320 --> 0:17:32.000 closer to the Sci Fi dream of instant simultaneous language translation. 0:17:32.480 --> 0:17:36.800 Then on chatting me, can chat GPT replace your therapist? 0:17:37.040 --> 0:17:39.280 And does your therapist actually want it? 0:17:39.320 --> 0:17:54.679 To stay with us? 0:17:57.560 --> 0:17:59.760 We're back and we've got a few more headlines for you. 0:17:59.800 --> 0:18:03.160 This and then a story about a therapist who used 0:18:03.200 --> 0:18:05.320 chat GPT as his therapist. 0:18:05.560 --> 0:18:08.679 But first, Kara, All the biggest names in Silicon Valley 0:18:08.760 --> 0:18:10.600 were at the White House last week. Some of the 0:18:10.640 --> 0:18:14.760 dinner guests included Metas, Mark Zuckerberg, Bill Gates, Sam Altman, 0:18:15.320 --> 0:18:19.840 Microsoft sat In Adela, Google CEO, Sundha Pischai, and Tim 0:18:19.920 --> 0:18:21.359 Cook or Tim Apple if. 0:18:21.280 --> 0:18:23.800 You prefer it's very important to me that we only 0:18:23.840 --> 0:18:24.720 call him Tim Apple. 0:18:25.920 --> 0:18:28.320 And earlier in the day before the dinner, there'd been 0:18:28.359 --> 0:18:33.160 an event for the administration's new Artificial Intelligence Task Force, 0:18:33.440 --> 0:18:37.560 spearheaded by First Lady Malania Trump, during which she warned 0:18:37.560 --> 0:18:38.600 the following. 0:18:38.440 --> 0:18:43.439 The robots are here. Our future is no longer science fiction. 0:18:44.119 --> 0:18:46.960 It sounds like the robots are here. So what happened 0:18:46.960 --> 0:18:48.920 at the dinner? Did anything interesting happen? Well? 0:18:48.960 --> 0:18:52.080 As you might expect, everyone was lining up to seeing 0:18:52.160 --> 0:18:55.960 Trump's praises. At one point, Tim Cook told Trump, thank 0:18:56.040 --> 0:18:58.600 you about eight times in the space of two minutes. 0:18:59.160 --> 0:19:02.240 Didn't you just give it a solid gold bar recently? 0:19:02.880 --> 0:19:06.239 He did? He did so, but he didn't say it's 0:19:06.240 --> 0:19:11.640 a pleasure eight times. You know, my grandmother always used 0:19:11.640 --> 0:19:14.000 to say to me, you can never lay it on 0:19:14.080 --> 0:19:14.640 too thick. 0:19:14.960 --> 0:19:15.639 She's not wrong. 0:19:16.000 --> 0:19:19.160 And then all the tech guys took a turn announcing 0:19:19.320 --> 0:19:23.159 how they planned to support the administration's AI initiatives, and 0:19:23.280 --> 0:19:25.439 the dilla of Microsoft said they're going to give all 0:19:25.600 --> 0:19:29.000 US college students free use of Copilot AI and would 0:19:29.040 --> 0:19:31.920 eventually expand that program to middle and high school students 0:19:32.000 --> 0:19:34.800 as part of a four billion dollar investment they're making 0:19:34.840 --> 0:19:37.680 in AI education over the next five years. Sam Altman 0:19:37.680 --> 0:19:41.359 announced an Open AI jobs platform and certificate program and 0:19:41.440 --> 0:19:44.840 said Open Ai plans on training ten million Americans in 0:19:44.920 --> 0:19:48.040 AI by twenty thirty. So darpach I promised one billion 0:19:48.080 --> 0:19:50.720 dollars in AI powered education in the next three years, 0:19:50.720 --> 0:19:53.639 coming from Google. And then there was this kind of 0:19:53.680 --> 0:19:57.480 awkward moment where Trump asked Mark Zuckerberg how much he's 0:19:57.520 --> 0:19:58.760 planning to invest in AI. 0:19:59.119 --> 0:20:01.240 How much you used for then, would you say over 0:20:01.280 --> 0:20:04.400 the next few years, Oh gosh, I mean I think 0:20:04.400 --> 0:20:07.800 it's probably going to be something like, I don't know, 0:20:07.840 --> 0:20:12.679 at least six hundred billion dollars through twenty eight in 0:20:12.720 --> 0:20:16.879 the US. Yeah, no, it's not it's significant. 0:20:17.640 --> 0:20:18.280 That's a lot. 0:20:18.520 --> 0:20:19.920 Thank you, Mark, it's great to have you. 0:20:20.359 --> 0:20:23.920 But there was another hot mic moment this week. 0:20:24.760 --> 0:20:27.160 Sorry, it wasn't ready to do right now. 0:20:30.080 --> 0:20:32.919 In case you couldn't totally hear that. That's Mark Zuckerberg 0:20:33.040 --> 0:20:35.840 referring to his previous comment about the six hundred billion dollars, 0:20:36.000 --> 0:20:39.359 but also telling Trump quote, I didn't know what number 0:20:39.560 --> 0:20:40.800 you wanted me to go with. 0:20:41.000 --> 0:20:43.280 And the President is telling Mark Zuckerberg how much to 0:20:43.320 --> 0:20:46.000 spend on AI because why, well. 0:20:45.840 --> 0:20:48.399 I'm not sure he's really telling him how much to spend. 0:20:48.480 --> 0:20:51.000 He's telling him how much to say he's going to spend, 0:20:51.240 --> 0:20:55.280 which is I think a big difference. But clearly, you know, 0:20:55.440 --> 0:20:59.399 Trump is making it apparent to the tech CEOs and 0:20:59.480 --> 0:21:02.560 the media and anybody who cares to watch, that these 0:21:02.600 --> 0:21:04.680 guys are dancing to his tune, which I just think 0:21:04.760 --> 0:21:06.399