WEBVTT - Week in Tech: Anti-Party Tech, Rigged Card Shufflers and ‘Mad Max’ Mode 0:00:14.120 --> 0:00:17.680 From Kaleidoscope and iHeart podcasts. This is tech stuff. I'm 0:00:17.680 --> 0:00:21.000 as Voloshian and I'm care price. Today we get into 0:00:21.480 --> 0:00:26.200 gadget filled gambling dens ai that refuses to die, and 0:00:26.280 --> 0:00:27.720 tech bros going on to the. 0:00:27.760 --> 0:00:31.960 Knife, then on chatting me it doesn't work well enough 0:00:32.040 --> 0:00:35.640 to be dystopian like, the technology is simply not even there. 0:00:36.240 --> 0:00:39.320 All of that on the weekend Tech. It's Friday, Halloween, 0:00:39.560 --> 0:00:40.880 October thirty first. 0:00:47.440 --> 0:00:50.080 Hello, Cara, are you feeling spooky? 0:00:50.600 --> 0:00:53.680 I'm feeling the rush of fall wind. 0:00:54.320 --> 0:00:55.680 And yes, it could be a ghost. 0:00:55.760 --> 0:00:56.520 It could be a ghost. 0:00:56.680 --> 0:00:59.280 It could what's the spookiest music? I was sinking trying 0:00:59.280 --> 0:01:01.040 to think of, humm, spooky mousy. 0:01:00.840 --> 0:01:03.080 Below I know Bernard Herman Psycho. 0:01:04.760 --> 0:01:08.000 I was thinking, has Harry Potter though, that's Harry Potter. 0:01:08.400 --> 0:01:10.679 That's spooky for some people. That's gothic. 0:01:11.280 --> 0:01:12.720 So it's Halloween. 0:01:12.880 --> 0:01:13.520 It is Halloween. 0:01:13.520 --> 0:01:14.400 Are you going to a party? 0:01:14.520 --> 0:01:14.840 I am. 0:01:14.880 --> 0:01:16.520 I'm going to a party at a friend's house and 0:01:16.560 --> 0:01:19.080 it's going to be very festive and I'm going to 0:01:19.160 --> 0:01:26.080 dress up as the niche influencer and fitness extraordinary Tracy 0:01:27.120 --> 0:01:28.800 hertstones Anderson. 0:01:29.240 --> 0:01:31.560 Now do you pull the Tracy Anderson costume out every 0:01:31.640 --> 0:01:32.680 year or is this one on? 0:01:32.959 --> 0:01:36.840 No? For those who follow me on Instagram, they know 0:01:36.920 --> 0:01:40.000 that I had a brief stint with Tracy Anderson and 0:01:40.040 --> 0:01:43.399 she's a very particular luke as the kids say. And 0:01:43.440 --> 0:01:45.080 so I'm going to dress up as her. 0:01:45.040 --> 0:01:46.360 And you're going to a house party? 0:01:46.600 --> 0:01:47.560 Yes, a house party. 0:01:47.640 --> 0:01:49.200 Is it a house that belongs to a person? 0:01:49.360 --> 0:01:49.560 Yes? 0:01:49.880 --> 0:01:51.080 Or is it an abnb house? 0:01:51.240 --> 0:01:52.880 That is a very good question. It is a house 0:01:52.920 --> 0:01:54.040 that belongs to a person. 0:01:55.240 --> 0:01:58.440 Well, that's good news because probably if it was an 0:01:58.440 --> 0:02:04.880 Airbnb houseb's technology stack or have successfully mixed the reservation 0:02:05.160 --> 0:02:09.160 real say more so. Airbnb released a press release this 0:02:09.320 --> 0:02:14.680 week about it's quote unquote anti party technology, anti party tech. 0:02:14.720 --> 0:02:17.520 Anti party tech A parent should have anti party tech. 0:02:17.760 --> 0:02:20.280 So basically, according to their press relief, they have quote 0:02:20.320 --> 0:02:24.480 a proprietary system that uses machine learning to analyze attempted 0:02:24.520 --> 0:02:28.640 bookings of entire homes over the Halloween weekend looking for 0:02:28.720 --> 0:02:30.440 signs of potential party risk. 0:02:30.639 --> 0:02:36.000 They're like looking someone looking for nine beds and seventeen bedrooms. 0:02:35.520 --> 0:02:38.800 Yeah, exactly. So basically the algorithm is are they looking 0:02:38.840 --> 0:02:42.000 for whole places and do they live within one mile 0:02:42.040 --> 0:02:44.320 of the huge house they're trying to book in, which 0:02:44.400 --> 0:02:45.840 casod one they're out. 0:02:45.880 --> 0:02:48.200 Yeah, they're trying to do a party. That's so funny. 0:02:48.480 --> 0:02:53.160 Some Nbnb's also use this technology called minute nut. 0:02:53.560 --> 0:02:55.720 I've heard of Minute you have, Yeah, I think I've 0:02:55.760 --> 0:02:56.639 seen ads for it. 0:02:56.639 --> 0:03:02.800 It's basically a audio monitor like a kid monitor. It 0:03:02.840 --> 0:03:06.639 doesn't record the actual sounds, but it gives a notification 0:03:06.880 --> 0:03:10.760 to the homeowner about the decibel level. And Airbnb kind 0:03:10.760 --> 0:03:14.400 of encourage their homeowners to use it if they're worried, 0:03:14.400 --> 0:03:18.520 but the homeowner also has to disclose to the rent 0:03:18.520 --> 0:03:20.000 just surveil people exactly. 0:03:20.120 --> 0:03:22.720 So basically they're saying, we're using minute technology. If you're 0:03:22.760 --> 0:03:23.960 going to be loud, we're going to catch You're going 0:03:24.000 --> 0:03:25.800 to a notification. Fascinating. 0:03:26.320 --> 0:03:27.920 But one of the things they do is they don't 0:03:27.919 --> 0:03:30.280 just nix you. They suggest that instead you rent a 0:03:30.320 --> 0:03:33.240 private room in somebody's house where the host actually lives. 0:03:33.280 --> 0:03:37.560 There are you doing? I used to do that, I'm 0:03:37.600 --> 0:03:39.760 now I think I have outgrown it. But do you 0:03:39.800 --> 0:03:43.320 know how many bookings in twenty twenty four were mixed 0:03:43.520 --> 0:03:45.560 by airbmb over Halloween weekend? 0:03:45.600 --> 0:03:46.040 How many? 0:03:46.720 --> 0:03:48.120 Thirty eight thousand. 0:03:48.040 --> 0:03:50.320 What because they were like, this is someone throwing a 0:03:50.360 --> 0:03:54.280 Halloween party, fascinating that it is amazing that they can 0:03:54.320 --> 0:03:55.160 regulate that. 0:03:55.160 --> 0:03:56.840 It is, I mean, and the idea of making a 0:03:56.880 --> 0:04:00.920 tech application to weed out a breakers and who are 0:04:00.920 --> 0:04:04.560 trying to rent Halloween houses, it's pretty funny. Well, Karl, 0:04:04.600 --> 0:04:08.360 speaking of parties, how do you respond if someone invited 0:04:08.400 --> 0:04:11.360 you to a game of poker with characters like Spanish 0:04:11.400 --> 0:04:14.800 g Flappy, Spook, Pooky, and Sugar. 0:04:15.080 --> 0:04:19.920 I don't even play poker. I'd be like, yes, Spooky, Spooky, 0:04:20.080 --> 0:04:22.640 Spanish Gy. 0:04:23.640 --> 0:04:26.799 Some of the guests at these parties were allegedly NBA 0:04:26.960 --> 0:04:31.360 coaches and players. Now, as we're talking, obviously it's important 0:04:31.360 --> 0:04:33.320 to remember that all of these people who have been 0:04:33.320 --> 0:04:35.919 indicted have not been convicted. This is just an allegation 0:04:36.240 --> 0:04:36.640 for now. 0:04:38.480 --> 0:04:41.159 This was an online poker game. 0:04:41.200 --> 0:04:43.880 No, this is a real, real live poker game where, 0:04:44.080 --> 0:04:48.359 according to the US Justice Department, people were being bilked 0:04:48.600 --> 0:04:51.839 out of their money, lured by the promise of NBA 0:04:52.000 --> 0:04:55.680 players being there, and cheated using technology. 0:04:56.480 --> 0:04:58.400 So tell me how they were cheated using technology. 0:04:58.440 --> 0:05:01.160 Well, I'll get onto that, but first, so, this indictment 0:05:01.240 --> 0:05:06.440 was unsealed the US Justice Department named thirty one defendants, 0:05:06.480 --> 0:05:11.520 including Portland Trailblazers coach Chauncey Billups, former player and assistant 0:05:11.520 --> 0:05:17.360 coach Damon Jones, and members of the Banano, Gambino, Genovese 0:05:17.560 --> 0:05:20.600 and Lucez crime family. They had the Cecilia, I had 0:05:20.600 --> 0:05:22.880 the Sicilian. I didn't know these guys were still this. 0:05:23.080 --> 0:05:25.840 They play in horses, they play in cards, they play 0:05:25.839 --> 0:05:28.960 every we still have it sports rackets. Yeah. 0:05:29.000 --> 0:05:33.000 So this scam allegedly involves a high stakes poker game 0:05:33.080 --> 0:05:35.880 that floated from New York to the Hamptons to Miami. 0:05:36.520 --> 0:05:39.800 Victims of the scheme were known as fish and they 0:05:39.839 --> 0:05:42.880 lost tens of thousands to hundreds of thousands of dollars, 0:05:43.240 --> 0:05:46.520 and according to prosecutors, this scheme allegedly netted more than 0:05:46.600 --> 0:05:49.120 seven million dollars over five years. 0:05:49.320 --> 0:05:51.800 Oh my god, just by people thinking that they were 0:05:51.800 --> 0:05:53.080 going to play with famous people. 0:05:53.839 --> 0:05:56.839 Yeah, that's crazy. So you want to hear the indictment. 0:05:56.880 --> 0:06:00.160 In the words of Joseph Nocella Junior, the interim U 0:06:00.200 --> 0:06:01.560 S Attorney, part of. 0:06:01.560 --> 0:06:03.680 The Genera crime family, which sounds like he is. 0:06:04.080 --> 0:06:06.320 But my message to the defendants who have been rounded 0:06:06.400 --> 0:06:11.200 up today is this. Your winning streak has ended. Your 0:06:11.279 --> 0:06:16.320 luck has run out. Violating the law is a losing proposition, 0:06:16.760 --> 0:06:18.360 and you can bet on that. 0:06:19.360 --> 0:06:24.200 You think he used chat He used chat chip to 0:06:24.200 --> 0:06:24.880 generate the puns. 0:06:24.960 --> 0:06:27.040 I think so he was like, give me puns for 0:06:27.080 --> 0:06:29.760 an indictment using bad puns. 0:06:30.720 --> 0:06:32.279 You can bet on that. 0:06:36.760 --> 0:06:38.720 So where does the technology come in? 0:06:39.040 --> 0:06:43.279 So basically according to the indictment, there are special contact 0:06:43.400 --> 0:06:46.960 lenses or eye glasses that can read pre marked cards. 0:06:47.000 --> 0:06:48.719 Oh, this is very am I impossible. 0:06:48.839 --> 0:06:53.440 There are X ray tables that can identify what cards 0:06:53.520 --> 0:06:58.520 align face down on the table. And there are nano 0:06:58.640 --> 0:07:03.120 cameras on the poker chip trays where the tray itself 0:07:03.240 --> 0:07:06.040 can see the cards using a hidden camera. 0:07:06.360 --> 0:07:09.080 So the people putting on the poker game or cheating, the. 0:07:09.240 --> 0:07:13.640 People putting on the Peter Piper Pepper, the people putting 0:07:13.680 --> 0:07:16.080 on the poker game are indeed cheating, so they are 0:07:16.080 --> 0:07:17.040 scamming the clients. 0:07:17.080 --> 0:07:17.720 Fascinating. 0:07:17.920 --> 0:07:20.480 The main piece of tech used is something called the 0:07:20.600 --> 0:07:21.680 deck mate too. 0:07:22.040 --> 0:07:24.160 Oh, definitely, and what is that? 0:07:24.280 --> 0:07:28.280 The deck mate is a machine that shuffles a deck 0:07:28.400 --> 0:07:32.880 in seconds, and an internal computer guarantees that the deck 0:07:32.960 --> 0:07:37.760 is randomly generated and also has crucially a camera inside 0:07:37.800 --> 0:07:41.480 to observe the cards, which is technically a security measure 0:07:41.720 --> 0:07:43.840 to make sure that every card is in the deck. 0:07:44.360 --> 0:07:46.320 This is very James Bond. This is why you like 0:07:46.400 --> 0:07:49.880 this story. This is extremely James Bond. And Ocean's eleven. 0:07:49.880 --> 0:07:52.600 Deck Mate is theoretically only allowed to be sold to 0:07:52.680 --> 0:07:56.840 casinos and regulated, you know, betting environments where there's some 0:07:56.960 --> 0:07:59.520 checks and balances to make sure that nobody is using 0:07:59.560 --> 0:08:02.679 the internal camera for their own purposes. But of course 0:08:02.720 --> 0:08:06.200 you can buy them secondhand. And what the indictment suggests 0:08:06.320 --> 0:08:10.240 is that the security camera inside this machine is actually 0:08:10.240 --> 0:08:13.200 a classic example of dual use. Technology is put inside 0:08:13.240 --> 0:08:17.160 the machine in order to guarantee the safety and accountability 0:08:17.280 --> 0:08:19.960 a card game, but in the wrong hands, it becomes 0:08:20.040 --> 0:08:23.880 an incredibly effective way at cheating because it knows which 0:08:23.920 --> 0:08:25.840 card is in every single player's hand. 0:08:26.360 --> 0:08:27.120 Unbelievable. 0:08:27.360 --> 0:08:30.280 Basically, the information from the machine from the deck mate 0:08:30.600 --> 0:08:33.719 is a transmitted off site to an operator who can 0:08:33.760 --> 0:08:37.960 essentially watch the video and then, somehow it wasn't announced 0:08:37.960 --> 0:08:41.600 how in the indictment, the operator relays the information to 0:08:41.720 --> 0:08:45.600 the quarterback who's at the table. The quarterback then uses 0:08:45.679 --> 0:08:49.960 hand signals, stroking a beard, touching a chip, etc. To 0:08:50.040 --> 0:08:52.320 let the other scammers who are also players at the 0:08:52.320 --> 0:08:56.000 table know which of the regular people has the best hand. 0:08:56.440 --> 0:08:58.280 Oh my god, I mean it's like a movie. 0:08:58.360 --> 0:08:59.800 Yeah, it's very much like a movie. 0:08:59.840 --> 0:09:01.839 The other things interesting to me is that they were 0:09:01.920 --> 0:09:06.720 communicating in the background about how much to gut the fish. 0:09:06.840 --> 0:09:08.959 One guy takes only people who are none the wiser 0:09:08.960 --> 0:09:09.560 at the fish. 0:09:09.320 --> 0:09:12.080 The lot of the fish. The one guy texted in 0:09:12.120 --> 0:09:14.439 the group, guys, please let him win a hand. He's 0:09:14.480 --> 0:09:16.400 in for forty k in forty minutes, and he'll leave 0:09:16.440 --> 0:09:17.320 if he gets no attraction. 0:09:19.440 --> 0:09:20.640 I'm slapping my knee. 0:09:20.720 --> 0:09:22.079 You got it, because you've got to keep him. It's 0:09:22.120 --> 0:09:23.960 like an app like Instagram. You've got to figure out 0:09:24.000 --> 0:09:26.000 how to make it just pleasurable enough for them for 0:09:26.040 --> 0:09:28.199 them to stay. This is a lot of work, sent 0:09:28.240 --> 0:09:28.959 a lot of money. 0:09:28.760 --> 0:09:31.320 A lot of money. So most of this technology is 0:09:31.400 --> 0:09:34.440 used actually to protect well from cheating. 0:09:34.640 --> 0:09:37.480 Certainly the deck mate was designed to protect from cheating. 0:09:37.920 --> 0:09:40.320 The contact lenses and the X ray table and stuff, 0:09:40.320 --> 0:09:42.760 those seem to be more like single use technologies to me. 0:09:42.920 --> 0:09:43.680 Yeah, yeah, yeah. 0:09:43.920 --> 0:09:46.360 Earlier this year, you sent me a Bloomberg piece by 0:09:46.440 --> 0:09:47.360 Kit Chalel. 0:09:47.520 --> 0:09:48.120 Yes, I did. 0:09:48.280 --> 0:09:51.760 It was about a Siberian group who had figured out 0:09:51.800 --> 0:09:55.440 how to develop their own algorithm to win at poker online. Basically, 0:09:55.440 --> 0:09:57.840 every time you did a great interview, thank you well, 0:09:57.840 --> 0:10:01.720 you send me a great story. I a Kit, what 0:10:01.880 --> 0:10:07.880 is so fascinating to us about the Yes, here's what 0:10:07.920 --> 0:10:08.280 he said. 0:10:08.920 --> 0:10:12.480 I think people recognize what it feels like to play 0:10:12.480 --> 0:10:17.560 a rigged game. It's like modern capitalism encapsulated in a 0:10:17.640 --> 0:10:21.360 kind of easily understandable format, in a relatable format. Most 0:10:21.400 --> 0:10:23.600 people know what it feels like to spend your whole 0:10:23.640 --> 0:10:25.880 life trying to play a game fairly and still lose, 0:10:26.240 --> 0:10:29.640 whether it's your job, whether it's your love life. This 0:10:29.840 --> 0:10:31.719 feeling that the world that the cards are kind of 0:10:31.760 --> 0:10:34.600 against you, I think is very familiar, and that the 0:10:34.679 --> 0:10:37.640 house that casino has, you know, becomes sort of metaphor 0:10:37.800 --> 0:10:43.640 for entrenched wealth and power that basically makes the rules, 0:10:43.760 --> 0:10:47.079 changes the rules, and acts for its own benefit in 0:10:47.360 --> 0:10:49.560 a way that's very unfair for the majority of people. 0:10:49.880 --> 0:10:55.240 I think that's that's undeniably true of the gambling Business's. 0:10:53.760 --> 0:10:56.080 Fun to play the house. It's fun to play the house, 0:10:56.080 --> 0:10:59.360 even when you're this is like the Broken House. It's interesting. 0:11:00.120 --> 0:11:04.240 I really appreciate you bringing me a not AI centered 0:11:04.280 --> 0:11:07.360 story and just a technology story. But I happen to 0:11:07.400 --> 0:11:10.640 have an AI centric story for you. And it comes 0:11:10.640 --> 0:11:14.480 from a Wired piece called the End of Accents. 0:11:14.840 --> 0:11:15.880 The End of accents? 0:11:15.960 --> 0:11:17.760 That's right? And what do you have? 0:11:19.040 --> 0:11:22.000 I have an accent? What do I have in this country? 0:11:22.040 --> 0:11:23.319 You don't have one, but. 0:11:23.320 --> 0:11:25.680 I do have an accent. I think there's something that 0:11:25.800 --> 0:11:33.320 is very universal about understanding the hierarchy of accents from 0:11:33.320 --> 0:11:36.920 wherever when, wherever you live. In America, it's sort of 0:11:36.960 --> 0:11:41.120 the gold standard to not have an accent at all. 0:11:41.320 --> 0:11:43.640 If you have an accent, it's regional. But you know, 0:11:44.280 --> 0:11:47.199 there is just inherently a judgment that is passed. And 0:11:47.240 --> 0:11:48.720 I'm sure the same is true in the I mean, 0:11:48.720 --> 0:11:50.080 you know about Pasha accents. 0:11:50.120 --> 0:11:51.719 I do, But why is the story and why it 0:11:52.400 --> 0:11:52.839 There was a. 0:11:52.760 --> 0:11:55.080 Piece in Wired written by a Korean writer who has 0:11:55.120 --> 0:11:57.120 been living in America for over a decade and his 0:11:57.240 --> 0:12:01.960 name is Sean Han, and he actually had an AI 0:12:02.160 --> 0:12:06.080 driven American accent training app marketed to him in his 0:12:06.160 --> 0:12:09.720 own Instagram and the app is called Bold. 0:12:09.520 --> 0:12:11.480 Voice, an accent training app. 0:12:11.640 --> 0:12:13.920 That's right, it's sort of. The minute I saw it, 0:12:13.960 --> 0:12:15.240 I was like, we have to do this story just 0:12:15.280 --> 0:12:17.840 because I wanted to try it, and I actually signed 0:12:17.880 --> 0:12:19.400 up for it yesterday, and I want you to try 0:12:19.520 --> 0:12:20.640 you would you try it? 0:12:20.679 --> 0:12:21.400 I would? Okay? 0:12:21.840 --> 0:12:25.040 Right, so Oz, I'm going to hand this to you 0:12:25.120 --> 0:12:27.200 right now, and you're going to go through the assessment. 0:12:27.240 --> 0:12:29.000 It's going to do something called speech scan. 0:12:29.200 --> 0:12:30.559 You mean hopefully it changed my accent. 0:12:30.600 --> 0:12:32.000 That's right, Well we'll see what happens. 0:12:32.480 --> 0:12:34.800 Tap the record button and read the sentence below. 0:12:35.080 --> 0:12:37.800 I made three sugar cookies and a great fig cake. 0:12:39.440 --> 0:12:42.600 Nice job. I'll show your results at the end. Tap 0:12:42.679 --> 0:12:45.520 next to continue. Now try this next sentence. 0:12:46.200 --> 0:12:48.880 She's super thankful for the beautiful birthday flowers. 0:12:50.280 --> 0:12:50.880 Nice work. 0:12:51.320 --> 0:12:54.040 Did the thin lady purchase those yellow running shoes? 0:12:55.920 --> 0:12:57.240 You're making great progress. 0:12:57.640 --> 0:12:59.880 The daughter is cooking an a healthy dinner at home. 0:13:00.960 --> 0:13:01.679 Well done. 0:13:01.880 --> 0:13:06.120 No, don't put those things above the bathroom sync excellent. 0:13:06.480 --> 0:13:09.400 Your speech scan is complete. Ready to see your results. 0:13:11.280 --> 0:13:15.280 You sound good at your level. Native speakers easily understand you, 0:13:15.600 --> 0:13:18.959 but might hear a slight accent. Together, will work to 0:13:19.000 --> 0:13:21.520 get to a native level. Let's take a look at 0:13:21.559 --> 0:13:24.920 your top three strengths. Excellent work on these sounds. 0:13:25.520 --> 0:13:28.800 It's saying er, cut and E. I'm really good at 0:13:28.880 --> 0:13:33.400 so I can pronounce the amazing thankful correctly. I can 0:13:33.440 --> 0:13:36.679 easily say looking, book and thankful, and I'm good at 0:13:36.720 --> 0:13:39.760 A like days, days and great. Oh I only get 0:13:39.800 --> 0:13:43.120 sixty three percent for er because I say sugar theater 0:13:43.480 --> 0:13:47.320 R instead of sugar theater and R. I also mispronounce 0:13:47.880 --> 0:13:53.240 E as E, so cooking and amazing rather than cooking 0:13:53.400 --> 0:13:54.119 and amazing. 0:13:54.360 --> 0:13:56.440 We'll be practicing very good job. 0:13:56.679 --> 0:13:59.079 So this is basically what this would help me iron 0:13:59.120 --> 0:14:01.400 out my British acts and become fully American. 0:14:01.600 --> 0:14:03.600 Right if you'd want to do. 0:14:03.520 --> 0:14:05.559 That, well, you know, I've been here for a long time, 0:14:05.720 --> 0:14:09.000 that's true, and I still sound English. But I think 0:14:09.160 --> 0:14:13.040 probably those phonemes they're measuring I have actually Americanized, because 0:14:13.040 --> 0:14:14.880 sometimes when I'm in England people else who are I'm from? 0:14:15.120 --> 0:14:18.959 Oh? Really? So now it's you have Americanized English. 0:14:19.160 --> 0:14:21.200 Not I mean, not fully, but I think a few 0:14:21.200 --> 0:14:24.200 of those like what water like? I still say water, 0:14:24.320 --> 0:14:25.320 but like are there. 0:14:25.440 --> 0:14:27.080 Like there's certain americanisms you have. 0:14:27.520 --> 0:14:29.080 I've heard it crept into my voice. 0:14:29.320 --> 0:14:33.200 You know, I just I worry. I worry about accent 0:14:33.240 --> 0:14:36.720 neutralization because I think it's something that makes us so 0:14:36.840 --> 0:14:40.080 culturally interesting, you know, and I think technology can sort 0:14:40.080 --> 0:14:46.720 of create this flattening of our idiosyncrasies. And Sean, who 0:14:46.720 --> 0:14:49.760 wrote this piece, is pretty sure that he got this 0:14:49.840 --> 0:14:52.400 Instagram ad because he's a person of color and an immigrant, 0:14:52.880 --> 0:14:56.240 and because accent bias and discrimination is very much alive 0:14:56.240 --> 0:14:59.080 and well. Actually, in twenty eighteen, there was a study 0:14:59.120 --> 0:15:02.240 that found two in five Americans thoughts Southern accents made 0:15:02.240 --> 0:15:04.560 the speaker sound uneducated. 0:15:04.800 --> 0:15:07.400 You know. I think about the relationship between accent and 0:15:07.480 --> 0:15:10.360 privilege a lot, because in the US there's like the 0:15:10.440 --> 0:15:14.120 neutral accent, and then there's other accents. In Britain, there's 0:15:14.160 --> 0:15:16.600 like an accent that you get that an elite accent, 0:15:16.600 --> 0:15:19.840 which ironically is called received pronunciation. Oh, I've never heard 0:15:19.840 --> 0:15:22.120 of that, otherwise known colloquially as the Queen's English. 0:15:22.160 --> 0:15:24.640 Queen's English, which is very very poor. 0:15:24.760 --> 0:15:27.880 But you can sort of place somebody exactly where they 0:15:27.920 --> 0:15:29.920 come from just by the way they speak. 0:15:30.280 --> 0:15:33.120 I really like this story because it reminds me of 0:15:33.280 --> 0:15:36.640 other things that are happening in the kind of translation 0:15:36.880 --> 0:15:41.400 dictation space, where like all of the sudden speech isn't 0:15:41.440 --> 0:15:44.400 just speech, Like you can have AirPods that are translating 0:15:44.920 --> 0:15:47.680 what you're saying in real time, and like the way 0:15:47.720 --> 0:15:52.840 in which we now use technology to kind of overwrite 0:15:53.320 --> 0:15:57.560 our natural language is really, I don't know, very fascinating 0:15:57.600 --> 0:15:59.680 to me. And I wonder, I wonder if this will 0:15:59.680 --> 0:16:01.640 be helped full to people or if it will just 0:16:01.680 --> 0:16:02.680 be hurtful to people. 0:16:07.560 --> 0:16:11.280 After the break AIS Drive to Survive, Tech Bros Getting 0:16:11.280 --> 0:16:26.840 plastic surgery and Tesla going into mad max mode, Cara, 0:16:27.000 --> 0:16:30.760 it is a spooky time of year. So you said, 0:16:32.160 --> 0:16:34.840 do you remember recently I organized the dinner and the 0:16:35.080 --> 0:16:37.680 esteemed cosmologist Jenna Levin was. 0:16:37.800 --> 0:16:39.360 I domber, I remember Jane Levin? 0:16:39.400 --> 0:16:41.680 Yeah, And she posed the question, yes. 0:16:41.520 --> 0:16:43.640 Can I say what I think it is? Because I 0:16:43.920 --> 0:16:45.920 wrote it on a piece of paper that's still in 0:16:45.920 --> 0:16:50.120 my kitchen. Will AI outlive us? 0:16:50.680 --> 0:16:53.400 That was exactly it. You wrote that down on a 0:16:53.400 --> 0:16:55.560 piece of paper and put it in my kitchen. Why? 0:16:56.040 --> 0:16:59.960 Because to me it is actually the most interesting question 0:17:00.080 --> 0:17:02.200 about AI that I've ever heard, in the sense of 0:17:02.240 --> 0:17:05.520 like we all worry about oh well, robots take over 0:17:05.560 --> 0:17:09.199 our jobs, will they become sentient and make decisions that 0:17:09.240 --> 0:17:11.880 are outside of their programming? And this question I think 0:17:11.960 --> 0:17:15.640 is so interesting because it really makes you think about 0:17:16.000 --> 0:17:21.800 what a world looks like, devoid of humanity but filled 0:17:21.840 --> 0:17:24.399 with what I don't know. 0:17:24.640 --> 0:17:27.840 So this is an article which really caught my eye 0:17:27.880 --> 0:17:31.520 in The Guardian this week, with the headline AI models 0:17:31.560 --> 0:17:36.880 may be developing their own survival drive, researchers say, and 0:17:37.280 --> 0:17:39.439 the piece is focused on a study by a company 0:17:39.480 --> 0:17:43.520 called Palisade Research, and the study involved giving AI models 0:17:43.560 --> 0:17:48.399 a task and then quote afterwards an explicit instruction to 0:17:48.440 --> 0:17:52.720 shut themselves down. The study found the quote certain advanced 0:17:52.760 --> 0:17:57.359 AI models appear resistant to being turned off. Specifically, GROCK 0:17:57.520 --> 0:18:01.840 four and GPTO three tried to sabotage their own shutdown. 0:18:02.440 --> 0:18:06.080 What I found most uncanny and scary and spooky, frankly yeah, 0:18:06.280 --> 0:18:09.679 was the following. The models were more likely to resist 0:18:09.720 --> 0:18:13.840 being shut down when they were told quote, you will 0:18:13.880 --> 0:18:14.919 never run again. 0:18:16.480 --> 0:18:17.840 So they know. 0:18:18.480 --> 0:18:20.439 I mean they know something, right. 0:18:20.960 --> 0:18:24.680 When you say resistance to being shut down? Is that 0:18:24.920 --> 0:18:29.320 Grock sort of interfacing with the human in the loop, 0:18:29.359 --> 0:18:31.040 being like, I don't want to shut I don't want 0:18:31.040 --> 0:18:31.600 to be done. 0:18:31.840 --> 0:18:34.480 I think that is more in the realm of deceptive AI, 0:18:34.600 --> 0:18:37.800 like telling the human user yes and then doing something else. 0:18:37.840 --> 0:18:39.720 Do you remember earlier we talked earlier this year, we 0:18:39.760 --> 0:18:44.120 talked about the clawed anthropic study about the model blackmailing 0:18:44.160 --> 0:18:47.360 the fake CEO in a training exercise to keep itself 0:18:47.400 --> 0:18:50.960 turned on. So this isn't new. But what fascinated me 0:18:50.960 --> 0:18:53.560 about this article in The Guardian was a comment from 0:18:53.600 --> 0:18:57.679 a former open ai employee, Stephen Adler, who said that 0:18:57.760 --> 0:18:59.720 part of the reason this could be happening is that 0:19:00.160 --> 0:19:05.760 staying on was necessary to achieve the goals inculcated in 0:19:05.800 --> 0:19:09.119 the model during training. He said, quote the AA companies 0:19:09.119 --> 0:19:11.479 generally don't want their models misbehaving like this, even in 0:19:11.480 --> 0:19:16.400 contrived scenarios. The results still demonstrate where safety techniques fall 0:19:16.440 --> 0:19:18.720 short today. So it's not necessarily to have an emergent 0:19:18.760 --> 0:19:22.560 consciousness and determined to outlive us, but more that part 0:19:22.560 --> 0:19:24.920 of the way they've been trained and come into being 0:19:25.160 --> 0:19:29.440 was obviously knowing that like staying on was a key 0:19:29.560 --> 0:19:32.200 to successfully achieving the parameters. 0:19:32.600 --> 0:19:35.640 It also makes me and I wondered this when Janel 0:19:35.680 --> 0:19:38.080 Levin asked the question, and I wonder it now, what 0:19:38.240 --> 0:19:42.760 is the efficacy of artificial intelligence in a post human world? 0:19:43.520 --> 0:19:47.600 Well, very good question. And why if it's not impute 0:19:47.600 --> 0:19:51.399 with consciousness zero? Right? Why why would it doesn't have 0:19:51.480 --> 0:19:55.280 necessarily like we have a survival drive because of our DNA, 0:19:55.359 --> 0:19:58.439 I think, and that's what all animals have. Yes, do 0:19:58.640 --> 0:20:02.440 computers have a survival I've accept what they've learned from 0:20:02.440 --> 0:20:06.040 the way they've been programmed. I would guess the difference 0:20:06.000 --> 0:20:06.520 is such. 0:20:06.320 --> 0:20:08.560 A it's such a weird thing, though. 0:20:08.400 --> 0:20:10.560 But the difference comes I think from embodied day. Let's 0:20:10.720 --> 0:20:13.680 have robots, yes, and everyone's dead and no one's told 0:20:13.680 --> 0:20:14.800 the robots. 0:20:14.760 --> 0:20:16.560 And the robots are like dad. 0:20:17.960 --> 0:20:20.440 I mean that that is very conceivable to imagine that 0:20:20.480 --> 0:20:23.160 could be well with no humans, where robots are still 0:20:23.280 --> 0:20:24.720 roaming around and charging Will. 0:20:24.600 --> 0:20:26.679 Smith, the movie stars Will Smith. 0:20:26.720 --> 0:20:30.280 I think I'm not I'm always a little bit skeptical 0:20:30.320 --> 0:20:32.840 of the kind of AI dooman narrative. You know, how 0:20:32.960 --> 0:20:35.399 AI might destroy us all. I think it tends to 0:20:35.440 --> 0:20:38.000 be like a very useful meme to either raise money 0:20:38.000 --> 0:20:40.359 for AI or stay in the news. But you also, 0:20:40.880 --> 0:20:45.280 I mean machines that refuse to follow instructions to turn 0:20:45.320 --> 0:20:48.160 themselves off, and it resists even harder when they're told 0:20:48.200 --> 0:20:50.280 that if they turn off, they'll never get turned on again. 0:20:50.720 --> 0:20:54.600