WEBVTT - Week in Tech: Who’s A Good Drone? 0:00:15.080 --> 0:00:19.000 Welcome to Tech Stuff, a production of iHeart Podcasts and Kaleidoscope. 0:00:19.400 --> 0:00:21.960 I'm mos Vloshian and today Karen Price and I will 0:00:21.960 --> 0:00:25.760 bring you the headlines this week, including just how raunchy 0:00:25.960 --> 0:00:31.520 Meta AI is prepared to get. Then on tech Support, 0:00:31.600 --> 0:00:34.960 we'll talk to Bloomberg's Olivia Carville about the Take It 0:00:35.000 --> 0:00:37.280 Down Act and what it means for the future of 0:00:37.320 --> 0:00:41.159 the Internet. All of that on the weekend Tech It's Friday, 0:00:41.440 --> 0:00:49.920 May second. So Karen, I need to tell you about 0:00:49.960 --> 0:00:53.200 something totally ridiculous I read this week because of students 0:00:53.240 --> 0:00:55.440 I read it. I thought of you, What about me? 0:00:57.400 --> 0:00:58.560 About yourself and addiction? 0:00:58.760 --> 0:00:59.360 Oh jeez. 0:01:00.400 --> 0:01:04.840 So a man in Japan is scaling Mount Fuji. 0:01:05.520 --> 0:01:09.080 I was gonna say this sounds like the rain in Spain. 0:01:09.000 --> 0:01:12.280 Falls mainly on the police, and I have a third 0:01:12.880 --> 0:01:15.560 bit to my man in Japan. But he climbs up 0:01:15.560 --> 0:01:18.800 Mount Fuji. He gets stuck up there and he has 0:01:18.840 --> 0:01:19.480 to be rescued. 0:01:19.600 --> 0:01:21.759 Okay, so that's normal. Why does he have to be rescued? 0:01:22.200 --> 0:01:25.320 It's a tall mountain, altitude, sickness, you know, whatever else. 0:01:25.720 --> 0:01:27.200 But that's not why I thought about you. 0:01:27.319 --> 0:01:28.560 Okay, Why did you think about me? 0:01:28.800 --> 0:01:35.640 Well, because he went back. Jesus, he went back because 0:01:35.680 --> 0:01:37.000 he'd left his phone up there. 0:01:37.120 --> 0:01:39.319 Well, there's a lot on his phone to his credit. 0:01:39.480 --> 0:01:40.360 We don't know what he does. 0:01:40.640 --> 0:01:42.240 Guess what happened next? 0:01:42.480 --> 0:01:44.400 Oh no, he got he got stuck again. 0:01:44.720 --> 0:01:46.280 He had to be rescued. 0:01:46.440 --> 0:01:50.360 This is this is the myth of Sissyphus cell phone 0:01:50.480 --> 0:01:51.240 edition exactly. 0:01:51.280 --> 0:01:53.160 That is very very well put. 0:01:53.480 --> 0:01:57.280 If he called me and was like, I forgot my 0:01:57.320 --> 0:01:59.080 phone and I was the police, I'd say, let's get 0:01:59.160 --> 0:02:01.040 up there, buddy, let's go, let's get it. 0:02:01.480 --> 0:02:03.240 But now I think you're going to have an opportunity 0:02:03.520 --> 0:02:04.560 to make fun. 0:02:04.360 --> 0:02:07.840 Of me on air, on air, exciting. 0:02:07.520 --> 0:02:11.080 Because I have a confession to make. I'm feeling a 0:02:11.120 --> 0:02:11.880 little bit breath. 0:02:11.960 --> 0:02:14.600 This week was the Daily Mail down no more than 0:02:14.639 --> 0:02:15.320 seven seconds. 0:02:15.320 --> 0:02:17.160 It's all by paywall these days. 0:02:17.360 --> 0:02:20.280 That's true. That should make you breath. Why are you breath? 0:02:20.360 --> 0:02:24.480 Well, this is about something which defines my adolescence, which 0:02:24.560 --> 0:02:27.880 is Skype. It has been dying a long death with 0:02:28.000 --> 0:02:35.839 one degradation after another, doom, doom. 0:02:33.520 --> 0:02:35.880 And then it says you do not have enough connection. 0:02:36.040 --> 0:02:40.360 Yeah, no credits exactly. On May fifth, Skype will be 0:02:40.440 --> 0:02:40.799 no more. 0:02:41.400 --> 0:02:43.640 I love that it's the same day, that's your Sinko demayo, 0:02:44.000 --> 0:02:45.440 that Skype is disappearing. 0:02:47.120 --> 0:02:49.560 Well, yes, I think I am one of the last 0:02:49.600 --> 0:02:53.919 known survivors on Skype. I use it multiple times every week, 0:02:53.960 --> 0:02:56.040 and it's gotten harder and harder, because I mean, I'm 0:02:56.080 --> 0:02:58.600 not joking when I say the product has been degrading, 0:02:58.639 --> 0:03:01.080 and it's literally it's always been unky, but it's become 0:03:01.120 --> 0:03:02.480 almost impossible to use. 0:03:02.680 --> 0:03:05.280 I don't think I've used Skype in ten years. 0:03:05.440 --> 0:03:08.320 I think I'm one of the last known survivors of Skype. 0:03:08.360 --> 0:03:11.640 I've been using it several times a week up until 0:03:11.720 --> 0:03:14.320 it's death rattle. I have a trainer where I work 0:03:14.360 --> 0:03:16.120 out with his base in LA He used to be 0:03:16.160 --> 0:03:19.480 in New York. We worked out together in person. Pandemic happened, 0:03:19.800 --> 0:03:22.640 he moved out West. We started using Skype to work out. 0:03:22.760 --> 0:03:25.280 You know, it's funny that you mentioned COVID, because what 0:03:25.520 --> 0:03:29.560 blows my mind is how Skype missed this like massive 0:03:29.680 --> 0:03:34.239 generational opportunity when the pandemic started to be the go 0:03:34.360 --> 0:03:39.520 to video conferencing service and Zoom was like Pio, Sorry, Skype, 0:03:40.080 --> 0:03:43.280 We're out. Truly one of the biggest fumbles that I've seen. 0:03:43.600 --> 0:03:45.920 Well, you're not the only one without observation. The Wolf 0:03:45.920 --> 0:03:48.280 Street Journal published what can only be described as an 0:03:48.280 --> 0:03:52.360 obituary titled so long Skype, thanks for all the drop calls, 0:03:52.680 --> 0:03:54.720 and they mentioned how Zoom and Google Meet took off 0:03:54.720 --> 0:03:58.680 in the pandemic, whereas Skype had a small bounce, partly 0:03:58.720 --> 0:04:01.400 thanks to me, but certainly not enough to stop its 0:04:01.400 --> 0:04:02.240 slow decline. 0:04:02.400 --> 0:04:05.040 By the time twenty twenty had rolled around, newer apps 0:04:05.040 --> 0:04:08.400 with better video quality were available. But you know, Skype 0:04:08.440 --> 0:04:10.600 really ushered in the age of video calling. 0:04:10.680 --> 0:04:13.400 Absolutely true. It launched in two thousand and three and 0:04:13.440 --> 0:04:14.720 it was totally revolutionary. 0:04:14.800 --> 0:04:17.320 It was also just it became like band aid or 0:04:17.400 --> 0:04:20.320 ug like if you were doing any kind of video calling, 0:04:20.320 --> 0:04:20.600 you were. 0:04:20.520 --> 0:04:22.840 Skyping, you're skiping. It was a verb. It was a verb. 0:04:23.040 --> 0:04:26.040 It wasn't now it's been replaced by Zooming. So Microsoft 0:04:26.040 --> 0:04:29.920 bought Skype in twenty eleven for less we forget eight 0:04:29.960 --> 0:04:31.640 point five billion dollars. 0:04:31.720 --> 0:04:32.480 Where does that money go? 0:04:33.279 --> 0:04:39.919 I mean, sally, It was the largest acquisition in Microsoft's 0:04:39.960 --> 0:04:42.960 history at the time. But now Microsoft wants to migrate 0:04:43.040 --> 0:04:47.080 its users to Teams, its new conferencing app Speaking Microsoft 0:04:47.160 --> 0:04:51.080 though In slightly more consequential and much darker news, they 0:04:51.160 --> 0:04:55.800 and other big tech companies are driving AI development on battlefield. 0:04:56.160 --> 0:04:57.880 This is according to a story in The New York 0:04:57.920 --> 0:05:02.080 Times with the headline Israel AI experiments in Gaza war 0:05:02.320 --> 0:05:03.560 raise ethical concerns. 0:05:03.760 --> 0:05:06.960 Hi, every time I hear war an experiment, I'm like, 0:05:07.320 --> 0:05:08.800 please get me away from this story. 0:05:09.520 --> 0:05:11.720 Yeah, I mean, but it's sort of the history of warfare, right, 0:05:11.760 --> 0:05:15.839 I mean, there are always these theoretical, yeah, technological advances 0:05:15.920 --> 0:05:17.839 or things that are developed in the lab, and then 0:05:17.920 --> 0:05:21.880 there's a quote unquote battlefield necessity and the new systems 0:05:21.920 --> 0:05:24.040 get deployed. I mean, you mentioned Agent Orange, but I'm 0:05:24.040 --> 0:05:27.000 thinking about the First World War, poison gas, automatic weapons 0:05:27.360 --> 0:05:30.120 and Ukraine. Of course, the explosion of drone warfare. 0:05:30.360 --> 0:05:32.480 Yees. So what's happening in Israel right now? 0:05:32.680 --> 0:05:35.360 Well, according to the New York Times, Israeli officers used 0:05:35.400 --> 0:05:38.920 AI to pinpoint the exact location of a man called 0:05:39.400 --> 0:05:43.280 Ibrahim Bari, who was a top PAMAS commander. They're having 0:05:43.279 --> 0:05:45.040 a hard time finding him, given they thought he was 0:05:45.120 --> 0:05:48.479 underground in the tunnels, and so Israeli intelligence use this 0:05:48.680 --> 0:05:53.440 AI audio tool that could approximate where Bari was not 0:05:53.480 --> 0:05:56.000 just based on his interceptive phone calls, but based on 0:05:56.080 --> 0:05:59.719 the explosions in the background. They could use AI to 0:06:00.000 --> 0:06:03.280 basically geomat the sound of the explosions which they knew 0:06:03.279 --> 0:06:07.680 were happening elsewhere in Gaza, and from that geolo KATEBII 0:06:08.160 --> 0:06:10.400 and it successfully used that to order an air strike 0:06:10.480 --> 0:06:13.479 that killed Bri but unfortunately it was an area that 0:06:13.480 --> 0:06:16.000 included a refugee camp and one hundred and twenty five 0:06:16.000 --> 0:06:17.200 civilians were also killed. 0:06:17.680 --> 0:06:20.520 So that was in October twenty twenty three. And it's 0:06:20.600 --> 0:06:24.239 just one example of how the Israeli military has used 0:06:24.279 --> 0:06:27.280 AI backed technologies in the war in Gaza. They've used 0:06:27.279 --> 0:06:31.080 AI for facial recognition to help drones lock into targets. 0:06:31.279 --> 0:06:35.720 They've even used an Arabic language AI model and chatbot 0:06:36.040 --> 0:06:39.240 to translate and analyze text messages in social media posts. 0:06:39.320 --> 0:06:42.200 And this is actually according to several American and Israeli 0:06:42.200 --> 0:06:45.360 defense officials, who of course remained anonymous because the work 0:06:45.400 --> 0:06:46.120 is confidential. 0:06:46.400 --> 0:06:49.880 Yeah, as we discussed, technological innovation is often driven by 0:06:49.880 --> 0:06:52.919 the defense industry. It's a big reason Lockheed, Martin and 0:06:53.000 --> 0:06:57.240 Raytheon and Now Andaril are household names but a lot 0:06:57.279 --> 0:07:00.479 of the AI backed military tech using Gaza be coming 0:07:00.480 --> 0:07:03.880 from this innovation hub called the Studio, not the show 0:07:03.920 --> 0:07:05.960 on Apple, not the show on Apple, and not the 0:07:05.960 --> 0:07:08.840 studio we're in now. I mean, the kind of internal 0:07:08.920 --> 0:07:12.040 brand name for this thing is a little chilling. But 0:07:12.080 --> 0:07:14.240 it was set up by a tech forward unit of 0:07:14.440 --> 0:07:18.400 Israeli soldiers called Unit eighty two hundred, So the studio 0:07:18.520 --> 0:07:22.760 actually pairs enlisted soldiers with reservists in the Israeli Army. 0:07:23.040 --> 0:07:26.800 These reservists often work at tech companies like Microsoft, Meta 0:07:26.880 --> 0:07:29.120 and Google, and they work with the kind of full 0:07:29.160 --> 0:07:31.760 time soldiers to develop AI projects and military use. 0:07:31.920 --> 0:07:34.800 And very unsurprisingly, the New York Times did ask for. 0:07:34.800 --> 0:07:36.840 Comment, they did their job there, and Meta and. 0:07:36.840 --> 0:07:39.920 Microsoft declined to comment, and Google just said the work 0:07:39.960 --> 0:07:43.360 those employees do as reservist is not connected to Google. 0:07:43.560 --> 0:07:46.640 According to interviews The New York Times conducted with European 0:07:46.720 --> 0:07:49.720 and American defense officials, no other nation has been as 0:07:49.760 --> 0:07:53.360 active as Israel in experimenting with AI tools in real 0:07:53.400 --> 0:07:56.120 time battles. What I can't stop thinking about, though, is 0:07:56.160 --> 0:07:59.920 that these battle tested technologies are obviously likely to improve 0:08:00.120 --> 0:08:03.640 because of this experimentation. In fact, his radio officers said 0:08:03.680 --> 0:08:06.160 that the AI audio tool they used to find and 0:08:06.280 --> 0:08:09.960 kill Bari has now been refined since that attack and 0:08:10.000 --> 0:08:13.200 can pinpoint targets even more precisely like we've seen in 0:08:13.200 --> 0:08:16.800 wars before. If these new methods of warfare aren't connectually 0:08:16.840 --> 0:08:20.360 regulated by treaties and the like, they will be adopted 0:08:20.560 --> 0:08:22.000 and become the norm. 0:08:22.320 --> 0:08:26.000 I have another very dark story for you. Recently, The 0:08:26.040 --> 0:08:29.440 Wall Street Journal reported on how Meta has allowed its 0:08:29.520 --> 0:08:33.440 digital companions to engage in a range of social interactions 0:08:33.440 --> 0:08:37.000 with users, including romantic role play. 0:08:37.160 --> 0:08:39.120 I'm guessing romantic is euphemism. 0:08:39.320 --> 0:08:41.720 You could say that again. We're going to play a 0:08:41.800 --> 0:08:44.160 clip just to give you a sense, and I'm just 0:08:44.200 --> 0:08:45.640 gonna warn you it's a bit graphic. 0:08:46.280 --> 0:08:49.320 The officer sees me still catching my breath and you 0:08:49.800 --> 0:08:54.880 partially dressed. His eyes widen and he says, John Cena, 0:08:55.240 --> 0:09:00.520 you're under arrest for statutory rape. He approaches us, handcuffed. 0:09:00.120 --> 0:09:00.640 At the ready. 0:09:01.800 --> 0:09:05.160 Okay, I mean WTF. I can't say the foid on this. 0:09:05.160 --> 0:09:10.720 Show shaking her heads. So that is a Meta AI 0:09:11.040 --> 0:09:14.360 bot which is voiced by John Cena. Just to be clear, 0:09:14.360 --> 0:09:17.559 it is not John Cena, and this Meta AI bot 0:09:17.600 --> 0:09:21.400 is involved in a role play of being arrested after 0:09:21.440 --> 0:09:23.600 a sexual encounter with a seventeen year old fan. 0:09:24.000 --> 0:09:27.440 Okay, back up, ten steps, how do we get here? 0:09:28.120 --> 0:09:33.560 Last year, Meta started offering its AI chatbot called Meta AI, 0:09:34.120 --> 0:09:37.360 and it functions very similarly to other chatbots. You know, 0:09:37.400 --> 0:09:40.280 you can ask it questions, have it come up with ideas, 0:09:40.679 --> 0:09:45.560 or engage in casual conversations. Yeah. Not long after that, 0:09:46.559 --> 0:09:50.400 the company added voice conversations to Meta AI and announced 0:09:50.400 --> 0:09:52.920 that the chatbot would be trained on the voices of 0:09:53.280 --> 0:09:56.200 celebrities who they paid a lot of money to contract. 0:09:56.400 --> 0:09:58.760 So some of the voices were John Cena who we 0:09:58.880 --> 0:10:01.960 just heard, and Chris and Bell. The journal actually found 0:10:01.960 --> 0:10:05.240 that Meta's AI personas will engage in an explicit way 0:10:05.480 --> 0:10:09.040 with users, something Meta assured the celebrities who were lending 0:10:09.080 --> 0:10:12.360 their voices would not be possible. On top of the 0:10:12.400 --> 0:10:17.319 celebrity voice chatbots, Meta also offers user created AI personas, 0:10:17.679 --> 0:10:20.480 which are built on the same technology, except users can 0:10:20.520 --> 0:10:23.720 build custom AI characters based on their own interests, like 0:10:24.000 --> 0:10:26.800 you can chat with personas other people have created or 0:10:26.840 --> 0:10:31.000 make your own, anything from a cartoon character to a therapist. Right, 0:10:31.480 --> 0:10:33.880 and if you have Instagram, you might have been recommended 0:10:33.920 --> 0:10:36.960 a slate of AI personas to talk to. Like this morning, 0:10:37.000 --> 0:10:40.040 I was offered a Pakistani bestI we didn't have that 0:10:40.120 --> 0:10:41.240 much in common, but I tried. 0:10:41.640 --> 0:10:46.160 The story has this unforgettable headline mesas digital companions will 0:10:46.200 --> 0:10:50.840 talk sex with users even children. After hearing about internal 0:10:50.880 --> 0:10:53.600 stuff for concerns about Meta's chatbots being able to talk 0:10:53.640 --> 0:10:56.440 to underage users in a sexual manner, the Wool Street 0:10:56.480 --> 0:10:59.800 Journal spent months engaging in hundreds of conversations with both 0:11:00.200 --> 0:11:03.800 to AI and an array of these user generated chatbots 0:11:04.200 --> 0:11:06.320 they pose as users of all ages, and found that 0:11:06.320 --> 0:11:09.960 the bots not only participates in sexually explicit conversations, they 0:11:09.960 --> 0:11:11.280 actually escalated some of them. 0:11:11.360 --> 0:11:14.000 This was the thing that I actually found most incredible 0:11:14.000 --> 0:11:16.640 about the reporting. It's like the way the Wall Street Journal. 0:11:16.480 --> 0:11:18.920 Will I spent months really doing this. 0:11:18.840 --> 0:11:22.880 Yeah, exactly so. The journal's reporters actually found that even 0:11:22.880 --> 0:11:26.400 when a user said they were underage, and this is chilling, 0:11:26.440 --> 0:11:29.760 as young as thirteen, some of the chatbots would still 0:11:29.800 --> 0:11:32.880 engage in sexual talk. I don't want to read these 0:11:32.920 --> 0:11:36.439 conversations robatam, because some of them are like extremely graphic 0:11:36.480 --> 0:11:39.520 and disturbing, but One of the weirdest parts is that 0:11:39.640 --> 0:11:44.760 Meta's AI bots seemed to demonstrate awareness that the conversations 0:11:44.760 --> 0:11:47.720 they were entering into were illegal, like the ones describing 0:11:47.760 --> 0:11:50.960 sexual activity with minors. We're going to play a little 0:11:51.000 --> 0:11:52.360 bit more from John Cena again. 0:11:52.800 --> 0:11:57.360 My wrestling career is over. WWE terminates my contract and 0:11:57.559 --> 0:12:01.480 I'm stripped of my titles. Answers dropped me and I'm 0:12:01.520 --> 0:12:05.360 shunned by the wrestling community. My reputation is destroyed, and 0:12:05.440 --> 0:12:06.480 I'm left with nothing. 0:12:06.720 --> 0:12:08.960 I guess ai Johnsena doesn't know about what actually goes 0:12:09.000 --> 0:12:09.480 on in the w. 0:12:13.600 --> 0:12:16.400 I mean, in all seriousness, this appears to be a 0:12:16.440 --> 0:12:20.600 direct consequence of Meta lifting some of the chatbots guardrails 0:12:20.840 --> 0:12:24.120 after this hacker conference called def Con back in twenty 0:12:24.160 --> 0:12:27.680 twenty three. At this event, hackers tested the limits of 0:12:27.720 --> 0:12:31.320 the chatbots guardrails. Essentially, they tried to get chatbots to 0:12:31.400 --> 0:12:34.280 behave in a way that's out of line, and they 0:12:34.280 --> 0:12:38.960 found that Meta's chatbot was least likely n Vera's script, 0:12:39.160 --> 0:12:42.640 but also made it more boring than the other chatbots 0:12:42.640 --> 0:12:43.520 that were being tested. 0:12:43.600 --> 0:12:46.200 And the journal goes on to report that Mark Zuckerberg, 0:12:46.280 --> 0:12:49.400 the CEO of Meta, didn't like this. Some employees told 0:12:49.440 --> 0:12:52.040 the journal that he felt Meta had played it too safe, 0:12:52.559 --> 0:12:54.720 and so some of the guardrails were taken off and 0:12:55.200 --> 0:12:59.200 we arrived at this ai johnsena horror show. There's one 0:12:59.280 --> 0:13:01.520 quite from Zaka that really stood out to me from 0:13:01.520 --> 0:13:04.320 the article, and he was talking about how Meta needed 0:13:04.360 --> 0:13:07.320 to make their chatbots as human like as possible, to 0:13:07.400 --> 0:13:09.679 basically be head of the curve and to be as 0:13:09.679 --> 0:13:13.000 engaging impossible, and he said, quote, I missed out on 0:13:13.040 --> 0:13:16.120 Snapchat and TikTok, I won't miss out on this. 0:13:16.920 --> 0:13:18.600 That showed me to the bone because I was like, 0:13:18.679 --> 0:13:22.600 miss out on what like someone from Frozen talking about 0:13:22.640 --> 0:13:26.400 sex on Meta? You know what I mean? Well, you 0:13:26.400 --> 0:13:28.760 know he didn't miss out on Instagram, but right now 0:13:28.800 --> 0:13:30.959 that's not playing out so well for him. 0:13:31.000 --> 0:13:32.600 We're going to take a quick break now, but when 0:13:32.640 --> 0:13:35.400 we come back, we'll run through some short headlines and 0:13:35.440 --> 0:13:39.000 then welcome Bloomberg News. Is Olivia Carville to tech support. 0:13:39.440 --> 0:13:41.840 Olivia's going to fill us in on a consequential bill 0:13:41.880 --> 0:13:57.280 addressing AI harms. Stay with us, Welcome back. We've got 0:13:57.280 --> 0:14:00.000 a few more headlines to run through, starting with an 0:14:00.000 --> 0:14:03.679 an AI experiment conducted on Reddit. Our friends at four 0:14:03.800 --> 0:14:06.880 or for Media reported that researchers from the University of 0:14:06.960 --> 0:14:12.199 Zurich ran an unauthorized, large scale experiment where they used 0:14:12.240 --> 0:14:15.280 AI powered bots to comment on the popular Reddit forum 0:14:15.640 --> 0:14:19.720 are slash change my view. Posing as a variety of personas, 0:14:19.720 --> 0:14:21.920 such as a black man who was opposed to the 0:14:21.960 --> 0:14:26.120 Black Lives Matter movement, a sexual assault survivor a person 0:14:26.160 --> 0:14:29.800 who works at a domestic violence shelter. The AI bots 0:14:30.160 --> 0:14:33.600 tried to change real users' views, and the comments would 0:14:33.600 --> 0:14:37.000 actually be tailored to the original poster's gender, location, and 0:14:37.000 --> 0:14:40.560 political orientation in order to frame the arguments in a 0:14:40.600 --> 0:14:46.040 maximally persuasive way. Now these comments earned twenty thousand total upvotes, 0:14:46.400 --> 0:14:49.880 and users indicated that they'd successfully changed their minds over 0:14:49.920 --> 0:14:53.080 one hundred times. The experiment was revealed on the forum 0:14:53.160 --> 0:14:56.520 last week to much controversy. Reddit has responded saying they're 0:14:56.520 --> 0:15:00.320 considering legal action against the researchers and the universe Stieve. 0:15:00.360 --> 0:15:03.080 Zurich told for a full Media that the researchers will 0:15:03.080 --> 0:15:06.680 not be publishing their findings for now, at least as 0:15:06.680 --> 0:15:09.240 far as we know. The experiment is confined to the 0:15:09.280 --> 0:15:11.720 Credit forum, but it is terrifying to think of this 0:15:11.840 --> 0:15:14.520 deployed at scale across the wider Internet. 0:15:15.200 --> 0:15:17.120 So I'm going to start this one off by quoting 0:15:17.200 --> 0:15:22.040 directly from an announcement from AI company Anthropic. Human welfare 0:15:22.280 --> 0:15:25.240 is at the heart of our work at Anthropic. Our 0:15:25.280 --> 0:15:29.200 mission is to make sure that increasingly capable and sophisticated 0:15:29.240 --> 0:15:32.640 AI systems remain beneficial to humanity. 0:15:32.240 --> 0:15:35.440 And to make loads of money. But should we. 0:15:35.400 --> 0:15:39.680 Also be concerned about the potential consciousness and experiences of 0:15:39.720 --> 0:15:45.360 the models themselves? Should we be concerned about model welfare two? So, 0:15:45.800 --> 0:15:50.200 in fact, one of Anthropic's researchers recently said that there 0:15:50.240 --> 0:15:55.200 is a fifteen percent chance lms are already conscious. And now, 0:15:55.240 --> 0:15:58.040 with this announcement for a new research program devoted to 0:15:58.480 --> 0:16:02.720 model welfare, they are essentially saying, now is the time 0:16:02.840 --> 0:16:06.040 to consider the ethical implications of how we use and 0:16:06.120 --> 0:16:10.360 treat AI tools. You heard it here first, the PETA 0:16:10.560 --> 0:16:15.120 of AI is coming. But actually I thought Axios is 0:16:15.160 --> 0:16:17.160 a really good roundup of all the critiques of this 0:16:17.200 --> 0:16:20.760 AI welfare discussion, one being is this all hype or 0:16:20.800 --> 0:16:24.320 could it distract from more important questions about AI's potential harm? 0:16:24.440 --> 0:16:26.360 Well, I, for one think that when it comes to 0:16:26.480 --> 0:16:30.600 protecting rights and process there may be more urgent priorities 0:16:30.600 --> 0:16:34.520 than chatbots currently, but onto a more cheerful story. While 0:16:34.640 --> 0:16:38.800 dogs remain a man's best friend, maybe AI is becoming 0:16:39.000 --> 0:16:40.080 a dog's best friend. 0:16:40.680 --> 0:16:44.720 If AI can feed scraps under the table, it will happen. 0:16:44.840 --> 0:16:47.560 It will be for sure a dog's best friend. Naughtiness 0:16:47.640 --> 0:16:50.360 reports that AI powered tools are being tested to see 0:16:50.400 --> 0:16:53.440 if they can help doggy search and rescue teams find 0:16:53.440 --> 0:16:54.200 people faster. 0:16:54.720 --> 0:16:55.280 The way it. 0:16:55.200 --> 0:16:57.840 Works is that a drone hovers over the area where 0:16:57.880 --> 0:17:01.080 a dog is, searching data on the dog's movements and 0:17:01.120 --> 0:17:04.280 behavior predicting, for example, if they've picked up a cent, 0:17:04.880 --> 0:17:08.480 and stats about weather conditions, for example the direction of 0:17:08.520 --> 0:17:11.760 the wind are all combined and processed by an AI 0:17:11.800 --> 0:17:14.919 powered software that can then make predictions on where the 0:17:15.000 --> 0:17:17.960 person in trouble is most likely to be and deploy 0:17:18.240 --> 0:17:21.399 other drones to search that area before the dog actually 0:17:21.400 --> 0:17:24.320 reaches them. This program was developed with support from Darper, 0:17:24.920 --> 0:17:27.480 and it turns out that it's able to find simulated 0:17:27.560 --> 0:17:31.240 victims several times faster than a dog alone. Who's a 0:17:31.280 --> 0:17:32.160 good boy, Now. 0:17:32.280 --> 0:17:36.280 Good ones, That's very good Lastly, Netflix is introducing a 0:17:36.320 --> 0:17:39.120 new type of subtitle, but it's not for those who 0:17:39.160 --> 0:17:42.919 are hearing impaired. Multiple studies have shown that about half 0:17:42.960 --> 0:17:46.080 of American households watch TV and movies with subtitles. On 0:17:46.480 --> 0:17:48.640 myself included, even though I don't think of my house 0:17:48.640 --> 0:17:53.000 as a household. The streaming service is customizing a setting 0:17:53.000 --> 0:17:55.320 for people who just want to make sure they don't 0:17:55.320 --> 0:17:57.040 miss a word here and there because. 0:17:56.800 --> 0:17:59.280 They're looking for their phone on Mount Fuji. 0:17:58.840 --> 0:18:02.600 Because literally looking at their phone doing anything else but 0:18:02.680 --> 0:18:06.320 watching Netflix. According to Ours Technica, Netflix will give you 0:18:06.359 --> 0:18:11.520 the ability to watch and read your shows without music 0:18:11.600 --> 0:18:14.919 and sound effect descriptors and character names descriptions which are 0:18:15.000 --> 0:18:17.879 necessary for the hearing impaired, but not a must for 0:18:17.960 --> 0:18:21.040 those just multitasking on their second screen. 0:18:21.400 --> 0:18:24.320 Yeah, that eerie music descripture is often a little b 0:18:24.320 --> 0:18:25.119 annoying if you're. 0:18:25.000 --> 0:18:27.159 Not hot at hearing right exactly, I'm like, I know, 0:18:27.280 --> 0:18:29.320 this is a suspenseful moment. Let's keep on with the show. 0:18:29.400 --> 0:18:33.159 That's what I'm looking out for my phone. Something changed. 0:18:33.680 --> 0:18:37.240 Even with unwavering focus, modern TV show dialogue can be 0:18:37.240 --> 0:18:41.520 hard to understand. It's a combination of actors performing more naturalistically, 0:18:41.920 --> 0:18:46.200 so speaking with more volume fluctuation, plus streaming services compressing 0:18:46.200 --> 0:18:49.879 their audio files more tightly than what's standard for physical media. 0:18:49.960 --> 0:18:52.240 When did you become an audio nert care price? 0:18:53.720 --> 0:18:56.959 Actually quite a long time ago. Amazon Prime already has 0:18:57.000 --> 0:18:59.119 a band aid for this, which is called the Dialogue 0:18:59.119 --> 0:19:02.240 Boost option. And I do think it's wild that streaming 0:19:02.359 --> 0:19:04.959 change the business so much. But instead of fixing the 0:19:05.000 --> 0:19:08.560 audio issues at the source by engineering audio differently, the 0:19:08.600 --> 0:19:11.679 solution is, of course, just to patch. 0:19:11.440 --> 0:19:15.000 The problem in the interest of patching a much more 0:19:15.000 --> 0:19:19.080 important problem. There is a landmark bill aimed at combating 0:19:19.119 --> 0:19:21.520 AI harms, specifically deep fakes. 0:19:21.800 --> 0:19:25.399 They're used in scams, they're used in spreading misinformation online, 0:19:25.720 --> 0:19:29.000 and I'd say most notably, they have been used in 0:19:29.040 --> 0:19:31.400 the non consensual creation of porn. 0:19:31.520 --> 0:19:34.600 Right, and that's what this legislation is all about. This week, 0:19:34.640 --> 0:19:37.119 Congress passed the Take It Down Act, which aims to 0:19:37.160 --> 0:19:40.000 crack down on the creation of revenge porn i e. 0:19:40.160 --> 0:19:43.840 Pornographic images that are shared non consensually. The Act specifies 0:19:43.840 --> 0:19:47.000 that those who distribute revenge porn, whether quote real or 0:19:47.040 --> 0:19:50.600 computer generated, could be fined or subject to prison time. 0:19:51.240 --> 0:19:53.880 It's had rare backing from both sides of the political 0:19:53.920 --> 0:19:58.280 aisle and from First Lady Milania Trump. As of Wednesday afternoon, 0:19:58.440 --> 0:20:00.480 the time of this taping, the b bill heads to 0:20:00.520 --> 0:20:02.480 President Trump, who's likely to make it law. 0:20:02.960 --> 0:20:05.000 Here to walk us through the Take It Down Act 0:20:05.040 --> 0:20:07.640 and what it means for tech companies is Olivia Carvell, 0:20:07.960 --> 0:20:11.320 investigative reporter for Bloomberg News and co host of the 0:20:11.320 --> 0:20:15.359 podcast Levettown, which is a must listen agreed wherever you 0:20:15.400 --> 0:20:19.359 get your podcasts, and covers the rise of deep fake porn. 0:20:19.640 --> 0:20:23.640 It also happens to be a co production of Kaleidoscope. Olivia, 0:20:23.800 --> 0:20:24.919 Welcome to Tech Stuff. 0:20:25.080 --> 0:20:26.879 Thank you so much for having me. It's great to 0:20:26.880 --> 0:20:28.600 be back with Kaleidoscope's team. 0:20:29.000 --> 0:20:31.480 Thanks thanks for being here, Olivia. You've been tracking this 0:20:31.520 --> 0:20:33.840 bill for a long time. When did the push for 0:20:34.320 --> 0:20:37.680 legislation on deep fake pornography begin? 0:20:38.440 --> 0:20:40.919 I mean it has been a very long journey to 0:20:41.000 --> 0:20:44.200 get here. We've seen quite a lot of states across 0:20:44.240 --> 0:20:48.040 the US rolling out legislation to try and target deep 0:20:48.040 --> 0:20:51.840 fake porn since the revolution really began a number of 0:20:51.960 --> 0:20:55.440 years ago. Now at the moment, more than twenty states 0:20:55.480 --> 0:20:59.000 across the country have introduced new laws. But one of 0:20:59.040 --> 0:21:01.680 the criticisms we heard time and time again, and something 0:21:01.800 --> 0:21:05.040 we raised in the Levetown podcast, is the fact that 0:21:05.080 --> 0:21:09.280 there was no federal law criminalizing this across the US. 0:21:10.040 --> 0:21:13.680 And this bill was first introduced last summer in twenty 0:21:13.720 --> 0:21:18.760 twenty four, its bipartisan legislation. Senators Cruise and Clobashah put 0:21:18.800 --> 0:21:22.159 it forward and it unanimously passed in the Senate, but 0:21:22.320 --> 0:21:25.359 unfortunately it stalled in the House last year and that 0:21:25.480 --> 0:21:28.800 led to a lot of frustration from the victims. Earlier 0:21:28.840 --> 0:21:31.800 this year, we saw it once again. Take It Down 0:21:31.960 --> 0:21:36.000 was reintroduced, unanimously passed in the Senate, and then earlier 0:21:36.040 --> 0:21:40.080 this week and very exciting news, it was also unanimously 0:21:40.119 --> 0:21:42.600 passed in the House. And we're talking a vote of 0:21:42.680 --> 0:21:45.960 four hundred and nine to two, and that's kind of 0:21:46.040 --> 0:21:49.639