WEBVTT - Tech News: AI Stalks Us on Friday the 13th 0:00:04.480 --> 0:00:12.440 Welcome to Tech Stuff, a production from iHeartRadio. Hey therein 0:00:12.600 --> 0:00:16.480 Welcome to Tech Stuff. I'm your host, Jonathan Strickland. I'm 0:00:16.480 --> 0:00:20.000 an executive producer with iHeart Podcasts and How the tech 0:00:20.040 --> 0:00:22.520 are you. It's time for the tech news for the 0:00:22.520 --> 0:00:29.000 week ending on Friday, September thirteenth, twenty twenty four. Happy 0:00:29.200 --> 0:00:32.559 Friday the thirteenth, everybody. I'm wishing all of you the 0:00:32.600 --> 0:00:37.040 best of luck and hopefully a notable absence of a hulking, undead, 0:00:37.080 --> 0:00:41.559 hockey mask wearing psychopath. We start off this episode with 0:00:41.760 --> 0:00:46.239 some stories about AI, as is our wont to do so. 0:00:46.360 --> 0:00:52.000 First up, Kylie Robinson of The Verge Robison, perhaps my apologies, Kylie. 0:00:52.240 --> 0:00:55.360 Kylie of the Verge has an article titled open AI 0:00:55.560 --> 0:01:01.400 releases OH one, its first model with reasoning abilities, which 0:01:01.440 --> 0:01:05.800 sounds pretty significant, and Kylie explains that one is the 0:01:05.840 --> 0:01:10.399 first of several planned sophisticated AI models that should be 0:01:10.440 --> 0:01:14.920 capable of answering complex queries. While in development, this was 0:01:14.959 --> 0:01:19.680 known to the AI crowd as the Strawberry Model. That's 0:01:19.720 --> 0:01:22.920 actually the first I've heard of that code name. But 0:01:22.959 --> 0:01:25.480 then again, I'm not exactly brought in to hear about 0:01:25.520 --> 0:01:28.880 all things AI. The model is said to be able 0:01:28.920 --> 0:01:32.759 to tackle multi step problems, which is something that earlier 0:01:32.800 --> 0:01:37.560 AI models have had troubled doing, and doing this comes 0:01:37.560 --> 0:01:40.720 at a couple of costs. There's the cost of time, 0:01:41.000 --> 0:01:44.160 because it does take a little longer for this model 0:01:44.240 --> 0:01:47.520 to create an answer, although it typically is still faster 0:01:47.640 --> 0:01:50.000 than humans would be able to do it. But then 0:01:50.040 --> 0:01:54.400 there's also the financial cost. These models aren't free to use, 0:01:54.600 --> 0:01:57.440 and they come with a fairly hefty fee for access. 0:01:57.600 --> 0:02:00.720 That fee is like three to four times more expensive 0:02:01.000 --> 0:02:06.520 than using the current AI models for other purposes. Now, 0:02:06.520 --> 0:02:10.160 according to open ai representatives, the one model is less 0:02:10.240 --> 0:02:13.960 prone to hallucinations than earlier models are, but it is 0:02:14.040 --> 0:02:17.480 still not immune to them. So the challenge of creating 0:02:17.480 --> 0:02:21.080 an AI model that you know is dependable and accountable 0:02:21.320 --> 0:02:24.280 without having to worry that it's just making stuff up, Well, 0:02:24.320 --> 0:02:29.560 that problem persists. It's also apparently pretty good at coding. 0:02:30.160 --> 0:02:34.200 It can't outperform the best of the best of human coders, 0:02:34.200 --> 0:02:36.760 at least not yet, but I feel like if open 0:02:36.800 --> 0:02:40.240 ai can convince companies that you know, their AI models 0:02:40.280 --> 0:02:44.320 are equivalent to exceptional coders. Maybe not the best, but 0:02:44.600 --> 0:02:49.040 really good coders. That's a pretty big sales bitch. Coders 0:02:49.280 --> 0:02:53.080 who don't have salaries, they don't have healthcare plans, they 0:02:53.120 --> 0:02:57.600 don't have stock options or retirement accounts or anything like that. 0:02:58.000 --> 0:03:00.840 You know. I know that AI ideally is meant to 0:03:01.080 --> 0:03:04.520 augment humans so that they can do their jobs better 0:03:04.760 --> 0:03:08.800 and they can offload the really tedious work to automated processes. 0:03:09.080 --> 0:03:11.560 But I still have concerns that, at least in the 0:03:11.600 --> 0:03:14.000 short term, there are going to be business leaders out 0:03:14.000 --> 0:03:16.720 there who will view AI as kind of a shortcut 0:03:16.800 --> 0:03:21.720 to downsizing staff and outsourcing work to the robots. Marie 0:03:21.760 --> 0:03:25.239 Baran of Newsweek wrote an article this week titled AI 0:03:25.400 --> 0:03:29.840 generated junk science is flooding Google scholar studied claims, and 0:03:30.080 --> 0:03:33.040 assuming that the study is correct, it illustrates one of 0:03:33.080 --> 0:03:38.000 the many concerns I have about generative AI. Essentially, the 0:03:38.080 --> 0:03:42.000 study found that AI generated science papers, fake science papers 0:03:42.000 --> 0:03:45.000 in other words, are showing up in Google scholar right 0:03:45.040 --> 0:03:49.800 next to legitimate, you know, scholarly articles within search results. 0:03:50.040 --> 0:03:53.680 So Google has been indexing pages that are made by 0:03:53.720 --> 0:03:57.640 AI bots, you know, like chat GPT. The Harvard Kennedy 0:03:57.680 --> 0:04:03.600 School Misinformation Review published this study and identified one hundred 0:04:03.640 --> 0:04:07.240 and thirty nine papers that are likely generated by AI, 0:04:07.680 --> 0:04:11.320 and more than half of those identified papers focused on 0:04:11.400 --> 0:04:15.640 topics that relate closely to stuff like climate change, health concerns, 0:04:15.680 --> 0:04:19.720 and other policy relevant subject matter. And the danger here 0:04:19.839 --> 0:04:23.279 is obvious. The papers may appear to be from legitimate 0:04:23.320 --> 0:04:27.080 researchers who followed rigorous scientific practices in order to you know, 0:04:27.160 --> 0:04:31.039 draw their conclusions, but in reality, they could just be 0:04:31.160 --> 0:04:35.279 propaganda that's pushing a specific point of view and disguised 0:04:35.360 --> 0:04:38.279 as a legitimate scientific paper and considering a lot of 0:04:38.279 --> 0:04:42.279 people quote unquote research by just looking for passages and 0:04:42.360 --> 0:04:46.240 papers that appear to confirm what the people already believe, 0:04:46.640 --> 0:04:49.120 you know, just cherry picking. In other words, this could 0:04:49.200 --> 0:04:52.480 lead to situations in which folks are citing fraudulent papers 0:04:52.680 --> 0:04:57.000 because those papers support specious arguments. Now y'all already know 0:04:57.120 --> 0:05:01.640 that I call out for critical thinking regular on this show, 0:05:01.839 --> 0:05:05.800 and this trend really drives home how important critical thinking is. 0:05:06.320 --> 0:05:09.919 I am just as guilty as anyone else of grabbing 0:05:09.960 --> 0:05:13.800 onto a source that appears to confirm my biases. I 0:05:13.839 --> 0:05:16.520 actually have to remind myself to take a closer look 0:05:16.560 --> 0:05:20.000 to one make sure that the source is legitimate, and 0:05:20.040 --> 0:05:24.080 then make double sure that what it's actually saying is 0:05:24.120 --> 0:05:27.039 what I think it says, because that's not always the case. 0:05:27.160 --> 0:05:30.880 I have been known to misinterpret stuff like that has happened, 0:05:31.240 --> 0:05:33.760 so I have to be careful about this too, and 0:05:33.800 --> 0:05:36.520 I don't always succeed, but it's good to keep it 0:05:36.560 --> 0:05:38.880 in mind, like this is something to strive for, use 0:05:38.920 --> 0:05:41.760 critical thinking, and it sounds to me like this is 0:05:41.839 --> 0:05:44.160 just going to get more challenging to do as time 0:05:44.240 --> 0:05:47.080 goes on. One thing that could help is if Google 0:05:47.200 --> 0:05:51.280 developed some tools to make Google scholar more reliable. You know, 0:05:51.400 --> 0:05:54.360 knowing that these things are going to exist, how can 0:05:54.440 --> 0:05:59.719 Google scholar better differentiate between legitimate papers and those that 0:05:59.760 --> 0:06:04.799 were generated by AI. That might involve improving transparency regarding 0:06:04.839 --> 0:06:08.360 how Google Scholar indexes and ranks scholarly results in the 0:06:08.400 --> 0:06:11.599 first place, or giving more tools so you can filter 0:06:11.800 --> 0:06:15.200 stuff out, like make sure that the articles you get 0:06:15.240 --> 0:06:18.919 are from legitimate peer reviewed sources. I suppose I should 0:06:18.960 --> 0:06:22.839 also throw in this potential pitfall. If future generations of 0:06:22.960 --> 0:06:27.360 AI models train themselves on fake scientific papers that were 0:06:27.360 --> 0:06:31.040 created by earlier AI models, then the quality of content 0:06:31.080 --> 0:06:34.920 on the Internet will further decline reasoning AI models like 0:06:34.960 --> 0:06:38.160 the aforementioned one would be more likely to create incorrect 0:06:38.200 --> 0:06:42.960 solutions if the training material it used included this sort 0:06:43.000 --> 0:06:47.200 of stuff, these fraudulent scientific papers, because AI can't just 0:06:47.360 --> 0:06:53.520 magically know what information is relevant and dependable and which 0:06:53.560 --> 0:06:57.320 one is just invented in order to push forward a narrative. 0:06:57.920 --> 0:07:00.200 Emma Roth that The Verge has an article that it's 0:07:00.240 --> 0:07:04.479 titled Google is using AI to make fake podcasts from 0:07:04.600 --> 0:07:07.240 your notes, and it kind of makes me think of 0:07:07.680 --> 0:07:10.480 a recent episode I did on tech Stuff. I titled 0:07:10.480 --> 0:07:15.240 it this episode was written by AI sort of, So 0:07:15.520 --> 0:07:18.240 in that episode, I use chat GPT to create a 0:07:18.320 --> 0:07:21.880 tech podcast episode not just not of tech stuff, just 0:07:22.280 --> 0:07:25.800 a tech podcast, and it was supposed to trace the 0:07:25.880 --> 0:07:28.840 history and the technology of airbags. And I read the 0:07:29.040 --> 0:07:33.440 entire generated episode out in that podcast episode of tech Stuff. 0:07:33.520 --> 0:07:36.160 I then spent the rest of the episode fact checking 0:07:36.200 --> 0:07:39.520 and critiquing the AI's work. Now, the most disturbing thing 0:07:39.640 --> 0:07:44.360 I encountered with this experiment was that the AI kept 0:07:44.480 --> 0:07:49.760 inventing fake experts to deliver various bits of information. Sometimes 0:07:49.760 --> 0:07:52.920 that information was wrong, and all the experts didn't really 0:07:53.000 --> 0:07:57.360 exist anyway. In Roth's article, it unfolds that Google created 0:07:57.400 --> 0:08:00.800 a feature in its notebook LM app that will take 0:08:00.880 --> 0:08:03.880 notes that you have written down. It will then generate 0:08:04.240 --> 0:08:07.960 this you know, AI created podcast hosted by a couple 0:08:08.040 --> 0:08:12.160 of AI bots posing as the hosts, and the podcast 0:08:12.520 --> 0:08:16.840 has hosts having a discussion about whatever the research topic is, 0:08:16.960 --> 0:08:19.920 and it uses your notes to create a conversation between 0:08:19.960 --> 0:08:23.520 these two bot hosts, and it kind of riffs off 0:08:23.560 --> 0:08:26.920 the information that you have gathered. It sounds like the 0:08:27.000 --> 0:08:31.280 AI is careful to only draw information from your notes, 0:08:31.600 --> 0:08:35.240 so the output you get should reflect the input that 0:08:35.320 --> 0:08:39.040 the bots relied upon. So in other words, if something 0:08:39.120 --> 0:08:41.600 is wrong in the episode, it would be because your 0:08:41.679 --> 0:08:45.800 notes have wrong information in them or incomplete information. The 0:08:45.880 --> 0:08:49.840 AI wouldn't necessarily be hallucinating or confabulating, you know, drawing 0:08:50.240 --> 0:08:54.600 from some source you've never seen either, And that's a 0:08:54.600 --> 0:08:56.880 good thing. It does remind me. In the case of 0:08:56.960 --> 0:09:01.240 the AI episode that I generated, Chad gbts was actually 0:09:01.360 --> 0:09:04.679 unable to share with me what the sources were that 0:09:04.760 --> 0:09:07.520 it was pulling information from. I asked it to, and 0:09:07.559 --> 0:09:11.240 it couldn't. Instead, it gave me a list of sources 0:09:11.240 --> 0:09:14.160 that the information might have come from, but there was 0:09:14.240 --> 0:09:17.040 no guarantee that any of the information used in the 0:09:17.080 --> 0:09:20.920 episode actually came from those sources. Now, in this case, 0:09:20.960 --> 0:09:23.800 with Google, the sources you or at least the notes 0:09:24.120 --> 0:09:27.520 that you've taken. And I think this approach is interesting. 0:09:27.640 --> 0:09:31.200 It doesn't strike me quite as off putting as what 0:09:31.280 --> 0:09:34.600 I experienced with chat GPT for one thing. To me, 0:09:34.760 --> 0:09:37.080 this feels more like a study tool. I mean, we 0:09:37.160 --> 0:09:40.480 all know that people have different learning styles, right, so 0:09:40.640 --> 0:09:43.319 I think this tool could potentially be good for someone 0:09:43.760 --> 0:09:47.240 who does take meticulous notes, but that doesn't really help 0:09:47.320 --> 0:09:51.480 them absorb the information, right they don't really understand they've 0:09:51.520 --> 0:09:54.320 got the notes, but it hasn't kind of sunk in. 0:09:54.880 --> 0:09:57.800 So I think this kind of approach could create a 0:09:58.160 --> 0:10:02.680 way of synthesizing and contect rualizing the information that could 0:10:02.720 --> 0:10:06.280 be more impactful depending upon the subject matter and you know, 0:10:06.360 --> 0:10:09.480 the learning style of the person involved. So as a 0:10:09.520 --> 0:10:13.520 studying tool, I think it's a pretty neat idea. Now, 0:10:14.000 --> 0:10:18.160 I also wonder if this will ultimately lead to people 0:10:18.240 --> 0:10:22.120 using this tool to create podcasts that are hosted by AI, 0:10:22.559 --> 0:10:25.520 which could be a problem. I mean, especially considering that 0:10:25.840 --> 0:10:28.400 it's going to be based on whatever notes were made 0:10:28.440 --> 0:10:31.600 to create the podcast, so you could do it to 0:10:31.840 --> 0:10:34.480 make them say whatever you wanted, or you know, not 0:10:34.640 --> 0:10:37.760 make them say, but they would say things drawn from 0:10:38.080 --> 0:10:40.880 your own perspective in the notes you created. Now, that 0:10:40.880 --> 0:10:43.560 could be funny if you were to create really weird 0:10:43.679 --> 0:10:47.400 notes about obviously fake stuff, not in an effort to 0:10:47.520 --> 0:10:51.440 mislead listeners, but rather as a way to entertain them. 0:10:51.760 --> 0:10:54.240 And I'm thinking about shows that are something along the 0:10:54.240 --> 0:10:59.000 lines of existing fictional podcasts out there, stuff like Welcome 0:10:59.000 --> 0:11:02.720 to night Vale or Old Gods of Appalachia or my 0:11:02.920 --> 0:11:06.800 friend Shay's podcast Kadi Womple with the Shadow People. And yes, 0:11:06.880 --> 0:11:09.440 that last one is real. Kadi Womple with the Shadow 0:11:09.480 --> 0:11:12.120 People is a real podcast, and yes I am plugging 0:11:12.160 --> 0:11:15.600 my friends podcast sort of. So if Southern gothic fantasy 0:11:15.679 --> 0:11:18.120 with a healthy dose of feminism is your kick, you 0:11:18.120 --> 0:11:19.920 should check it out. In fact, you should just check 0:11:19.960 --> 0:11:22.520 it out anyway, give an episode a listen, because Shay 0:11:22.640 --> 0:11:25.920 is a great storyteller, and you know, maybe it's your jam, 0:11:25.960 --> 0:11:28.600 maybe it's not, but yeah, I could see this tool 0:11:28.679 --> 0:11:31.720 being used for that kind of thing. That arguably that 0:11:31.960 --> 0:11:36.040 does bring into question the artistry although you would still 0:11:36.080 --> 0:11:38.360 need to put in the work to create the source 0:11:38.880 --> 0:11:42.720 notes that the hosts are drawing from. So it's a 0:11:42.760 --> 0:11:45.800 gray area for me. Like generally, I tend to be 0:11:45.920 --> 0:11:49.800 pretty negative or pretty critical at least about generative AI, 0:11:50.280 --> 0:11:52.559 But if it comes down to something like this, where 0:11:52.640 --> 0:11:56.800 you have done an enormous amount of work to build 0:11:56.840 --> 0:12:02.400 the source material, I'm not quite as adamant the generative 0:12:02.400 --> 0:12:06.760 AI is a bad tool to use in this context. 0:12:06.800 --> 0:12:08.839 But maybe I just need to think on it more. 0:12:09.200 --> 0:12:13.160 Now back to artificial intelligence in general. So on Science Alert, 0:12:13.480 --> 0:12:17.080 David Neil has an article that's titled AI chatbots have 0:12:17.160 --> 0:12:21.720 a political bias that could unknowingly influence society. Now, I 0:12:21.760 --> 0:12:24.240 don't think this should come as a surprise because bias 0:12:24.280 --> 0:12:27.880 has been a big issue in AI for decades. Some 0:12:28.080 --> 0:12:31.960 experts have argued that not all bias is bad, right, Like, 0:12:32.280 --> 0:12:35.640 you might build an AI model that is quote unquote 0:12:35.720 --> 0:12:40.440 biased to pick out instances of say, medical images that 0:12:40.480 --> 0:12:43.440 could indicate a health hazard, like the presence of say 0:12:43.440 --> 0:12:48.080 a tumor, for example. But unintended bias is bad, and 0:12:48.160 --> 0:12:50.760 we've seen lots of examples of that with AI, like 0:12:50.840 --> 0:12:55.320 with facial recognition, technologies and the like. Neiled cites a 0:12:55.600 --> 0:12:59.959 computer scientist named David Rosato from Otago Polytechnic and New 0:13:00.080 --> 0:13:04.240 Zealand who uses various political questionnaires to test different AI 0:13:04.320 --> 0:13:07.840 chat models to see where they fall on the political 0:13:07.880 --> 0:13:11.360 spectrum based upon their responses to these questionnaires, and according 0:13:11.400 --> 0:13:15.520 to his results, the models all fell somewhere left of 0:13:15.720 --> 0:13:19.760 center on political matters, and they tended toward a more 0:13:19.840 --> 0:13:23.839 libertarian point of view rather than an authoritarian point of view. 0:13:24.160 --> 0:13:26.320 None of the models were coming out as like hard 0:13:26.480 --> 0:13:29.840 left evangelists or anything like that, but the bias was 0:13:30.280 --> 0:13:34.400 present and it was significant. Rosatto doesn't believe that the 0:13:34.400 --> 0:13:38.880 bias was intentional, but rather it's sort of an emergent quality. 0:13:39.200 --> 0:13:41.960 And why is it emerging at all? Well, the best 0:13:42.000 --> 0:13:44.600 guess is that the material used to train the AI 0:13:44.760 --> 0:13:49.439 models skews left more than it does right. Not all 0:13:49.440 --> 0:13:52.080 of it, but that overall, when taken as in it, 0:13:52.320 --> 0:13:55.120 you know, a hole, it skews more left. That there 0:13:55.120 --> 0:13:58.439 are more pieces written from a left of center perspective 0:13:58.520 --> 0:14:02.559 than right of center. This in turn imbalances the material, 0:14:02.960 --> 0:14:06.240 so that leads to a bias in the models. And 0:14:06.280 --> 0:14:08.800 it kind of makes me think about how matter and 0:14:08.840 --> 0:14:13.440 antimatter are. So when matter and antimatter come into contact 0:14:13.440 --> 0:14:16.840 with one another, they annihilate each other. So if there 0:14:16.920 --> 0:14:20.720 had been a perfect balance of matter and antimatter at 0:14:20.800 --> 0:14:24.120 the dawn of time, there would be no universe to 0:14:24.120 --> 0:14:26.200 speak up, because it would have all blowed up before 0:14:26.200 --> 0:14:29.080 it could even get started. But for some reason, there 0:14:29.160 --> 0:14:33.320 was a little bit more matter than there was antimatter, 0:14:33.560 --> 0:14:37.320 and we got the universe. So with these AI models, 0:14:37.360 --> 0:14:41.080 the training material had more left of center perspective material 0:14:41.400 --> 0:14:45.520 than otherwise, which y'all you know, I lean left so hard. 0:14:45.560 --> 0:14:48.120 I walk around at a forty five degree angle. But 0:14:48.920 --> 0:14:53.040 I don't think having a biased perspective in the tools 0:14:53.120 --> 0:14:58.680 themselves that are meant to provide and contextualize information is 0:14:58.720 --> 0:15:02.240 a good thing, even though that bias kind of leans 0:15:02.280 --> 0:15:05.400 toward the way that I view the world. I don't 0:15:05.400 --> 0:15:07.640 think a bias is good in that respect. It needs 0:15:07.680 --> 0:15:11.240 to be as objective as it possibly can, in my opinion, 0:15:11.800 --> 0:15:15.080 So if the bias means we can't rely on the 0:15:15.120 --> 0:15:18.600 results provided, that ends up being a big problem, especially 0:15:18.680 --> 0:15:22.360 considering how gung ho everybody is on AI. Now. Despite 0:15:22.400 --> 0:15:25.520 these findings, we have also seen examples of generative AI 0:15:25.720 --> 0:15:29.440 engaged in recreating some really ugly biases as well. I'm 0:15:29.440 --> 0:15:32.720 thinking primarily of image generating models that tend to be 0:15:32.760 --> 0:15:35.920 guilty of perpetuating racial stereotypes. So I guess you could 0:15:35.920 --> 0:15:38.600 see this issue as a wake up call regarding our 0:15:38.640 --> 0:15:42.520 own prejudices and biases on top of the issue we 0:15:42.560 --> 0:15:45.560 have with AI. Okay, we've got a ton more news 0:15:45.600 --> 0:15:47.520 to get through. Let's take a quick break to thank 0:15:47.560 --> 0:15:59.720 our sponsors. We're back and we're not done with AI 0:15:59.880 --> 0:16:03.240 just yet. Just Weather Bed of the Verge has a 0:16:03.320 --> 0:16:07.680 beast titled Meta fed its AI on almost everything you've 0:16:07.720 --> 0:16:11.400 posted publicly since two thousand and seven. And yeah, that 0:16:11.520 --> 0:16:14.080 article starts off with a whammy. In fact, I'm just 0:16:14.120 --> 0:16:16.480 going to quote whether Bed at the beginning of the article. 0:16:16.720 --> 0:16:20.760 Who writes quote Meta has acknowledged that all text and 0:16:20.880 --> 0:16:24.960 photos that adult Facebook and Instagram users have publicly published 0:16:25.280 --> 0:16:28.480 since two thousand and seven have been fed into its 0:16:28.600 --> 0:16:33.080 artificial intelligence models. End quote. Now this is significant for 0:16:33.600 --> 0:16:36.720 many reasons, one of which is that Meta executives had 0:16:36.760 --> 0:16:40.440 previously sort of denied that this was the case when 0:16:40.480 --> 0:16:44.200 asked by Australian legislators if user data was being exploited 0:16:44.280 --> 0:16:46.840 in this way, but ultimately they did cop to the 0:16:46.880 --> 0:16:51.080 practice when lawmakers really cornered them with pretty direct questions 0:16:51.080 --> 0:16:55.000 that they couldn't just deflect. So essentially, unless users had 0:16:55.320 --> 0:16:58.360 set their content to something other than public, you know, 0:16:58.440 --> 0:17:01.560 like friends only or private or whatever, then that content 0:17:01.680 --> 0:17:04.480 was up for grabs and Meta grabbed it for the 0:17:04.480 --> 0:17:08.040 purposes of training AI. Meta didn't go so far as 0:17:08.080 --> 0:17:10.960 to explain if there's a cutoff for when the data 0:17:11.000 --> 0:17:14.840 scraping happened. So, for example, assuming that it does go 0:17:14.880 --> 0:17:17.160 all the way back to two thousand and seven, can 0:17:17.200 --> 0:17:20.840 the bots scrape everything that was ever posted to the platform, 0:17:21.040 --> 0:17:24.800 at least publicly? And could that be the case even 0:17:24.840 --> 0:17:27.520 if the person who posted that stuff was a minor 0:17:27.680 --> 0:17:31.040 at the time, so all of those posts, including images, 0:17:31.440 --> 0:17:34.040 could be up for grabs. Now, Meta has said it 0:17:34.160 --> 0:17:37.080 does not scrape profiles of users who are under the 0:17:37.119 --> 0:17:41.560 age of eighteen. Fine, but what about users who today 0:17:41.880 --> 0:17:45.040 are adults but have been on Facebook long enough so 0:17:45.080 --> 0:17:48.280 that the earliest days of their Facebook use was when 0:17:48.280 --> 0:17:51.520 they were under the age of eighteen? Did the data 0:17:51.560 --> 0:17:55.240 scraping include their data from that time? I mean, yeah, 0:17:55.280 --> 0:17:58.400 today they're adults. But when they posted those things back 0:17:58.440 --> 0:18:01.320 in say, two thousand and seven, they well, that question 0:18:01.440 --> 0:18:04.840 is more murky, and to be truthful, the Meta representatives 0:18:04.880 --> 0:18:08.280 didn't really have an answer for it, and that's very concerning. Now. 0:18:08.280 --> 0:18:11.800 Meta does allow users to opt out of this data 0:18:11.840 --> 0:18:15.320 scraping practice if those users happen to live in the 0:18:15.359 --> 0:18:19.640 European Union, where regional laws mandate that Meta create this 0:18:19.720 --> 0:18:23.920 option to opt out. Likewise, laws in Brazil required Meta 0:18:23.960 --> 0:18:27.000 to cut out the data scraping for AI there as well, 0:18:27.280 --> 0:18:29.840 But everywhere else in the world, it's fair game, baby. 0:18:30.040 --> 0:18:33.520 If it ain't expressly against the law, Meta is doing it, 0:18:34.000 --> 0:18:36.440 which might be food for thought for all the other 0:18:36.560 --> 0:18:39.439 countries out there, at least the ones with any interest 0:18:39.440 --> 0:18:44.320 at all regarding protecting citizen data from massive corporations. Metta 0:18:44.400 --> 0:18:46.560 is also in the news here in the United States, 0:18:46.640 --> 0:18:50.639 as Republican Congressman Tim Wahlberg has some pretty harsh words 0:18:50.680 --> 0:18:53.880 for the company after it responded to concerns over how 0:18:53.920 --> 0:18:57.040 it has hosted ads for illegal drugs, which I talked 0:18:57.080 --> 0:19:00.400 about in an earlier Tech Stuff episode, so to kind 0:19:00.400 --> 0:19:04.760 of summarize, Legislators had sent an inquiry to Meta following 0:19:04.800 --> 0:19:08.240 reports from the Tech Transparency Project as well as the 0:19:08.240 --> 0:19:12.760 Wall Street Journal that detailed how advertisements for illegal drugs 0:19:12.800 --> 0:19:17.400 were appearing on Facebook and Instagram, both prescription drugs and 0:19:17.760 --> 0:19:21.760 recreational drugs, like log on and suddenly there's an ad 0:19:21.760 --> 0:19:25.840 for cocaine on your Facebook feed. And this means that 0:19:25.920 --> 0:19:29.640 Meta was not only providing a platform that these illegal 0:19:29.720 --> 0:19:33.200 ads got to use, but Meta itself was profiting from 0:19:33.280 --> 0:19:37.400 these illegal advertisements. Now, Meta didn't create the ads, They're 0:19:37.440 --> 0:19:40.520 just hosting them, but they are profiting from them because 0:19:40.520 --> 0:19:44.480 the advertisers have to pay Meta to have this space. Right, 0:19:44.800 --> 0:19:48.119 So the lawmakers sent more than a dozen questions to 0:19:48.160 --> 0:19:50.640 Meta to really get down to how big an issue 0:19:50.640 --> 0:19:54.359 this is, how prevalent is this problem, and what the 0:19:54.400 --> 0:19:58.120 heck is Meta doing about it? And Meta essentially responded 0:19:58.160 --> 0:20:01.600 by saying, and to be clear, I am paraphrasing like 0:20:01.840 --> 0:20:05.240 crazy here, but they said essentially like, yeah, you know, 0:20:05.359 --> 0:20:09.760 that's crazy. We agree that's crazy. But Meta is all 0:20:09.800 --> 0:20:12.240 about doing its part to fight illegal activity. This is 0:20:12.240 --> 0:20:15.800 a big issue beyond any one platform. This is this 0:20:15.880 --> 0:20:17.720 is major. This isn't just us, This is this is 0:20:17.720 --> 0:20:21.240 a big problem. Now, Wahlberg was not buying this and 0:20:21.280 --> 0:20:24.920 called the response unacceptable. I agree with him. He went 0:20:24.960 --> 0:20:28.800 on to say, quote Metta's response not only ignores most 0:20:28.800 --> 0:20:31.600 of the questions posed in our letter, but also refuses 0:20:31.640 --> 0:20:35.199 to acknowledge that these illicit drug ads were approved and 0:20:35.400 --> 0:20:39.120 monetized by Meta and allowed to run on their platforms 0:20:39.320 --> 0:20:42.880 end quote. The director of Tech Transparency Project, Katie Paul, 0:20:43.080 --> 0:20:47.359 also accused Meta of deliberately sidestepping questions of accountability in 0:20:47.400 --> 0:20:49.600 an effort to deflect and to claim that this is 0:20:49.680 --> 0:20:53.360 just a bigger issue, much bigger than Meta and its platforms. 0:20:53.800 --> 0:20:58.000 And CEO Mark Zuckerberg recently said on a podcast interview 0:20:58.200 --> 0:21:00.720 that he thinks in the past he has made the 0:21:00.720 --> 0:21:04.479 mistake of accepting responsibility for stuff that subsequently he believes 0:21:04.520 --> 0:21:07.480 his company wasn't actually guilty of doing. Like he said, 0:21:07.480 --> 0:21:09.679 we got to stop saying we're sorry for stuff that 0:21:09.840 --> 0:21:12.359 isn't our fault. Essentially is how it came across to 0:21:12.440 --> 0:21:16.280 me anyway. But this particular subject seems pretty cut and 0:21:16.359 --> 0:21:19.760 dried to me, because either Meta was selling ads to 0:21:19.800 --> 0:21:24.359 clients who ultimately made those ads about illegal drugs, or 0:21:24.440 --> 0:21:28.440 Meta did not do that. So either Meta profited off 0:21:28.480 --> 0:21:33.200 of illegal advertisements or it didn't. There's not grey area here, 0:21:33.800 --> 0:21:36.600 you know. And sure Meta could argue that the scale 0:21:36.680 --> 0:21:39.240 of its business as such that it can't police every 0:21:39.320 --> 0:21:43.200 ad or ensure that an ad that was sold actually 0:21:43.280 --> 0:21:45.720 ends up being for whatever it was sold to be. 0:21:46.400 --> 0:21:51.960 But shouldn't they because advertising is their business, that's what 0:21:52.080 --> 0:21:55.120 Meta does. It's where the vast majority of the company's 0:21:55.119 --> 0:21:57.720 revenue comes from. So it seems to me that the 0:21:57.720 --> 0:22:02.800 company absolutely should prioritize that it ensures that its core 0:22:03.000 --> 0:22:07.560 business is legal. Maybe I'm being unreasonable here. I don't 0:22:07.560 --> 0:22:11.560 think so, but maybe. David Shepherdson of Reuter's has a 0:22:11.640 --> 0:22:16.359 piece titled TikTok faces crucial court hearing that could decide 0:22:16.400 --> 0:22:19.760 fate in the US. So you might remember that lawmakers 0:22:19.800 --> 0:22:22.439 here in the United States decided that TikTok would have 0:22:22.520 --> 0:22:26.040 to divorce itself from its parent company, Byteedance, which is 0:22:26.160 --> 0:22:29.040 headquartered in China, if it is to be allowed to 0:22:29.160 --> 0:22:33.600 continue to operate in the United States, and subsequently, TikTok 0:22:33.640 --> 0:22:38.160 has argued that such a separation is technologically impossible. Now 0:22:38.200 --> 0:22:41.520 that's a claim I personally find hard to believe, though 0:22:41.560 --> 0:22:44.359 I do think any separation would require huge changes to 0:22:44.400 --> 0:22:48.000 how TikTok operates. It also said it's legally impossible and 0:22:48.160 --> 0:22:52.080 financially impossible. Well, next week, the US Court of Appeals 0:22:52.119 --> 0:22:55.240 for the District of Columbia will hear arguments about that 0:22:55.400 --> 0:22:58.560 legal side from TikTok's legal team, and they're saying that 0:22:58.840 --> 0:23:02.280 this law is unconstantitutional that violates the First Amendment, which 0:23:02.320 --> 0:23:04.679 is also known as the freedom of speech. Well, no 0:23:04.760 --> 0:23:08.080 matter what the outcome is of this particular case, I 0:23:08.119 --> 0:23:10.480 think chances are pretty good it's going to get pushed 0:23:10.560 --> 0:23:13.320 upward to the Supreme Court. Both the US Department of 0:23:13.440 --> 0:23:16.960 Justice and TikTok's lawyers have asked the Court of Appeals 0:23:17.000 --> 0:23:20.280 to render a decision by no later than December sixth, 0:23:20.600 --> 0:23:22.840 and that will be just a little over a month 0:23:22.880 --> 0:23:25.400 before the nationwide ban is to go into effect, which 0:23:25.440 --> 0:23:29.400 is on January nineteenth. That's assuming that, you know, there's 0:23:29.440 --> 0:23:32.960 no challenge to this, and it provides very little time 0:23:32.960 --> 0:23:35.040 for the Supreme Court to get involved to make a 0:23:35.080 --> 0:23:37.359 decision as to what might happen once it gets to 0:23:37.400 --> 0:23:40.600 the Supreme Court. That beats me, because I mean Donald 0:23:40.600 --> 0:23:45.520 Trump appointed three Supreme Court justices during his presidency and 0:23:45.920 --> 0:23:49.680 At that time, Trump was actively trying to ban TikTok 0:23:50.040 --> 0:23:53.440 by executive order. That was a big deal. However, since then, 0:23:53.520 --> 0:23:57.399 Trump has flip flopped about TikTok. Whether that has anything 0:23:57.440 --> 0:24:01.600 to do with a billionaire TikTok and who made significant 0:24:01.600 --> 0:24:05.720 campaign donations to the GOP, I can't say. But certainly 0:24:05.760 --> 0:24:09.280 Trump's own explanations of well, you know, kids really like it, 0:24:09.480 --> 0:24:12.119 that doesn't seem to be a compelling reason for his 0:24:12.760 --> 0:24:16.320 one to eighty degree turn on TikTok. Anyway, I have 0:24:16.480 --> 0:24:20.480 no clue if the current Supreme Court would side more 0:24:20.560 --> 0:24:24.120 on what past Trump said back when he was president 0:24:24.280 --> 0:24:26.240 or what he says now when he's trying to be 0:24:26.280 --> 0:24:29.520 president again. But that decision will happen after the election, 0:24:29.800 --> 0:24:33.119 so maybe that could have an impact, I have no 0:24:33.200 --> 0:24:37.320 way of knowing. Yesterday, the Food and Drug Administration here 0:24:37.320 --> 0:24:39.880