Week in Tech: Careless People, Mindful Chatbots

Published Mar 21, 2025, 9:00 AM

How do you calm down a chatbot? This week in the News Roundup, Oz and producer Eliza Dennis dig into the book that Meta doesn’t want you to read, chatbot reactions to stressful stimuli, and the new home of Pokémon Go data. On TechSupport, 404 Media’s Joseph Cox discusses a tool with surprising data scraping capabilities that is used by US agencies like ICE.

Welcome to tech Stuff, a production of iHeart Podcast and Kaleidoscope. I'm os Valoshan. Today will bring you the headlines this week, including unearthed executive drama meta. Then in the tech Support segment, we'll talk to four of Form Media's Joseph Cox about the aggressive data scraping carried out by an Ice contractor all of that on the weekend. Tech is Friday, March twenty first. This week with covering everything from the future of AI assists science to gaming, to anxious chatbots and executive upheaval in Silicon Valley. Eliza Dennis, our producer, is here with me. Hello us, So let's start off with some meta drama. The former director of Global Policy for Facebook, Sarah Wyn Williams, wrote this memoir called Careless People, a cautionary tale of power, greed, and lost idealism. The phrase careless people, I believe, is a reference to the Great Gatsby.

It is, indeed, and it is not subtle. Fitzgerald referred to his wealthy characters as careless people who quote smashed up things and creatures and then retreated back into their money.

You know, it's interesting because people talk about this time as a new Gilded Age, when Williams obviously alluding to that, but also her allegations against her former bosses meta range from the geopolitical to the sexual. There are multiple allegations of sexual harassment in the book, and some involve the form of Facebook COO Cheryl Sandberg.

Cheryl Sandberg of lean In fame and if I remember correctly, I believe that's essentially about being a girl boss at home and at work.

Well, when Williams says that, in addition to writing Lean In, Sandberg had an inappropriate relationship with her twenty six year old assistant, and that this employee, a cooting to the book, was quote very conscious of the benefits of being Cheryl's little doll, as she calls it, and having Cheryl tell her she loves her right.

And to my knowledge, Sandberg has yet to comment on any of these allegations.

Beyond sort of personnel relationships. Win Williams also alleged that Mark Zuckerberg and Facebook were willing to do just about anything to make their platform available in China and reap the benefits of that enormous population of potential users. This reportedly included plans to comply with the Chinese Communist Party's censorship requirements and to appoint a chief editor. His job will be to take down unacceptable posts and share user data with the Chinese government. Now this is according to the book, but also to a whistleblower complaint that win Williams filed with these Securities and Exchange Commission. METTA is taking legal action against win Williams. They argue that she's being paid by anti Facebook activists and that the book violates a non disparagement agreement that she's signed during her Facebook days. On the non disparagement issue, an arbitrator recently ruled in Meta's favor, saying that Wim Williams can't promote the book, but you can still buy it Careless People remains on sale. Our next headline is a little bit more uplifting, and it's about a startup that wants to use AI to speed up scientific discovery. So the New York Times are about a company called Lila Sciences. The headline is the Quest for AI Scientific Superintelligence, and the story is about Lilah, whose mission is to quote soul humanity's greatest challenges. Has been working secretly for the past two years on an AI program trained on scientific data, scientific process, and scientific reasoning. And then they let this AI software loose in an automated lab to run its own experiments.

I don't know exactly how this works, but I love the image I've created in my head.

Likewise, and in fact, the results seem promising. With the help of a few human scientists to assist, Lila's AI made novel antibodies to fight diseases, and developed new materials for capturing carbon in the atmosphere. And it made these discoveries quickly. Studies that would normally take years reportedly took months.

Using AI to speed up the scientific process is very popular these days. I just want to shout out the Nobel Prize in chemistry went to two scientists who used AI to predict and create proteins.

Yes, that was the alpha fold initiative coming out of Google's Deep Mind. Lila Sciences is attempting to systematize AI driven scientific discovery, and their goal is to create, as I mentioned, what they call scientific superintelligence. There are early signs of success, but the times reported or perhaps worn, should I say that since Lyla's been operating in secret, outside scientists have not been able to evaluate its work. But Liila Science's secret is out and they're getting more lab space, and I'm sure we'll hear more about them soon. In a fun aside, George Church, the Harvard geneticist also behind Colossal Biosciences, the company that recently created the Wooly Mouse, recently joined Lilah as chief scientist.

We love the Wooly Mouse. I can't wait to see where Lila Sciences goes.

Speaking of going place to Eliza, do you remember Pokemon Go?

I definitely do, though I admit I never.

Played well for Media reports that Niantic, the company that made Pokemon Go, is now selling the game to scope Lee in a three point eight five billion dollar deal. Scopely is a subsidiary of Savvy Games, which is wholly owned by Saudi Arabia's public Investment Fund. Now, if you're wondering why you should care, it's because Pokemon Go uses players' locations as part of the gameplay. In fact, this deal will pull in location data from one hundred million players. So what happens to all that data?

It's a great question, do you know?

It may be that it's a three point eight five billion dollar question. We don't know, but Scope he assured for or form media that user data will remain private and handle it on US servers. But Niantic and Scope Lee say they have a partnership that goes beyond Pokemon Go. Niantic is building a quote large geospatial model using Pokemon Go data and spinning it into a separate business called Niantic Spatial. Scope Lee has invested a further fifty million dollars in this nascent mapping business, and the idea is to make an AI model trained on millions of geolocated images from around the world from places that only pedestrians can go. So unlike data from say Google street View, Pokemon Go's view is not limited by what you can see from a car on the road. Right.

This is so wild to me how data from a game just might one day help robots navigate the world or something like that.

Something else that's wild, though, is a new study shows that chat GPT can show signs of distress when shown disturbing material.

So, as we've been told over and over that we're not supposed to answer promortize chatbots, how am I supposed to do that?

With this news?

It seems like it may be anthropomorphizing themselves, but Chatchipt models have been integrated into therapy apps to give people advice or be a source of support. And researchers conducted this experiment where they first had Chatchipt read very boring material. I mean, we're talking about vacuum cleaner manuals. But then the AI therapist that integrates Chachipt was given a quote traumatic narrative that described a scenario like an intruder breaking into an apartment or a soldier in a firefight, and after each activity, the chatbot was given this questionnaire called the State Trait Anxiety Inventory, which is used in mental health care to basically analyze anxiety. The research has found that after seeing a traumatic narrative with the soldier, Chatchipt answered the questionnaire in a way that signaled severe anxiety.

I genuinely don't know how to respond because I feel bad for the chatbot.

Well, in that case, you'll be pleased to hear that. The researchers then gave the chatbot information on deep breathing and mindfulness exercises, and the chatbot's anxiety score came down to vacuum manual levels. Wow, so time to take a deep breath for all of US, and I'm going to run through the rest of this week's headlines. Google struck a deal to buy cybersecurity startup Whiz for thirty two billion dollars, making it the biggest acquisition in Google's history. Wiz was founded only five years ago in twenty twenty, and it scans cloud service providers data for potential security risks. If the transaction is approved, Whiz will join Google Cloud, which plays a critical infrastructure role in supporting Google's AI ambitions. Our friends at four or for Media reported that human generated content is getting drowned out by aislop. AI generated videos are cheap and easy to produce, and they are getting views of The video of this horrifying creature with a giraffe head and a spider body has over three hundred and sixty two million views on Instagram reels. AI generated content is apparently overwhelming algorithms and turning social media feeds into fever dreams. Finally, tech Crunch reports that people have discovered Google's newest AI model, Gemini two point zero Flash, has a controversial use case, convincingly removing watermarks from stock media outfits, including Getty images. This new feature from Gemini is labeled as experimental and quote not for production use. I'm sure everyone will follow those rules. Coming up, we hear from four or four Media's Joseph Cox about a tool US agencies used to scrape social media data. Stay with us. If you're listening to this show, you're probably someone who cares about or at least pays attention to data privacy. Every day, our online movements being tracked in some manner by companies who do targeted marketing, social media platforms, and government agencies. For Americans, this type of tracking has gone on in the background of people's digital lives for decades. Since two thousand and one, the US government has ramped up surveillance efforts, including tracking americans phone calls, emails, search histories, with the stated goal of preventing terrorism, and as technology has evolved, the idea has lived on. In recent years, agencies like immigration and Customs enforcement have been working with one company in particular to get the data they want. Here to tell us more is four or four Media's Joseph Cox. Joseph, welcome back to the show.

Thank you so much for having me.

So you wrote an article which as soon as I saw the headline, I wanted to have you on the show. The headline was the two hundred plus sites an ICE surveillance contractor is monitoring. Who is the contractor, what are the sites, and what's the story?

Sure? So the contractor is a company called Shadow Dragon. I seriously doubt many people have heard of this company relatively obscure, but over the past months and years they have become sort of a go to contractor for not just immigrations and Customs enforcement, but the DEA, the State Department, and other US government agencies. And what they do is basically something called open source intelligence OSANT, or social media monitoring, right, And typically we used to think that was monitoring sites like Facebook, Instagram, Twitter, all of those sorts of things, and shadow Dragon does that as well, but as you say by reading out the headline, there are many many more sites as well. Essentially, any sort of open website that you may perform some sort of activity on there's a chance it's on this list.

So all day, every day, we create these digital breadcrumbs, and essentially there's somebody hoovering them up to process them and repackage them in some cases for the US government.

So what the shadow Dragon tool does It is called social net the specific tool. It allows the automation and streamlining of searching those sites and social networks for a target. So let's say I'm in ICE or another government agency. I'll enter a phone number. It will then take that and see, well, there's that link to accounts on Facebook, or this site, is this username available on this social network or then this obscure platform as well, And some of these sites are much more revealing than others. Again, it's not just the big social networks. There's stuff like hiking apps on here as well. So if you want to find out where somebody's located, oh, they take a load of hikes in southern California or something like that. I think people would be, on one hand, not surprised, because, as you say, people listening probably care about data privacy see well. On the other I think they might be surprised in that you think it's just this sort of benial, insipid information you're putting out there, But when it's an aggregates pulled by a tool like this, it can be really really revealing. It almost becomes more than the sum of its parts.

Jose, can you help me understand because obviously everyone's been very focused on the case of Mahamud Khalil, the Columbia student with the green card who looks like he may be deported. Is this tool about identifying people whose profile, who's on activity contravenes the current administration's priorities or is it more about, hey, this person is a target, help me build a stronger case against them.

So at the moment, it's almost certainly the latter, as in ICE has access to this tool, and specifically Homeland Security Investigations as well HSI part of ICE, and they're able to use this tool for whatever part of their mission they wish to do so. Axios also reported somewhat recently that the State Department is going to be using AI to scan the social media feeds of students to see if it can detect what it deems as anti Israel or pro mass views. That's not to say Shadow Dragon is being specifically used for that. We don't know exactly what TALL the State Department may or may not use, but it absolutely sits in that context. You're right in that there are these deportations happening of various people, students and then other people visiting the US, and Shadow Dragon is very much at all there is available to US agencies if they do wish to use it for their mission.

How did you first hear about Shadow Dragon.

I first heard about shadow Dragon probably a couple of years ago at this point, because I was actually covering other companies in this space. So I've covered the sale of location data for a long time. I've covered the gathering of social media monitoring, and I got sent some email that showed that ICE was moving from one provider, a company called Babble Street, over to shadow Dragon, And the actual reason was because it was cheaper. At the end of the day, even when you're talking about surveillance systems, sometimes it's just about you know how much it actually costs to one of these agencies. So that move happened a while ago at this point, But then every so often I go back and check, well, his shadow Dragon got in a new contracts, and sure enough, every few months there'll be something in there. And they did just continue this contract with ICE as well. So it started very very quiet, but now they've absolutely become sort of a pretty it seems reliable solution for the US government.

So the company is shadow Dragon, but the product is Social net can you just explain a bit more detail how it actually works.

Yeah. Sure, So somebody will log into social net and they get the sort of graph interface of all of these dots and lines. Honestly, it almost looks like the stereotypical like hacking or surveillance move right, And of course that's funny on one end. On the other, I don't know, it's very easy for an analyst to process. I think they have this interface in front of them now, then enter some information, maybe somebody's name, they email address, their phone number, they use the name, they can click a few buttons and it will then search all of these different sites for information related to that. Did this person us a username on this dating site? Did they use the same username on this fetish website for example? That's hypothetical, but those are some of the websites that are included in this tool. It's not just the mainstream stuff. Now, an analyst could do that manually if they wished, but that would require a lot of time. And this streamlines the entire process of mapping out someone's online profile, their relationships, and potentially some of their movements as well. Well.

And how do you get this list of the two hundred sites that shadow Dragon is pulling from, and can you give an example of what some leative sites are.

Yeah, so I can't go to too much detail, just beyond you know, we obtained this list and then I've verified it. But when I first got the list, I was quite blown away by the length of it, really, and I started putting all of those sites into sort of buckets. So you'll have ones that focused on hobbies like I mentioned the hiking site all Trails, and a book, sort of fan website. Even chess dot com is on there as well, do a lingo. You have payment ones like cash app and PayPal, communication apps like discorder, WhatsApp. Now they're all going to have different data in different degrees available to the tool. Like there's absolutely no indication that it has access to, you know, private messages on any of these services, and I don't think that's the case. It's purely publicly available information. But again, people may not know what they're exposing in advertently or not on one of these sites.

And these sites that you mentioned, I mean, are they sort of collaborating with shadow Dragon? Do they know this is happening? What is the range of responses from these sites? That shadow Dragon is pulling from, so.

They're not collaborating in that shadow Dragon is going and basically grabbing the data. But there is still great variety in the responses from the companies. I actually think the most illuminating one and the most helpful one came from chess dot com, who told me on the record that, you know, we didn't know about this. We don't like it. Any sort of data scraping has to be done within the law, and if it's in within the law, it's okay, but generally we don't like it. And then you have the big companies like Meta and Snap and all of those sorts of people just pointing to their terms of service saying we don't like people scraping our website. And then I actually had a bunch who never got back to me. Essentially, even when it seems like a pretty I'm not a pr professional, but it seems like a pretty easy to just reply and say, here's our terms of service, we don't like scraping. But yeah, a lot of companies didn't even acknowledge the request for comment.

On this one. So there's really these somebody don't really have a choice. I mean, their choice is to sue Shadow Dragon for violating their terms of service, which is an unlikely thing to happen because it's not directly harming their business unless there's a kind of groundswell of consumer complaint.

Yeah, I think that's fair when it comes to the lawsuits, like there would have to be some sort of tectonic shift around that. We have seen WhatsApp, for example, sue NSO Group, which was or is a company that delivers packing tools to iPhones. But that's much more aggressive. That's much more active that the.

Group behind the Pegasus product, that spyware product that was able to be covertly and remotely put onto people's.

Phones exactly, and WhatsApp has sued them. But that's sort of you know, one company versus another company. It's not as broad as this. That being said, you know, companies try to scraping all of the time. They don't like it when you make multiple accounts, for example, they don't like it when you use a VPN to log into Facebook. I can't tell you how many times Facebook has closed my account and they have to make a new one every single time. So they do have all sorts of measures to stop weird, potentially suspicious or just unusual activity, and this would fall into that. It's just a question of whether these services can actually detect that specifically shadow Dragon or not, or of course whether they actually want to or not, because if they started doing that, I don't know, US government agencies might not be too happy.

When we come back how shadow Dragon fits into the largest surveillance ecosystem used by the US government. To stay with us, have any specific examples of how shadow Dragon has been used in the real world.

We only have what shadow Dragon says publicly in its marketing material because the agencies are very tight lipped or simply don't respond when talking about that sort of thing. Obviously, when it comes to ice in HSI, there's going to be an immigration component there, so I feel it's safe to assume there's a chance it's being used for those. And then some of the concrete cases that shadow Dragon has mentioned is fighting child exploitation, potentially even mapping out the opioid crisis as well. So don't get me wrong, there are clearly useful use cases for this technology. It's just that people don't really know it's going on necessarily, or that you know their government is buying this tool or again that all of this sort of data is publicly exposed.

And how practically does the US government actually use this tool, and let's use ICE as an example. Is it like there's like a shadow Dragon consultant and the ICE agent sends their request to the shadow Dragon employee and the Parodragon employee generates a report or is this more like a kind of plug and play platform.

It's much more a plug and play platform, almost a software as a service tool. The agency will buy licenses to use it for a year, two years, whatever. They'll log in just like you log into any sort of internet service or piece of software, and then they can use it within the parameters of what they've been allowed to do. And that's probably good for the agencies. I don't know if they necessarily want to give up information about who they're targeting to a third party, like a contractor, but it absolutely puts the honest on ICE or whoever is using it to be just very very careful with this tool as well. And I think just more broadly, we've seen repeatedly over the years that not specifically with shadow Dragon, but just other social media surveillance tools, we've seen authorities use it to monitor protests, and there was an article just in the Intercept recently that I think it was the LAPD use as similar tool to monitor protests there as well. So this is very very long running. Social media surveillance has been going on for many years, but now the playing field has expanded where it's not just the big platforms, it's every platform.

How does shadow Dragon compare to pan andeer as a technology?

So Palenteer is also used by government agencies and some private companies to link all sorts of disparate data that could come from various places together, where shadow Dragon is much more about We're going to go out and pull the data for you. It's not just about the oh look we're shadow Dragon and we linked this phone number to this Facebook profile or whatever. It's that shadow Dragon is making what I presume would be pretty reliable technology to go grab data from these websites. And that's not you know, it's not the hardest thing in the world, but I don't think it's necessarily trivial. Even that shadow Dragon is going to have to custom code all of these small little tools that go and grab that data from these websites, and presumably if you're selling to the US government. It has to be quite good, quite reliable, So there's a sizeable amount of work going on here to we even grab that data in the first place.

How much do you think about the scope for abuse here.

I absolutely do think about it, and it would be very much a case by case basis. It would depend on the agency, the individual official, all of that sort of thing. What makes this sort of maybe not a form issue, but a quite complicated one is that, yes, it may be violating terms of service of individual websites, but like, it's not illegal. There is no Fourth Amendment search going on here. It's not like the agents have kicked down the door and they're searching somebody's house or they're breaking to somebody's phone searching it there. Technically and legally, it is all public data, so there's not much room for legal abuse because it's perfectly legal generally speaking. That being said, agencies can't just rely on that idea that well, it's all public so it's fine. It can still be abused by certain people in the unfortunate correct context.

Suit me out a little bit. How does this fit into the kind of wider surveillance ecosystem and how it's being used by the US government, And I think it's a timely moment to discuss that, given that this Macamu Khalil case is kind of top of mind for so many people.

Yeah, so I see it on the lower end of sophistication when it comes to US surveillance tools. You imagine, on this lower end you have social media monitoring tools like this, Somewhere in the middle you'll have wire taps. After that, maybe you have direct hacking tools, and right at the top you have I don't know, NSA, mass surveillance programs like Edward Stoden Reveal or whatever. And even though social media monitoring like this maybe on the lower end of the spectrum, I think it's still vitally important for people to at least know about it so they can decide whether they care or not. And maybe they don't, and that's absolutely fine. But the key thing here is that these sorts of tools are also available to local law enforcement sheriffs, state police as well. Is not purely a federal agency thing. So while the federal agencies may only get the very sophisticated tools, basically everybody gets social media monitoring and that does open up the potential room for abuse as well, again, depending on what the individual person using the tool is doing well.

Joseph, we're both we're both Brits. We didn't live through the Second World War, but sure we're both familiar with the phrase loose lip sync ships, and so yeah, be careful what you put out there exactly.

I think there is the message that people should take away. Joseph, thank you, Thank you so much.

That's it for this week for tech Stuff. I'm mas Vlosen. This episode was produced by Eliza Dennis and Victoria Domingez. It was executive produced by me Carrot Price and Kate Osborne for Kaleidoscope and Katrina Norvel Fireheart Podcasts. Bahid Fraser as I Engineer and Kyle Murdoch mixed this episode and wrote our theme song. Join us next Wednesday for tech Stuff The Story, when we'll share an in depth conversation with one of the most interesting people working in and around tech. Please rate, review, and reach out to us at tech Stuff podcast at gmail dot com. We want to hear from you.

In 1 playlist(s)

  1. TechStuff

    2,443 clip(s)

TechStuff

TechStuff is getting a system update. Everything you love about TechStuff now twice the bandwidth wi 
Social links
Follow podcast
Recent clips
Browse 2,440 clip(s)