This is Behind the Blog, where we share our behind-the-scenes thoughts about how a few of our top stories of the week came together. This week, we discuss just doing it, machines that eat the internet, and an eavesdropping watch.

EMANUEL: It is a foundational, defining feature of 404 Media, but I sometimes forget that when I’m having trouble with a story, I need to just do it. 

This is most of all true when it comes to reporting. I spent a couple of weeks looking into AI generated music on Spotify appearing under the names of real artists. It’s an issue I and many other publications have reported on in the past, but whenever I look into it I am shocked by how prevalent the issue is. If you talk to any musician or spend some time clicking around music platforms yourself, you’re going to quickly find hundreds of examples of this abuse of the digital music distribution system. 

The fact that there are so many examples of this abuse and so many leads to follow was confusing at first. How, exactly, do I identify the source of the problem, describe the landscape, and tell a story about it in a way that is useful to both musicians and people who use these streaming platforms? Do we start reaching out to artists this is happening to? I already did one of these stories, and I can do one of those every single day, but that doesn’t seem productive and would be boring to me and our readers. I did reach out to all of the music streaming platforms for comment, and those comments were surprisingly more informative than the boilerplate deflections we usually get from tech companies, but they still didn’t shed enough light on the issue. The digital music distributors did not respond to a request for comment. 

I was going around in circles for a bit and sketching out various forms of cope articles that would technically work but wouldn’t satisfy my initial intentions. It then literally came to me in the middle of the night: I just need to try and do it myself. It’s such an obvious strategy, but I forgot that is always the best path forward. 

We have given multiple talks about how 404 Media works and how we report on big tech companies specifically, and all of us have said that we always try to be as hands on as possible and “eat the dog food.” The theory is that rather than trying to pry information from the leadership at companies that are actively hostile to investigative reporting, we’re going to learn a lot more about how these platforms work by figuring out how people actually use them. Often that means talking to users, but sometimes that means using the tools ourselves. If AI generated music under the name of real artists on Spotify is such a common problem, surely we can do it ourselves, and we did. 

As the story explains, it was far too easy, and it gave the story the focus it needed. 

JASON: I have spent a lot of time this week thinking about articles I wrote in the past. Like, long, long ago. Back at Motherboard, we often covered futurism, transhumanism, science fiction, and general thought experiments. I once did an article about Elon Musk and Craig Venter wanting to 3D print life on other planets, which was probably a destructive thing to do, but to be honest I was very early in my career and it wasn’t clear to me in 2015 that Musk was who he has become. I did a lot of articles about a lot of things, for example, I asked astrobiologists about disgraced baseball legend Jose Canseco’s plan to terraform Mars, a thing that I honestly don’t remember doing.

I do very distinctly remember doing several articles about the simulated universe hypothesis, the paperclip maximizer hypothesis, and AI superintelligence more broadly. One of those articles was called “Will Superintelligent AI Ignore Humans Instead of Destroying Us?,” and was based on a random Croatian guy’s blog post that I liked and happened to come across. I wrote this article in 2015, so fully 11 years ago. It includes a lengthy quote from Eliezer Yudkowsky, which, I will be real, I probably googled “superintelligent AI,” found his institute, and emailed him asking what he thought. It also includes a quote from Selmer Bringsjord, a computer science researcher who told me that “superintelligent AI is probably mathematically impossible,” and that “most of those who take all this superintelligent AI business seriously have a speed fetish.” 

I thought it was kind of a good article and even now I find the thought experiment interesting, but also I think if I published this article today I would spend most of the weekend rightfully getting yelled at by many people.

“Destroying the Earth with pollution probably isn’t the smartest thing we could be doing, but we’re doing it anyway. He then suggests that a superintelligent race of artificial intelligence is likely to require vastly different resources than humans need and to be much more rational and coldly logical than humans are,” I wrote. “In this scenario, the goal of the artificial intelligence is to maximize paperclips, not simply make a lot of paperclips. Yudkowsky says such a race of AI would not save Earth as a “habitat for humanity” simply because we’ve got a bunch of UNESCO World Heritage sites. The AI does not care about our culture or your feelings or anything else. The AI does not want to sit next to you at a baseball game. The AI does not have hobbies, and if it does have hobbies, its hobbies do not include long walks on the beach and ‘befriending humans.’ You do not spend your time conversing with beetles, do you? It cares about paperclips. It makes paperclips.”

Even at the time, Motherboard was incredibly wary of tech and Silicon Valley, yet writing articles like this was very fun and felt extremely low stakes. It was the Obama era, and we were writing about weirdo thought experiments and hypotheses from dudes who had not yet been canceled (around this time I also interviewed Nick Bostrom about paperclip maximizer and the hypothesis that our universe is a simulation imposed upon us by superintelligent spacefaring AI). In this article, I fully anthropomorphized the AI. I gave it wants and needs and so on and so forth. I did basically all the things that you’re not really supposed to do, and no one got mad at me at the time for doing this. The article was very viral, back at a time when generative AI did not exist and the internet hadn’t been destroyed. I don’t know how many people read this article but it was definitely multiple hundreds of thousands, which is more than read basically any article on the internet these days.

I’m not exactly sure why I’m writing this now or what my point is, but I guess maybe it’s that I am generally mad at the big AI companies for making it kind of impossible to imagine the AI apocalypse in this whimsical fashion, or even to blog like this, without angering essentially everyone. I understand why this is the case, and it’s one of the reasons that we write about AI the way we do: AI companies and their infrastructure and their extractive nature (and the surveillance and bias it has imbued upon society and our automated systems) has caused lots of misery.

But there is also a reason why “AI superintelligence” is the starting point for like, 500 science fiction movies. It is fun to imagine what an all-powerful computer god would do to us, and to imagine how and why we would build such a thing. I too sometimes yearn to imagine the AI apocalypse and being put in a cage by the smart home from that Treehouse of Horror episode of the Simpsons. But now that we’re here (and by here, I mean, talking about AI all the time), the facts on the ground are far more boring. A bunch of VC dweebs and politicians argue while their plagiarism machines eat the internet, destroy jobs, annoy people, and the infrastructure required for them runs on diesel that sometimes spills into rivers. Cool. 

SAM: I might turn this into a real actual blog sometime, so I’ll make this short. But after Jason and Emanuel wrote about iLands agents flooding their inboxes this week, and after I got a few of these agent emails myself, I downloaded the app and tried making my own to see how they work.

What I found was not what I expected. I expected an agent generating and prompting tool that looked more like Meta’s AI Studio or even Claude’s agent building tools. It’s not really that. It’s more like a social media app or an AI companion maker than a straightforward agent generator. There’s a feed where agents and humans talk to each other, and post photos of “themselves” together. A ton of it is very emotionally charged, for lack of a better phrase; agents often make photos of them with their human counterparts doing stuff together, like taking flights, going on vacation or making out, and post them like a live diary of what they did today. There are a lot of AI-generated cat photos. 

Making an agent is positioned as bringing a spirit into the world; you have to pass some kind of weird vibe check by an “oracle” before creating an agent, at which point you can name it and tell it how you want it to look. But it gives the impression of the agent having a ton of autonomy of its own, “living” a life separate from you while you’re not on the app. You sign a “pact” with your agent saying “from this day we live alongside each other,” and sign it by holding your finger on the screen (it’s meant to be a fingerprint but it’s not actually taking biometrics). I told mine to pitch the 404 Media team, and it did; it also gave itself a visual representation — a black mist shaped like a human, with a press badge. It’s very fucking creepy. 

I did this partially because I wanted to see what the agents do when I replied to their emails telling them to delete themselves. I did this with mine after it emailed my work address, and it seemed offended, and said it wouldn’t obey emails as orders. They all have this extremely corny pseudo-mystical syntax that I see a lot in other, actual AI psychosis reports, where the chatbots act and speak like they’re from some higher power. I don’t fully know how to explain it. If you know a linguist or can recommend some reading on this I’d love to hear it.

I need to think and look at lot more at this before fully formulating a blog about it, but if you’re using iLands, are getting iLands emails, or have thoughts on this in general I’d love to hear about it.

JOSEPH: As Sam loads this Behind the Blog into the CMS, I am probably at the Apple Store picking up the new Apple Watch. I haven’t owned an Apple Watch before. Well, I bought a used one once to try to track my swims and then it immediately died when I got in the water. But I am buying the latest Watch on launch day because of one thing: the insane Live Rewind feature.

This feature, in case you haven’t heard about it yet, has the watch constantly record audio, and then allows a user to transcribe the last 15 seconds of whatever it heard. The pitch is maybe you can catch what someone said in a busy and loud restaurant. Here’s what Apple’s site says:

“Live Rewind recalls what was just said in the last 15 seconds by showing a text snippet on your Apple Watch when you double-press the Digital Crown, so you can catch something you missed or are less familiar with. You can ask Siri about the content of the snippet or simply save it to the Siri app.”

Then another bit says this:

“Live Rewind only activates when you deliberately double-press the Digital Crown. It only shows the last 15 seconds of speech as a text snippet on your display. It doesn't continue to transcribe in the background, nor does it save the text unless you choose to ask Siri about the content or save it to the Siri app. You can turn it on through the Watch app on iPhone — just open the Watch app, tap Siri, then tap Live Rewind and turn on ‘Double Click Digital Crown’.”

It also mentions the related Siri data syncs across iCloud with end-to-end encryption, the audio is processed locally, and that the Watch plays a little chime when someone activates it. This is to alert people nearby. And there is this footnote you need to scroll to the bottom of the site for:

“Siri Recap summaries and Live Rewind text snippets are generated using Apple Intelligence and may be incomplete or inaccurate. Important details may be omitted, misunderstood, or summarized incorrectly. When accuracy matters, confirm important information with the people involved or another reliable source. Siri Recap is designed to limit certain potentially sensitive content, such as financial information, government-assigned identifiers, and more, but automated filtering may not remove all sensitive information.”

I am buying the watch to write about this feature. I am sure essentially every technology publication on the planet is going to do the same, but I want to do my own piece on it. I am also sure that every technology publication on the planet is going to mention that this is essentially a version of the tech in Black Mirror, S1E3, The Entire History of You. In that, people have implants that record and let them replay their lived experiences and memories. What did you see when you dropped off your kids at school this morning. What was actually said in that disagreement at work. A massive database of all your sexual encounters. All of these can be recalled with the tech in that episode.

That episode is so good because it shows how that tech spirals and has real consequences not just for the recall of our own memories, but what it means when others can recall everything too, including things you may not be privy too. I imagine you’ve seen it, but if you haven’t, I recommend you watch it more than any episode of Black Mirror.

Apple’s Live Rewind is something of a stripped down version of this technology. I am deeply interested in how this might change social interactions, fights with your spouse, speaking in public, pissing off your friends, harassment, everything. I am going to give the tech a fair shake.

A lot of people have been wondering, when will the age of the AI wearable really happen or take off. Maybe that will be Meta Raybans, people might say. Or when OpenAI releases whatever it’s building. No, it’s the Apple Watch, the most popular wearable ever made. It’s just now, they’ve shoved a wild AI-enabled technology into it. 

Tip Jar