Dead internet theory can essentially be boiled down to the idea that the entirety of the internet is fake. Bot activity and false stories dominate the whole thing in a big attempt to manipulate you the consumer/citizen. You are left haunted by memories of the "good old days" when the internet was fun and real.
This seemed laughable when the theory was introduced in 2016, even in the face of some evidence. That same year a security firm, Imperva, found bots to be 52% of all internet traffic. I don't think many took this hypothesis very seriously, even though we all had been seeing the effects of bots. It wasn't exactly the war against the bots that we envisioned as a kids. No terminators, just a bunch of social-media-driven-post-liking bots and algorithms designed to boost YouTube accounts.
Now, large language models have stepped into the arena and they are accessible. I'm on the home stretch of a bachelor's degree and the usage is rampant on campus. Sometimes answers are non-sensical, but mostly they are believable. If we'd modify and apply the old Turing test as, "if we can't tell if it's human or not, it must be human," there would be a whole lot of the internet that is seemingly not a bot because it could plausible be human. Most versions of this complicated matrix of word math is seemingly coherent and believable. Not only that, but the emotional believability is impressive.
It's getting sophisticated, but will it last? How might this play out?
Well, I think there is a tipping point. We might be in the golden age of large language models. Sure they can get more correct and refined, but eventually feedback will start to introduce itself as distortion in the system. Have you ever pointed a camera that was connected to a monitor at itself? It becomes an infinite loop of distortion. The little bit of randomness introduced through the screen begins to take over the screen as it looks like a vortex out of the old show Sliders. If you haven't seen that, think about pointing a microphone at a speaker that is playing the sound from the microphone. You get a nice ear piercing scream from this feedback.
AI feeds on the internet, but AI writes to the internet. Since we don't want to produce the same exact results with each query in the language model, we introduce a certain degree of randomness. Eventually in a feedback loop this randomness will start to introduce nonsensical stuff and start getting amplified. Maybe it won't look like a vortex of static or result in an ear piercing screech, but it could look like someone dumped a bunch of Spaghetti-Os on each page.
Still, we'll have these social media bots jumping around and liking posts/pages, but that's a much different and more complex type of problem. On Hive for instance, there are multiple forces at play. You want engagement, but you also want the undying loyalty of autovotes. For most users there is a mismatch between token rewards, and dopamine rewards from just BS'ing with people on the internet. You hate the bots, especially when other people use them, but also you love them because they bring eyes to your page and tokens in your wallet. The dopamine reward right now from tokens > from a fellow user's post on average. But if we're using typical social media as a model, potential monetary gain isn't enough either, as likes are manipulated as part of a psychological long game of consumerism/persuasion.
The internet might be dead for now, but I'm still here (extremely rarely), and maybe that's the key. Maybe we all need to step away more and let the bots sort it out. They're not terminators, but I'm no terminator either.