Web4/Heaven—The Simulation
[This is a thought experiment / ranty idea. I don’t necessarily agree with everything I say hear but try to think the idea through to its end. Inspired by the hypothetical practice of heaven banning, hence why I call it Heaven.]
Web4 is the internet that is completely (or at least mostly) replaced by AI generated content. Any content you consume or interactions you have are replaced by AI: GPT-30 generates responses to you comments on Web4-Reddit and posts you favorite tweets on Web4-Twitter. Dalle-10 generates any image you might choose to look at in the internet and VideoDalle creates your favorite Web4-YouTube videos. Every internet user has their own version of the internet, completely tailored to them, perfectly adjusted to maximize everyone’s enjoyment of time spent on the internet, learned from the users’ individual feedback. Why would you need real Reddit posters or Instagram influencers, when Web4 Instagram contains only images that are perfectly generated specifically you?
You might say, posting a comment under a YouTube can provide helpful feedback to the content creator, but your comments under Web4-YouTube videos can also influence what videos VideoDalle will show you in the future.
Some people think we already live in a simulation — this is that but for the internet. Even if you know that your whole online experience is simulated, how many of your online interactions ever make it into the real world? Does it matter whether there are actually real people behind those?
Sure, there is still space to get posts from your real life friends on Web4-Facebook. Maybe you even choose to have generated content be interspersed with real people, this way you can still make real life friends on the web. Maybe fake YouTube influencers will tell you about real-world news stories, courtesy of MultiModalGPT-20.
No more negative interactions on the internet (unless you want some). Presumably content could even be optimized to fulfill some kind of specific goal you set: Learning, intellectual growth, limiting time spent on the internet.
Having access to the entire internet and sufficiently capable content generation AI systems should be sufficient to create a simulated version of it.
Is this a dystopian vision? Do you take the red pill or the blue pill?
I’m planning to do this with LW/AF. Full multimodal Web4 would be fun—I hope it happens, at least somewhat, before the end.
Some prior work:
GPT-3 simulations of Google search and Wikipedia I made a while ago
This paper made GPT-3 subreddit simulations for refining “social computing designs”
Of course, GPT-4chan
I should clarify that by “I hope it happens” I mean “I hope we get to experience this conditional on a fixed outcome”, not “this would be good in expectation”.
“Your own version of the internet tailored to you” really undersells it. You program the boundary conditions of a massive multi-agent simulation. You can go down any not-yet-existent rabbit hole (latent multiverse branch). This encompasses all fields of research; it encompasses code. Now replace “you” with “anything that writes”. Web4 would accelerate timelines if not directly foom. A self-programming simulation hallucinated by superhuman AI (GPT-30) is even more of a ticking time bomb than base reality.
If this is done using RLHF I particularly expect it to end in wireheading / pathological mode collapse / misaligned takeoff.