1 min read

The AI Feedback Apocalypse

There are some dates being thrown about. 2026 is the one MIT is throwing out there. I like that they’re making predictions on things, because then it’s not just me that’s wrong.

So imagine this scenario. AI text-to-image generators run amok even more than they are now. Image after image, by the millions or even billions now slowly and then soon very quickly flood the Internet as we now know it. Let’s say half or more of the images you see in a day are now generated. Yet, the generators don’t stop, they keep on going but now are referencing the images that are generated because they can’t tell. The generators then enter infinite recursion, doing nothing but feed off of each other.

I wonder if anyone besides AI generating companies will actually care. Eventually it will just be treated like ads, smudges of things here and there on the periphery of what you’re really doing, annoying and discarded like a wrapper of a really disappointing sandwich. Maybe the coming tsunami of generated content will just make us stop reading, looking and watching for the most part, abandoning the calls to righteous panic as much as the conspiracy lures.

I wonder how anyone could possibly think that anyone will treat a generated image any better than spam, which it sort of is. I wonder how people could repurpose this visual spam as they once saved the junk ads that came in the post for dogs to shit on or help get the fireplace going.