You know when you make a photocopy of a photocopy, the quality degrades? The more generations removed from the master, the worse the quality gets with each iteration, right to the when it becomes barely legible.

Well, turns out that the same thing is happening with Artificial Intelligence (AI). The first AI models were trained on human-generated material. Lots of it. 

But now, the internet is starting to get flooded with AI generated stories, images, and other content. And the AI models are being trained using this new material. 

Do you see the photocopy analogy yet? Basically, the more AI is trained on AI-generated content the worst it starts to become. This new AI-generated content is starting to muddy the digital waters with no reliable way of determining real from fake, so AI companies could soon find themselves hitting a dangerous wall. This is partly due to neither AI companies or their users are currently required to put AI disclosures or watermarks on the AI content they generate — making it that much harder for AI makers to keep synthetic content out of AI training sets.

A growing pile of research shows that training generative AI models on AI-generated content causes models to erode. In short, training on AI content causes a flattening cycle similar to inbreeding; in fact the AI researcher Jathan Sadowski last year dubbed the phenomenon as "Habsburg AI," a reference to Europe's famously inbred royal family.

Soon we will reach a tipping point on the quality of results that AI produces from us. The more AI generated content that is being used to train the models, the further away from the “original photocopy” we will get and the results will be gibberish!