(Re)Generative AI?

While reading still another article about the challenges of Generative AI (GenAI), legal and otherwise, it got me pondering the word “generative”. This had me thinking that AI may be “Generative”, but it’s certainly not “Re-generative” and unlikely ever will or even can be. If the foundation of GenAI comprises the Large Language Models (LLMs) from which it is trained, and we understand that those LLMs are digital content of varying types scraped from the internet, and that content source is created by humans, then what happens when humans are disincentivized to create any new content? No music, no art, no writing, no need … just prompts. Hmm. Yet, like any system on our planet, man-made or not, if no new energy is reintroduced into it then entropy sets in.

I can hear the detractors, “yeah, but the AI will create its own new content.” Okay, then maybe I’m wrong and there will be regenerative AI since all that remains for the GenAI to learn from is what it generates itself. Is that like “eating your own dog food”? I’m not sure how useful that sort of regenerative AI will be considering a key purpose of this technology is to help humans more quickly assess information to aid our decision-making. At least, that’s what I see as its main value. If that data is incestuously generated then there’s not much new from which to discern ideas. Probably nonsense content would be generated sending us prompters into equally nonsense choice-making (humans are already experts at nonsense choice-making).

I see it like a novice wheat farmer who year-after-year over decades plants, grows, and reaps wheat then wonders why over the years the yield dwindles while pests and disease arises. That wheat soil, like LLMs, needs to be regenerated with a diversity of biomass, else entropy sets in.

What’s my point? There is certainly usefulness in the things humans create, but they all have limits and need balancing against their impact because there’s no escaping natural laws.

As I said, just pondering.