The AI Death Loop

Image: Badly edited by author

We can argue till the cows come home about whether Generative AI is good/bad, useful/not useful, pro-humanity/humanity-destroying. The answer likely lies in middle.

But one thing we can probably all agree is that AI models or programs interacting with other AI models or programs is a waste of time, resources and space, seemingly only capable of producing soul-sucking, bland, paint-by-numbers sludge that’s slowly infecting our every day lives. As we continue to hand over of our agency to AI, we’re creating more and more scenarios where AI ends up meeting another AI.

It’s happening everywhere; AI bots responding to social posts produced by AI bots, students turning in AI-generated work that’s then being marked by an AI program, job applicants using AI to write their application only to have it rejected at random by an AI software.

In most cases, the outcome is worse than it was pre-AI.

If you wonder what this looks like, here it is: a death loop of flattery, apologising, the shifting opinions — known as AI sycophancy — and endless back and forth that achieves absolutely nothing. (h/t

Frederick Woodruff

)

It’s a painful watch. The looping “you’re right” and “let me know” would drive you insane. The ridiculous regurgitation of what it just heard, the obvious pattern matching at play, the attempt to keep everything fluffy and airy. In what way is this an improvement? It’s not, of course — it’s a cost-cutting exercise.

A trillion dollars wasted on training models, buying chips, building data centres, all to make a customer service call even more infuriating than speaking to a human being. If this is what we can expect from agentic AI — basically autonomous AI that can make decisions, plan, and take action with minimal human effort — I don’t think it’s going to be the revolution it’s promised to be (or the one that AI companies are counting on to stay afloat).

Imagine handing off your customer service compliant to an agent, who then proceeds to have a 4 hour long conversation with another AI, both unable to make a decisive move, stuck in a death loop of agreeability, niceties and vibe checking.

Give me an uninterested human who barely speaks my language any day.