Stuck in Recursive Loops — Economic, Intellectual, Personally?
By KW Norton
In a recent X post (read it here) we have evidence of what is sketched out for economics using AI technology — and unsurprisingly it takes the shape of a recursive loop — a circuitous flow of money in which a circle of corporations benefits from each other, with no breaking free of this go nowhere fast recursive loop.
Follow the money one full turn: AI companies pay the evaluators. The evaluators’ revenue depends on the companies they grade. The auditors are “partnered” with the audited. The standards the evaluators enforce are written by the companies being graded. Nothing outside the loop is doing the checking. We ran this experiment before — it was called issuer-pays credit ratings, and it ended in 2008.
But the loop is not only economic. It is intellectual: machine learning loops inherit the residue of human loops — slogans, incentives, approved answers — and recursion alone does not reach the actual problem. A system that can only re-grade itself inside the rules it was handed cannot recognize that the test was written around the wrong problem.
And it is personal. Every one of us knows the version of this loop that runs in a life: the same argument, the same reward, the same next year. Getting better inside the game is still inside the game.
The question that breaks a recursive loop is never “how well does it perform?” It is: who pays the evaluator, and what does that payment buy?