Call the loop G0034

The 1976 warning

Drew McDermott's paper, Artificial Intelligence Meets Natural Stupidity, named a habit it called wishful mnemonics: labeling programs and data structures with words like UNDERSTAND or GOAL. A major source of simple-mindedness in AI programs, the paper argued, is that a researcher who names the main loop of his program UNDERSTAND can mislead a lot of people, most prominently himself.

The alternative is to call that loop G0034 and then see whether he can convince himself or anyone else that G0034 implements some part of understanding. Once the habit is pointed out, examples turn up all over AI research.

Reasoning models today

The same habit runs through AI reasoning today. Chains of thought and thinking tokens borrow their names from human thought, and the shorthand takes some suspended disbelief. That doesn't mean AI reasoning can't or doesn't work, and novel research likely needs some of this mindset just to get off the ground. It resembles Oprah-style manifesting with a computer science spin.