Token Economics Versus the 20 Watt Brain: Why Inference Costs, Not Intelligence, Will Cap Clinical AI, & Whose Job It Is to Decide If Machines Should Diagnose Us When They Reason Better but Cost More

AI models hit 80-85% accuracy on brutal NEJM diagnostic cases. Unaided generalist physicians averaged around 20% on the same cases. So why isn’t smarter AI running on every patient? The answer is the economics of thinking.

There are two AI cost lines. Training: one-time capex, you build the model, done. Inference: ongoing opex, you pay every time the mod…