The Feedback Cost Map

The timing of AI disruption is better predicted by feedback cost than by discipline name. A domain changes quickly when candidate outputs can be turned into reliable training or evaluation signal. It changes slowly when feedback is expensive, delayed, noisy, or missing.

This gives a rough map. Formal systems, code, and combinatorial search have near-zero feedback cost once the verifier exists. Protein structure, molecular simulation, and some materials problems have fast computational or experimental proxies. Clinical trials, wet-lab biology, and engineering deployment have medium to high feedback costs. Macroeconomics, epidemiology, climate policy, and many institutional sciences have slow, confounded, or non-repeatable feedback. Mathematical taste and theory choice often have no ground-truth signal at the time the decision must be made.

The map cuts across academic labels. “Biology” contains both fast-feedback structure prediction and slow-feedback clinical intervention. “Mathematics” contains both formal theorem proving and open-ended conjecture formation. “Physics” contains simulation-tight subproblems and theory-selection problems where the relevant evidence may take decades.

This also explains why AI progress can look broad while remaining uneven. Models improve rapidly where the learning loop closes. They may still fail to produce comparable progress where success cannot be scored cheaply. The bottleneck is often the absence of a usable reward channel.

Blog Use

Use this for predictions. It replaces vague claims such as “AI will disrupt science” with a map of feedback costs.

Source Trail

  • ../ai_views.md, “The feedback cost map predicts disruption timing more precisely than subdomain”.
  • Max Welling, Information Bottleneck interview, cited in the source note but still needing exact bibliographic verification.
  • Noga Alon et al., “Remarks on the disproof of the unit distance conjecture”, arXiv:2605.20695, for a concrete low-feedback-cost mathematical case.
  • See reference_inventory.md for full source status.