Measurability Distortion
AI changes what work can be done and which work gets rewarded. When some tasks become easy to score, automate, and display, institutions tend to reallocate attention toward them. Measurable work starts to look like the work that matters.
This is a familiar failure mode of benchmarks. A benchmark begins as a proxy for progress. It becomes a target. Then it becomes a training distribution, a funding signal, and a hiring criterion. AI strengthens the loop because it produces visible gains precisely where the proxy is sharp.
The same mechanism can operate in mathematics and science. Problems with crisp verifiers, public leaderboards, formal checkers, and short feedback cycles become easy for institutions to evaluate. Problems requiring taste, synthesis, long-horizon theory choice, or messy contact with the world remain harder to score. Resources move toward the first class even when practitioners know that the second class contains much of what matters.
The distortion compounds through training. New researchers learn the taste of the environment that rewards them. If the environment rewards AI-tractable problems, the next generation may internalize tractability as importance. The field can then become more productive by its metrics while narrowing its actual range.
Blog Use
Use this for the institutional mechanism. It explains how local use of AI can reshape research priorities without anyone explicitly choosing that outcome.
Source Trail
- Johan Commelin et al., “Shaping the Future of Mathematics in the Age of AI”, arXiv:2603.24914.
../ai_views.md, “Tractability colonizes significance” and “Measurability distortion is self-reinforcing”.- Leiden Declaration on Mathematics and AI, cited in the source note but still needing bibliographic verification before publication.
- See
reference_inventory.mdfor full source status.