CONCEPT
The Myopia of Learning
March and Levinthal's 1993 diagnosis of the three structural biases — temporal, spatial, and failure-averse — that make learning systems favor the near, the certain, and the measurable over the distant, the uncertain, and the meaningful.
Learning systems are structurally myopic. They discount the future, favor the local over the distant, and systematically underweight negative outcomes from unexplored alternatives while overweighting positive outcomes from current strategies. The myopia is not pathological; it is the normal operation of a well-functioning learning system. At each individual decision point, the exploitation choice is better-supported by available evidence. No individual decision-maker is making an error. The error is emergent — visible only at the system level, over horizons longer than any individual decision, and only to an observer who can see the entire trajectory. AI intensifies every mechanism of this myopia, compressing the feedback loop between action and observed outcome from months to minutes, making exploitation returns so visible and exploration costs so evident that the asymmetry becomes civilizational.
In The You On AI Field Guide
The three mechanisms operate in parallel. Temporal myopia discounts the future: an exploitation strategy producing measurable returns this quarter