The Useful Knowledge Feedback Loop

In The You On AI Encyclopedia

The framework explains why scientific knowledge alone does not produce economic growth. A society can possess extensive propositional knowledge — the late Roman Empire, Song Dynasty China, medieval Islamic civilization — without sustaining the feedback loop. The loop requires channels efficient enough to move knowledge through each stage at a pace that keeps the cycle going. When any stage slows, the entire loop slows.

The Industrial Enlightenment accelerated the loop by attacking the conversion bottleneck. Scientific societies, patent law, the Encyclopédie, technical education — each reduced the cost of converting understanding into technique and thereby sped the cycle. The result was not just faster invention but a qualitatively different relationship between science and practice, in which each stage fed the others at an unprecedented rate.

The AI transition attacks the conversion bottleneck more aggressively than any previous innovation. Large language models reduce the cost of converting propositional knowledge into prescriptive capability to near zero for problems specifiable in natural language. The consequence, Mokyr's framework predicts, is an acceleration of the entire loop. The Nobel Committee's 2025 citation made the prediction explicit: 'AI could reinforce the feedback between propositional and prescriptive knowledge, and increase the rate at which useful knowledge is accumulated.'

The implications are profound and ambivalent. A faster feedback loop means faster accumulation of useful knowledge — which means faster solutions to problems that previous generations could not solve. It also means faster obsolescence of existing prescriptive knowledge, faster displacement of workers whose skills are commoditized, and faster pressure on institutional structures that were designed for slower cycles of change.

Origin

The framework was developed in The Gifts of Athena (2002) and elaborated across Mokyr's subsequent work. The Nobel Committee's 2025 citation established it as the central theoretical contribution to which AI had added new urgency.

Key Ideas

Four-stage cycle. Propositional → prescriptive → new data → new propositional → expanded prescriptive.

Speed governed by bottleneck. The slowest stage determines the pace of the entire loop; speeding any other stage produces no acceleration if the bottleneck remains.

Historical bottleneck: conversion. For most of human history, converting understanding into technique was the binding constraint — the stage where ideas went to die.

AI attacks the bottleneck directly. Natural language interfaces collapse conversion costs for a significant class of problems, releasing the feedback loop to cycle faster than any previous era allowed.

Acceleration is ambivalent. Faster useful knowledge accumulation means faster displacement, faster institutional pressure, faster obsolescence — gains and costs moving at the same accelerated pace.