Concept · part of Who owns the filters

Suppose a newspaper’s front page is rigged. Swap the editor and tomorrow’s page is different. Now suppose a school curriculum is rigged for twenty years. Swapping the curriculum fixes nothing for the people already taught: they read every new fact through what they learned.

The book’s capture taxonomy turns this into a rule. Capture means tuning a filter against what it was meant to serve. Some filters only shape the surface a person meets: what a feed ranks, how an AI tool is configured, what an archive holds. Others install the equipment people (or AI models) read with: schooling, an AI’s training data, the goal it was trained toward. The book calls the second kind consumer-key capture and argues it is harder to recover from. The damage sits in the reader, and recovery usually takes a generation.

The rule generalizes an earlier claim, the asymmetry between captured training and captured archives. It also sets a priority for institutions that want trustworthy AI: defend training data and objectives first, and accept weaker defenses on configuration, which can be re-tuned cheaply. The book notes that the commercial landscape does the reverse.

Read the full argument

How it connects


← Capture taxonomy (substrate × source × target asymmetry) · Who owns the filters (8 of 9) · Capture asymmetry (training vs. preservation) →

Explore all the ideas · Read the book in order · Open the interactive map