Hank Green proposes a homemade benchmark — the “shelter bench” — as a counterweight to the technical evals that dominate AI discussion (drawing pelicans, novel proteins, beating grandmasters). The question: how will this technology help solve the housing crisis?
The point is not that housing is the most important problem in the world. The point is that housing is a structurally honest problem to test against. The blockers are well-understood and they are not information-shaped: landowners holding power, zoning that protects scarcity, neighbors who fear change, property values that benefit from constraint. AI can perhaps speed up permitting or design review at the margins — but margins are not where the bottleneck lives.
Verbatim: “It’s like saying that steam engines are going to fix hurricanes. Like there are problems that exist in different universes.”
Generalization: the shelter bench is a usable diagnostic for any “X will change everything” claim. Pick a problem whose blockers you understand well and which you know not to be information-bound. Ask honestly how X helps. If the answer is “at the margins,” then X is a real but bounded technology, not a civilizational solution. The test prevents the conflation of technical capability with civilizational impact — the move described in 2026-04-26-powerful-technology-transforms-society-but-does-not-necessarily-solve-problems.