Y.-K. Guo

Well-meant policies keep backfiring: people treat a new rule as a problem to solve for their own advantage, the best-resourced adjust first, and the costs drift back onto the people the rule meant to protect. I propose rehearsing policy with AI: models play the people it touches, including enforcers, intermediaries, and a red team hunting for loopholes, so drafters can see who takes the detour and where the costs land before revising the text. It is a stress test, not a prophecy.

A dream I had after writing Taste: a room at home that shows a different scene each time it opens, in groups that barely differ, and picking the best of each group is how I make a living. A choice that gets recorded becomes the next default, so taste starts being harvested the moment it is used. What can be learned is a preference. What cannot is a commitment, or the line between options and what is not an option at all.

When options get cheap, the value is in the choosing. Industrialization pushed ordinary goods down to cost, and AI is doing the same to ordinary intelligence. What still carries a price is the choice made under abundance: what to make, what to leave out, what counts as done. That choosing is taste. It does not spend efficiency; it decides whether what efficiency produces still has a price.

I was a technology optimist for years. What frightens me is not that AI takes the work. It is that cheap, fast, always-present intelligence demotes the person from subject to the AI's sensors and actuators.

I tried to replace runtime configuration with compile-time composition, one binary per scenario. The paradigm held and the code ran, but I stopped — not because it was wrong, because it was early.