Answers are grounded in this paper and cite their sources.
Two imaginary technologies, an augmentation pill and a dish of lab-grown neurons, show which AI policies will still work a decade from now and which popular tools expire first. Investigation, liability and testing all depend on being able to see why something happened. AI is steadily eroding that ability. This essay argues that the point where it is lost, the visibility threshold, should organize AI policy. The rules that feel most natural today break at that point: label the machine's work, and make whoever caused a harm pay. What survives is a set of seven less glamorous commitments that are cheap to build now and close to impossible to retrofit later. One of them is a line rather than a safety measure. It marks a small class of decisions that should stay with people even if the machine never errs. The essay is a short companion to the report The Pill and the Dish, which gives the full analysis and policy designs.
No takes yet. Share an insight, caveat, or question.
Verlyn Fischer (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: