Answers are grounded in this paper and cite their sources.
Root essay of The Bush Book. A girl pulls the petals off a daisy: "loves me? loves me not?" But fortune-telling never checks the petals against what actually happened, so it never learns. Today's AI is such a daisy: we pull its petals (like / dislike), and it never learns how its advice turned out. The essay proposes to measure generations of AI not by power but by whom the machine learns from and whom it protects. The Fortune-Teller (first generation) learns from approval. The Resident (second) learns from verified accounts by specific people of how its advice turned out. The Chart Room (third) plays out a step in a model of the world before it is taken and shows its price for everyone the step will touch. A machine learns only from what is in its data and in the world's answers: no computation adds information that is not there (the data-processing inequality), while the outcomes of advice exist mostly with the people who received it. The missing piece is a channel by which outcomes come back from people, verified and credited. Whether training on such lessons reduces dangerous errors without reducing helpfulness is a testable hypothesis, not a proof. In English and Russian; the two editions match paragraph by paragraph. Author: Igor Kapustin. With the participation of AI assistants: Claude, ChatGPT, Mistral. License: CC BY 4.0.
No takes yet. Share an insight, caveat, or question.
Igor Kapustin (2026) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: