History‑trained AI Advises Policy

Training language models on a focused historical corpus (e.g., Bismarck’s correspondence and chronology) and then prompting them about modern crises can produce structured, argument‑style advice that mimics historical actors. Experiments reveal both promising analytical help (chronology, causal framing, decision counterfactuals) and risks: confident but misleading analogies, 'jagged' competence across topics, and the temptation for policymakers to substitute model‑narratives for nuanced expert judgment. — If governments and advisers start using purpose‑trained historical AIs to justify or design policy, that could change how states learn from the past — amplifying some lessons, suppressing others, and institutionalizing algorithmic analogy as a mode of strategic reasoning.