‘The Lock-in Hypothesis: Stagnation by Algorithm’ — the published precursor on dual human↔AI influence entrenching today’s values. ICML 2025.
Status: shipped · People: Tianyi Alex Qiu, Zhonghao He
The paper argues that influence running in both directions, humans shaping models and models shaping humans, can entrench whatever values are current rather than letting them improve. It is the conceptual precursor to the Martingale and Psychohistorians lines, which take the same worry and try to measure it: Martingale on a single agent’s belief updates, Psychohistorians on a population’s. The public code is in Modeling-Lock-in.
The project is in maintenance. Nothing here is being extended; it stays on the hub because the later work refers back to it.
Also here
The pages below need a passcode.
- Log — the run record.