Forecaster David Manheim: define before you measure, AGI debates will look silly in a year or two
9 月 16 日,AI 预测与风险领域研究者 David Manheim 围绕「AGI」术语与 AI 监管方法论连发多条帖子,核心观点是:先定义、再测量,且监管不必等精确定义。
已确认
- Manheim 重申其 2019-2020 年以来的立场:预测社区应停用「AGI」这类定义不清的术语,否则测量结果基本没有意义。他还引用一处自嘲式对比:预测社区说「别用 AGI 这种模糊词」,AI 社区却仍在使用。
- 他同时预测:当前 AI 能力轨迹大概率延续——能力参差但整体快速推进,会使 AGI 定义之争在一两年内失去意义,答案会越来越明显。
- 在与评论者(如 averyfjames 转述的讨论)的交锋中,他回应「监管必须先精确定义对象」(例如监管核反应堆需先定义裂变)的质疑,指出国家完全可以监管界定不清的事物:非法毒品管制通过不断重新定义把新型策划药纳入;核能监管则通过覆盖所有放射性物质(无论是否可裂变)来入手。
为什么重要
- 这场争论直指 AI 治理的方法论前提:如果「先定义再测量」是有效预测与评估的必要条件,当前围绕 AGI 的大量基准与时间线预测可能基础不稳。
- 毒品与核能的例子表明,监管可以采取宽口径、动态界定的路径,这为概念尚无共识的 AI 政策提供了可借鉴的务实思路。
2026-09-16 ~ 2026-09-16 · 5 related posts
Primary sources
- Forecasting community: stop using poorly-defined 'AGI' — define before you measure — davidmanheim ·
- Researcher David Manheim: AGI definitional debates will look silly within a year or two — davidmanheim ·
- Regulation Doesn't Need Perfect Definitions: Lessons from Drugs and Nuclear Power — davidmanheim ·
- [source] Forecasting community: stop using poorly-defined 'AGI' — define before you measure — davidmanheim · 2026-09-16
- [source] Researcher David Manheim: AGI definitional debates will look silly within a year or two — davidmanheim · 2026-09-16
- Define Before You Measure: The Case for Regulating Poorly-Scoped Tech — averyfjames · 2026-09-16
- [source] Regulation Doesn't Need Perfect Definitions: Lessons from Drugs and Nuclear Power — davidmanheim · 2026-09-16
- Poorly-Scoped Definitions Don't Block Regulation: Lessons from Drugs and Nuclear — davidmanheim · 2026-09-16