自动化偏见Automation Bias
Automation Bias
人这头的毛病：AI 说得顺畅，人就懒得检查照单全收。The human-side flaw: smooth AI output makes people skip checking and accept it wholesale.
国际 AI 安全报告指出：人们会表现出自动化偏见。即不加仔细检查就接受 AI 的建议。
它与模型偏见是两回事：一个错在模型，一个错在人的使用方式。
AI 输出越流畅自信、界面越权威，人越容易放松核查。
流程上的对策：高风险决定强制二次验证，而不是依赖「看起来靠谱」。
The International AI Safety Report notes people exhibit automation bias — accepting AI suggestions without checking them carefully.
It is distinct from model bias: one is the model being wrong, the other is how humans use it.
The more fluent and confident the output and the more authoritative the interface, the more people relax their checking.
Process countermeasure: mandatory second verification for high-stakes decisions, not reliance on seeming reliable.
模型出错只是事故的一半，另一半是人不再核查。知道这个毛病，才会在关键流程里强制设检查点。Model error is only half of many incidents; the other half is humans no longer verifying. Knowing this flaw is why critical workflows need enforced checkpoints.
看站群怎么分工：这一站回答「是什么」，其他站回答「选哪个」How the site family divides the work ↗
回到名词地图看第2站的其他 35 个词 →Back to the map: the other 35 terms in station 2 →
- International AI Safety Report 2026《Extended Summary for Policymakers》（automation bias 表述）International AI Safety Report 2026, Extended Summary for Policymakers ↗