People can over-rely on an automated recommendation because it appears systematic, authoritative or convenient. In AI-assisted work, this may reduce independent checking or cause a person to overlook a model's uncertainty and limitations.
Avoiding the bias does not mean rejecting every automated result. Effective oversight calibrates trust to evidence and task conditions, preserves domain expertise and makes it practical to question both model outputs and human assumptions.
ELI5
Automation bias happens when people trust a computer's answer more than the evidence warrants. A polished answer can still be wrong.
For example, a reviewer might accept an assistant's summary without checking the source even though the summary conflicts with a known fact. The right response is to investigate the evidence, not automatically prefer either side.
Does automation bias mean computers are always less reliable than people?
No. It concerns unjustified reliance, not a claim that human judgment is always superior.
Can human review still suffer from automation bias?
Yes. A nominal review step is weak if the reviewer routinely accepts the automated recommendation without meaningful checks.
