A starting point for discussion
The idea
If Zarsky is right that people who review AI decisions and can overrule them may have implicit bias—unnoticed assumptions that can unfairly affect judgment—enhancement might reduce that bias while also improving people’s thinking abilities. The author presents this as a possible way to involve human reviewers without losing benefits of the decision process that are attributed to AI systems.
Place in the reasoning
AI assessment · Connected to reasoning
The claim is explicitly conditional on Zarsky’s account of bias in human review; it presents enhancement as a possible response, not a demonstrated remedy.
Proposals to improve people’s abilities and require human review have limits
Adds a conditional rationale for enhancement in human review, while not establishing that enhancement would reduce reviewer bias in practice.
What this means and why it matters
What this means
The example concerns people who review decisions made by AI and are allowed to overrule them. If Zarsky’s concern about their unnoticed assumptions is correct, enhancement might reduce those assumptions’ effect while also improving people’s thinking abilities. This is a proposed possibility, not a demonstrated effect.
Why it matters
The example addresses a specific concern about human review. It does not establish that enhancement would remove bias or preserve all benefits attributed to AI decisions.
Illustrative example
Hypothetically, an intervention might help a reviewer notice an assumption that could unfairly affect their judgment when deciding whether to overrule an AI decision. The passage does not show that such an intervention exists or works.
Conditions & assumptions
Source coverage: Full readable text of the author’s manuscript from an accessible copy, not the publisher’s formatted version. The passage gives this as a possible example: if implicit bias—unnoticed assumptions that can unfairly affect judgment—is a serious problem when people can review and overrule decisions made by algorithms, enhancement might aim to reduce the bias while improving people’s thinking abilities. Addresses: The conditional example about possible bias when people review decisions made by algorithms.
Questions this connects to
AI-suggested relationships. Authors have not necessarily responded to one another.