Fairness assessment measures outcome differences across groups using metrics such as demographic parity, equalized odds, and calibration. These metrics can conflict, so the applicable definition depends on context.
Assessment requires demographic data the system may not collect.
Alternative Names:
Bias Assessment, Fairness Evaluation
Why it Matters?
Fairness is more directly relevant to systems making or influencing decisions about people than to document analysis tools, which is where most legal AI sits. It becomes relevant where AI informs settlement valuation, since a model trained on historical outcomes will reproduce whatever demographic patterns those outcomes contained. That is a real concern for damages modeling and outcome prediction applications rather than for extraction and summarization.
Frequently Confused with
Related terms
Frequently asked questions
Which legal AI applications raise fairness concerns?
Can fairness metrics conflict?
