Current progress and boundaries
Public agencies have tried automated classification and risk analysis, but system transparency, data representativeness, and accountability often prompt concern.
Why it remains difficult
Historical data may reproduce existing inequities, and data sharing across agencies can blur accountability. Explanations must also be practically useful to ordinary people.
Paths worth exploring
Conduct impact assessments and consult stakeholders first. Preserve human decision authority, an appeal channel, version records, and explicit conditions for discontinuing use.
How to evaluate success
Audit error rates and resource impacts across groups, appeal outcomes, audit findings, and timeliness of remedies. Accuracy must not be treated as the only standard.