AI fairness requires domain experts who understand the social context of the training data.

Evidence

No matter how well-collected the criminal justice data, any technology that builds upon it must be vetted by experts who can contextualize the data. This is necessary because, at the least, technology should operate fairly, and fairness must be defined by key stakeholders. Once it is so defined, then context becomes vital for making sure that the legal technology employs the fairness definition that the key stakeholders think is the appropriate one to use (see Selbst et al., 2019). — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law, 1182 (#corroborates, via linked claim)

Categorization

• XREF: connects to: Distinguishing legitimate from illegitimate predictors requires domain expertise in policing bias.; Algorithmic fairness requires stakeholder-defined meaning plus expert contextual vetting — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law: A Definitive Look at…, 1182; Bias-reducing legal AI requires ongoing human translators to iterate between models and practice — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law: A Definitive Look at…, 1198 (+3 more)

Broader Connections

• Algorithmic fairness requires stakeholder-defined meaning plus expert contextual vetting #corroborates (0.93) — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law, 1182