Ethical limitations of algorithmic fairness solutions in health care machine learning

Abstract

Artificial intelligence has exposed pernicious bias within health data that constitutes substantial ethical threat to the use of machine learning in medicine.1,2 Solutions of algorithmic fairness have been developed to create neutral models: models designed to produce non-discriminatory predictions by constraining bias with respect to predicted outcomes for protected identities, such as race or gender.3 These solutions can omit such variables from the model (widely regarded as ineffective and can increase discrimination), constrain it to ensure equal error rates across groups, derive outcomes that are independent of one's identity after controlling for the estimated risk of that outcome, or mathematically balance benefit and harm to all groups.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 290 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.