Who owns algorithmic explanations?

Skeptical scholarship questions whether anyone can simply claim ownership of an algorithmic explanation. Post-hoc or simplified explanations may not expose the actual basis of a model’s output, may be too demanding for ordinary users, and may fail to guide action or justify blame. This view favors outcome-focused oversight and practical remedies over a broad right to explanation.

• Prevailing explainability methods may fail to reveal the actual reasons underlying automated decisions or support moral and legal judgments about them.
• A legal right to explanation may be restrictive, unclear, or mismatched with the technical explanations produced by machine-learning research.
• Causal and rule-based explanations may be infeasible for social prediction and too intellectually demanding for many decision-subjects.
• Explanations can become governance instruments shaped by institutional power, so the key question is who is required to explain what, to whom, and for what purpose.
• If explanations are monetized, providers may optimize them for profit, potentially corrupting their informational and accountability functions.

Featured sourceMonetization could corrupt algorithmic explanations | AI & SOCIETY | Springer Nature Linklink.springer.com6 sourcesView sources