Four Bottomless Errors and the Collapse of Statistical Fairness

Statistical fairness conflates fairness with equality: The paper argues that statistical fairness approaches mistake treating similar cases similarly (equality) for actual fairness, which limits how we can think about just treatment in AI systems.

The framework relies on flawed ethical reasoning: The author identifies that statistical fairness uses a "perspectival ethical view which functions by negating others and their viewpoints," suggesting the approach dismisses legitimate alternative perspectives on what fairness means rather than engaging with them.

Pre-defined social groups constrain rather than enable fairness: A core problem is that statistical fairness requires deciding group categories in advance, when fairness should theoretically work unconstrained and allow group composition to emerge from fair treatment principles rather than being imposed beforehand.

These errors are structural, not fixable: The paper contends that because these four mistakes are fundamental to how statistical fairness operates, attempting to fix individual problems only makes things worse—the entire approach would need to be abandoned rather than reformed.

The collapse opens new possibilities: Rather than ending the conversation about AI fairness, the author suggests this critique points toward alternative approaches more aligned with classical fairness concepts dating back to Aristotle, though the paper only briefly sketches these directions.