For calibrated binary predictors, any collaboration rule that is guaranteed to be at least as accurate as the worst agent must essentially always defer to a single agent.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A No Free Lunch Theorem for Human-AI Collaboration
For calibrated binary predictors, any collaboration rule that is guaranteed to be at least as accurate as the worst agent must essentially always defer to a single agent.