Researchers have developed the first implementation of the causal perception framework, which explores how agents with differing models of a system infer different probability distributions and perceive fairness. This work operationalizes structural and parametrical causal perception, proposing algorithms for computing interventional and counterfactual distributions and measures to quantify disagreement. Experiments using the German Credit dataset demonstrate that causal perception impacts accuracy and fairness in multi-expert decision-making, showing that bias is situated with respect to an agent's specific model and cannot be ignored in fairness assessments. AI
IMPACT Introduces a new framework for understanding and potentially mitigating situated bias in AI decision-making systems.
RANK_REASON Academic paper detailing a new framework and its implementation. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
