PulseAugur
LIVE 09:09:48
research · [2 sources] ·
0
research

Study finds Shapley value benchmarks for AI explainability misaligned with human utility

A new paper examines the evaluation of explainable AI (XAI) methods, specifically Shapley value variants, in high-stakes scenarios like fraud detection. Researchers found that standard quantitative metrics for XAI do not align with human understanding or decision utility. While the tested XAI formulations did not improve analyst performance, they did increase decision confidence, raising concerns about automation bias. AI

Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →

IMPACT Current XAI evaluation metrics may not reflect real-world human utility, potentially leading to overconfidence and automation bias in critical decision-making.

RANK_REASON Academic paper on XAI evaluation methods.

Read on arXiv cs.AI →

COVERAGE [2]

  1. arXiv cs.LG TIER_1 · In\^es Oliveira e Silva, S\'ergio Jesus, Iker Perez, Rita P. Ribeiro, Carlos Soares, Hugo Ferreira, Pedro Bizarro ·

    Rethinking XAI Evaluation: A Human-Centered Audit of Shapley Benchmarks in High-Stakes Settings

    arXiv:2604.22662v1 Announce Type: new Abstract: Shapley values are a cornerstone of explainable AI, yet their proliferation into competing formulations has created a fragmented landscape with little consensus on practical deployment. While theoretical differences are well-documen…

  2. arXiv cs.AI TIER_1 · Pedro Bizarro ·

    Rethinking XAI Evaluation: A Human-Centered Audit of Shapley Benchmarks in High-Stakes Settings

    Shapley values are a cornerstone of explainable AI, yet their proliferation into competing formulations has created a fragmented landscape with little consensus on practical deployment. While theoretical differences are well-documented, evaluation remains reliant on quantitative …