arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2601.12913cs.AIcs.LGcs.NE
收藏

可操作的可解释性必须根据对称性来定义

Actionable Interpretability Must Be Defined in Terms of Symmetries

University of Oxford(牛津大学);ETH Zurich(苏黎世联邦理工学院);University of Cambridge(剑桥大学)

Pietro Barbiero, Mateo Espinosa Zarlenga, Francesco Giannini, Alberto Termine, Filippo Bonchi, Mateja Jamnik, Giuseppe Marra

AI总结:

本文论证AI可解释性研究存在根本性问题,提出可操作的可解释性应基于四种对称性来定义,以形式化可解释模型并统一可解释推理。

摘要:

本文认为,人工智能(AI)中的可解释性研究从根本上来说是不恰当的,因为现有的可解释性定义未能描述如何正式测试或设计可解释性。我们提出,可操作的可解释性定义必须根据*对称性*来制定,这些对称性指导模型设计并导致可测试的条件。在概率视角下,我们假设四种对称性(推理等变性、信息不变性、概念封闭不变性和结构不变性)足以(i)将可解释模型形式化为概率模型的一个子类,(ii)产生可解释推理的统一形式(例如,对齐、干预和反事实)作为贝叶斯逆的一种形式,以及(iii)提供一个正式框架来验证是否符合安全标准和法规。

英文摘要:

This paper argues that interpretability research in Artificial Intelligence (AI) is fundamentally ill-posed as existing definitions of interpretability fail to describe how interpretability can be formally tested or designed for. We posit that actionable definitions of interpretability must be formulated in terms of *symmetries* that inform model design and lead to testable conditions. Under a probabilistic view, we hypothesise that four symmetries (inference equivariance, information invariance, concept-closure invariance, and structural invariance) suffice to (i) formalise interpretable models as a subclass of probabilistic models, (ii) yield a unified formulation of interpretable inference (e.g., alignment, interventions, and counterfactuals) as a form of Bayesian inversion, and (iii) provide a formal framework to verify compliance with safety standards and regulations.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑