【摘要】 可解释性是人工智能技术发展过程中越来越重要的伦理要求之一。现有的研究从围绕指导决策的人工智能可解释性原则开始转向开发人工智能可解释性工具，以帮助用户理解人工智能系统。然而，从批判建构主义的视角来看，这些可解释性工具只关注实质可解释性，不仅忽视了概念自身的争议性，还完全忽略了形式可解释性。可解释性工具在实现实质可解释性的同时，会掩盖形式不可解释性问题，并且实质可解释性概念在本质上具有争议性，这些共同造成了可解释性工具的困境，因此可以通过采取混合式的可解释性定义方式、赋予用户定义权等途径消解当前可解释性工具的困境。更多还原

【Abstract】 Interpretability is one of the more and more important ethical requirements in the development of artificial intelligence technology. Existing researches have shifted from starting around AI Interpretability principles that guide decision making to developing tools for AI interpretability to help users understand AI systems. However, from the perspective of critical constructivism, AI interpretability tools only focus on improving substantive interpretability, ignoring not only the controversial nature of the concept itself, but also the formal interpretability entirely. Interpretability tools can cover up the problem of formal interpretability while realizing substantive interpretability, and the concept of substantive interpretability is controversial in nature, which together result the dilemma of interpretability tools. Therefore, the dilemma of interpretability tools can be resolved by adopting a hybrid definition approach of interpretability and giving the users the right of definition of interpretability.更多还原