The Conflict Between Explainable and Accountable Decision-Making Algorithms

Explainability alone may not ensure accountability: The paper argues that making algorithms explainable doesn't automatically solve responsibility issues in high-stakes decisions like healthcare enrollment and hiring. Instead, post-hoc explanations (explanations generated after a decision is made) can actually obscure who is truly responsible for algorithmic outcomes.

Explainability can shift blame away from developers: When XAI systems provide explanations, they may be perceived as "blameworthy agents" themselves, which can inadvertently shield algorithm developers and designers from accountability for their choices in building and deploying these systems.

Vulnerable people may be wrongly blamed for algorithmic decisions: The paper warns that patients and other subjects of algorithmic decisions might be incorrectly held responsible for outcomes because explainable systems create a false impression that these individuals have control over or can meaningfully challenge the algorithms affecting them.

There is a risk of using algorithms and people as "scapegoats": Designers may deliberately use both the algorithm itself and affected individuals as convenient targets for blame, allowing companies and developers to escape legal and moral responsibility for harmful decisions.

Hard regulation is necessary to protect accountability: The authors recommend that legal requirements and regulations must go beyond simply demanding explainability and instead focus on ensuring developers cannot evade responsibility for algorithmic decision-making systems.