Fundamental considerations for the use of explainable AI in law enforcement

Explainable AI (XAI) methods have the potential to make the use of AI in law enforcement more understandable, and ultimately more trustworthy. We argue that explanation requirements differ strongly between use cases and between stakeholders ranging from law enforcement officers to affected persons. While no currently known XAI method provides a guarantee to fully reflect the functioning of an AI model, XAI methods are currently the most promising means to bridge the gap between human and AI after increasing the human’s AI literacy. Even though the benefits of XAI vary strongly with the accuracy of the AI system and need to be balanced against incurring risks, like automation bias, we argue that not using XAI implies larger risks than exploring the technologies’ benefits and further developing it. In order to overcome existing shortcomings, we advocate for more collaborations between law enforcement agencies, academia, and industry.