A Granular Computing Approach to Provide Transparency of Intelligent Systems for Criminal Investigations

Criminal investigations involve repetitive information retrieval requests in high risk, high consequence, and time pressing situations. Artificial Intelligence (AI) systems can provide significant benefits to analysts, by sharing the burden of reasoning and speeding up information processing. However, for intelligent systems to be used in critical domains, transparency is crucial. We draw from human factors analysis and a granular computing perspective to develop Human-Centered AI (HCAI). Working closely with experts in the domain of criminal investigations we have developed an algorithmic transparency framework for designing AI systems. We demonstrate how our framework has been implemented to model the necessary information granules for contextual interpretability, at different levels of abstraction, in the design of an AI system. The system supports an analyst when they are conducting a criminal investigation, providing (i) a conversational interface to retrieve information through natural language interactions, and (ii) a recommender component for exploring, recommending, and pursuing lines of inquiry. We reflect on studies with operational intelligence analysts, to evaluate our prototype system and our approach to develop HCAI through granular computing.