An Interventionist Approach to Explainable Artificial Intelligence

This project aims to develop a new approach to explaining and understanding decisions generated by artificial intelligence (AI).

Popular approaches rely on counterfactuals, which focus on how an outcome would change, given different inputs. Such explanations are criticised in philosophy for failing to provide causal understanding. Interventionism is a theory of explanation from philosophy designed to yield such understanding. This project aims to develop new strategies for explaining AI decisions using interventionism. Expected outcomes include improved understanding of AI and better AI decision-making. Anticipated benefits include new knowledge and support for government to use AI effectively while protecting the interests of individuals.