Artificial intelligence (AI) plays an increasingly prominent role in decision making in domains critical to society such as criminal justice, healthcare, and misinformation identification. It is crucial that AI systems be able to explain the basis for the decisions they recommend in ways that humans can easily comprehend, thus serving as a bridge between humans and AI. While most current computational research in generating explanations focuses on the AI side, little attention has been paid to how humans provide and interpret explanations. This project advances our understanding of natural language explanations formulated by humans, and then moves on to develop improved algorithms for human generation of explanations and human-machine collaborations on explanations.
First, the project will develop computational approaches to understanding human explanations by leveraging a unique large-scale corpus of naturally-occurring explanations with human annotations highlighting the persuasive elements of an argument. Additional datasets with annotations of explanations that draw on psychological theory of effective explanations will be created. Second, the project will build algorithms that learn from these natural language explanations so that AI systems can generate explanations that follow human style and so are more easily interpreted and compelling. Third, the project will develop best practices for soliciting human explanations where an AI system collaborates with the human to generate more effective explanations.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.