A team of Rensselaer and IBM researchers was recognized with the best resource paper award at the 2020 International Semantic Web Conference (ISWC) on November 6 for their paper "Explanation Ontology: A Model of Explanations for User-Centered AI".  Their research, supported by the IBM AI Horizons Network, produced a publicly available ontology that models the role of explanations that an AI system may offer to support its insights and recommendations.

A simple example might be an AI system that monitors the health of a user and recommends eating fewer carbohydrates in the coming week. If the users asks why, the system might explain that the user is pre-diabetic and the system sees a rising trend in blood sugar levels, which raises risks.

The AI community has been increasingly aware of and interested in enabling the ethical, equitable, and safe use of insights generated by AI.  As AI systems become more pervasive in our society, their ability to provide users with information that explains why they generated their results has been a topic of particular interest. The creation of this ontology marks an essential first step towards helping AI system designers make informed choices on what kind of explanations AI systems need to provide to their users.

The ontology the team created covers a range of explanation types identified in the literature, and accounts for relationships between explanation types, the system interface, and user attributes.

The team’s work is aimed at using semantics to go beyond representing information about a specific problem, to enabling AI systems to become more useful and trustable partners for the people who use them. While the paper illustrates how the ontology can be used in healthcare AI applications, the ontology has the potential to support diverse explanation types in many other domains as well.

The Health Empowerment by Analytics, Learning and Semantics (HEALS) project is proud of the recognition this paper has received at this prestigious venue, and continues to pursue research topics related to the application of the Explanation Ontology in health care.

Rensselaer authors are Shruthi Chari, a computer science doctoral student, Deborah McGuinness, a Tetherless World Senior Constellation chair, Oshani Seneviratne, the director of health data research, and Daniel Gruen, a senior scientist. IBM Research authors are Morgan Foreman, a healthcare data scientist and Amar Das, the program director of Integrated Care Research at IBM Research at the time the research was conducted.