EAGER: Subsequent Similar Cases to Unexpected, Exceptional Cases

University Of Massachusetts Amherst, Amherst MA

Investigators

Abstract

Given that change is ubiquitous, it is necessary to understand how systems can learn in response to unexpected, exceptional, and provocative events. This project addresses reasoning about strikingly novel, perhaps disruptive cases, so that subsequent cases similar to the harbinger case do not take an intelligent agent by surprise; rather, the agent can guard against harm and/or lay groundwork to reap reward. The project explores similarity assessment among novel, possibly high-impact cases and uses hypothetical reasoning to explore the space of possible future cases and the implications of these cases on an agent's environment. The extent in time and quality of this exploration (e.g., to include representation change) are conditioned on the costs and benefits associated with anticipating (or not) new events like the original disruptive event. These issues are being explored in two application domains: (1) law where there are numerous areas with apt historical episodes of surprising exceptional cases provoking dramatic change (e.g., warrantless search), and (2) multi-agent distributed resource planning that allow controlled experimentation with models, model-revision policies, and hypothetical problem scenarios (e.g., the Producer-Consumer-Transporter domain). The research is a fundamental step in understanding and representing the actionability facet of concepts, particularly in the face of problematic or atypical instances. It contributes to efforts to promote critical thinking by showing how critical cases can provoke change. It furthers our knowledge of how insightful hypotheticals, so called 'what if' situations, can be used to shed light on the ramifications of actions, classifications, and policy decisions.