UMBC ebiquity

Virtual Patients as Intelligent Agents

Speaker: Sergei Nirenburg

Start: Friday, May 08, 2009, 01:00PM

End: Friday, May 08, 2009, 02:00PM

Location: 325b

Abstract: Virtual patient (VP) environments are becoming very popular as an educational, reference and assessment tool for the medical profession. Technologically, however, state-of-the-art VP environments devoted to training cognitive skills are still little more than decision trees whose nodes present static material and allow users a choice of which daughter node to visit next. The Maryland Virtual Patient (MVP) project seeks to simulate medical encounters at a higher level of verisimilitude. To this end, they are modeled in a society of intelligent agents where the human user plays the role of attending physician and other medical personnel and the VP itself are realized as artificial intelligent agents. The VP is viewed as a “double” agent – a combination of a “body”, a simulator of normal and pathological physiological processes, and a “mind”, a model of human cognitive capabilities of perception, reasoning and action. The project also seeks to formalize the clinical experience of master physicians in diagnosing and treating diseases. Knowledge of these “best clinical practices” is used to support automatic tutoring.

The VP is a knowledge-based model and simulation of a person suffering from one or more diseases. Physiologically, it undergoes normal and pathological processes and responds realistically to expected and unexpected (e.g., user error-caused) internal and external stimuli. On the cognitive side, its perception apparatus includes experiencing symptoms and understanding natural language input from the user; its reasoning is goal- and plan-based with elements of reasoning by analogy and individual preferences based on context and a model of character traits; its arsenal of actions includes verbal, mental (memory management and learning) and simulated physical actions.

In this talk, we will briefly discuss the current capabilities of the MVP environment and illustrate them with a demonstration of a training encounter in the latest version of the MVP system.

The MVP project is a collaboration between ILIT and a research team at the University of Maryland School of Medicine led by Dean Bruce Jarrell and Dr. George Fantry. The ILIT team members include: Steve Beale, Marjorie McShane, Jesse English, Ben Johnson and Matt Rodatus.