January 20, Thursday 10:30, Education Building Room 570
Title: Artificial Minds and Simulated Bodies
Lecturer : Sergei Nirenburg
Lecturer homepage : http://www.cs.umbc.edu/CSEE/people/nirenburg.shtml
Affiliation : Department of Computer Science, University of Maryland Baltimore County
Our research group develops artificial intelligent agents that are capable of perception, decision making, language understanding, mental and verbal action and learning. They are distinguished by having varying degrees of knowledge about the world in which they are immersed as well as a model of personal characteristics. They are intended for operation in mixed human-computer agent teams carrying out real-world tasks. In this talk I will describe two proof-of-concept applications of our theory of intelligent agency, both in the domain of clinical medicine. Time, technology and courage permitting, I will also show live demos of these systems.
MVP (Maryland Virtual Patient) is a training environment in which a human trainee plays the role of attending physician while artificial agents play the roles of the patient, tutor and supporting medical personnel. Many things set our virtual patients apart from today.s mainstream virtual patients, which are almost completely preprogrammed though often visually appealing. Our virtual patients are .double agents. in that they feature models of the body and the mind. Specifically, they are endowed with
a) simulated physiology that tracks the progression of a disease and responds appropriately to external medical interventions; and
b) the model of a mind, manifesting itself in the patient.s ability to initiate action as well as remember, understand and dynamically react to user input, and learn through experience.
In the CLAD (CLinician.s ADvisor) environment the artificial agent gives advice to a clinician during an interview with a human patient. This advisor agent is purely cognitive (that is, its body is not modeled) but it develops and manipulates a mental model of both the mind and the body of the human patient. A core prerequisite of generating high-quality advice is CLAD.s ability to simulate the expected progression of each specific patient.s disease, based on information already gathered about him or her, and to predict the patient.s likely responses to different treatment options. CLAD is not aimed at replacing the clinician by a machine but rather is intended to reduce the clinician.s cognitive load.