HOME       LINKS     RESEARCH     PUBLICATIONS

An Approach to Evaluating Human Characteristics in Agents by Norling & Sonenberg

Norling & Sonenberg - An Approach to Evaluating Human Characteristics in Agents

This paper provides a brief overview of three issues concerning agents that are able to behave in a human-like way in complex social environments. The first issue is the design and implementation of an agent that is able to make realistic decisions. The authors use a belief-desire-intention (BDI) agent, which they have modified to have naturalistic decision making. The inclusion of the Recognition-Primed Decision model (RPD) was designed to counter the usual implementation of BDI agents that simply choose the first available plan that satisfies their intentions. The second issue raised in this paper was how to select a testbed that would allow effective evaluation of the agent. Some of the concerns that were raised in regards to testbed selection were the difficulties in implementing a satisfactorily complex environment as well as finding domain experts to evaluate the behaviour of the agent. The domain chosen in this paper was Quake 2, the domain experts being experienced game players. The final issue raised was how to evaluate the behaviour of these agents, when the characteristic that is being evaluated is not quantitative. For example, it is relatively easy to measure increases in an agent’s speed or accuracy, but to measure the agent’s decision making ability requires subjective input from human subjects.

Go to Norling & Sonenberg's "An Approach to Evaluating Human Characteristics in Agents"

http://www.cs.mu.oz.au/agentlab/publications/Author/NORLING-E.html

Reference details for Norling & Sonenberg's "An Approach to Evaluating Human Characteristics in Agents"

Norling, E. & Sonenberg, E.A., 2002, An Approach to Evaluating Human Characteristics in Agents in Proceedings of the International Workshop on Regulated Agent-Based Social Systems: Theories and Applications, Hamburg, pp 51-60.