Automated Interpretation of Agent Behaviour

Software comprehension, which is essential for debugging and maintaining software systems, has lacked attention in the agent community. Comprehension has been a manual process, involving the analysis and interpretation of log files that record agent behaviour in the implemented system. This paper describes an approach and tool to automate creating interpretations of agent behaviour from observations of the implementation execution, thus helping users (i.e. designers, developers, and end-users) to understand the motivations of agent actions. By explicitly modelling the user’s comprehension of the implemented system as background knowledge for the tool, feedback can be provided as to whether the user’s comprehension accurately represents the implementation’s behaviour and, if not, how it can be corrected. Additionally, with the aid of the Tracer Tool, many of the manual tasks are automated, such as verifying that agents are behaving as expected, identifying unexpected behaviour and generating explanations for any particular observation.