Explaining Cognitive Agents

13^th November 2018, 13:00 Ashton Lecture Theater
Vincent Koeman
Interactive Intelligence
Delft University of Technology

Abstract

Cognitive agents derive their choice of action from their beliefs and goals. This potentially provides these agents with the capability to self-explain their behaviour in terms of these concepts. Applications that are built using cognitive agents that can self-explain their behaviour can provide users with a better understanding of that application. An example application domain is that of a negotiating agent that is able to explain why it offers the bids that it does in a negotiation. Another example is that a recommender agent developed using cognitive agent technology could explain the recommendations it makes to its user.

Agents that can explain their choice of action are useful not only for end users, but also for the developers of such agents. An agent developer needs to identify defects in the agent programs (s)he writes. Debugging an agent program requires analysing the behaviour generated by a cognitive agent program that is unexpected or undesired. In order to understand how this behaviour is generated, it is of great value to a programmer if an agent can explain why it chose to do something it did. I will focus on explanations aimed at agent developers. We believe that an explanation facility for developers is a step towards an explanation facility for end-users of applications built using cognitive agents.