Inexplicable agent behavior is perhaps the biggest obstacle to users accepting agent-based systems. It is critically important that the user develops a good understanding of the agent behavior. An area of future research involves creating tools to help users develop satisfactory agent models. There are two approaches to solving the problem. First, the internal state of an agent should be accessible and comprehensible to the user. Second, the agent has to be able to provide meaningful explanations for all its actions.
More work needs to be done in providing users with a better view of the internal state of the agent. One way to facilitate this is to allow better views and interfaces. For instance, consider the following situation that arose when testing the system with real users. Sometimes the user wanted to see the list of newsgroups searched by each profile separately, while at other times the user liked to see all the newsgroups searched by the entire population of profiles. The user thus needed different views of the internal state. In future implementations, perhaps, the user should also be allowed to experiment with ``what-if'' scenarios. For example, the user might want to know what kinds of articles will be retrieved if an extra keyword is added to the profile.
The second method for enhancing the understanding of the agent is to provide explanations for the agent's behavior. In the current implementation, the only information stored after the scoring process is done is the final score of the document. An alternative, for instance, could be to remember the five terms which contributed the most to the document score. Thus, a reasonable explanation would be available to justify why the document was selected. Interfaces must also be provided for the user to query any action of the agent.