The design of MOO agents: Implications from a study on multi-modal collaborative problem solving.
4. Conclusions
Table 1 summarizes the suggestions with respect to the design of MOO agents. Some of them are technically easy to implement (e.g. reasoning on MOO positions) while others require further research. This list of skills extends the agent reasoning: the agent does not only reason to solve the problem, it also reasons in order to interact properly. They illustrate they additional effort of reasoning involved for building a shared understanding. The boundary between reasoning on the task or on interaction is shallow: for instance, the probability of disagreement or misunderstanding concerns both the task knowledge and the interaction among partners. We attempted to further investigate the relationship between problem solving and interaction. We observed that different problem solving strategies lead to different profiles of interaction: for instance, pure 'forward' chainers often fill the whiteboard with facts during the first hour, then discuss inferences in the MOO, the MOO being used during the first hour only for management. Backward chainers start very early to discuss inferences in the MOO and write them down on the whiteboard (which is still not full). However, the relationship we observed between interactions and problem solving is very complex. In future investigations, we intend to use MOO-Observers, i.e. agents which automatically compute the indicators that we computed by hand when analyzing the protocols such as the rate of acknowledgment, the rate of co-presence in acknowledgment, the delay in acknowledgment, the complexity of dialogue patterns. In order to avoid natural language processing, we could used a semi-structured interface where the users pre-classify their utterances by using sentence openers and specify to which utterance they acknowledge (Baker & Lund, 1996).
- Match knowledge persistence with medium persistence;
- Determine to what degree a piece of knowledge is shared before any grounding act;
- Determine what the partner can see according to his position;
- Translate dialogue moves in terms of MOO actions
- Dynamically allocate the negotiation of content to communicative tools (media)
- Maintain and use the whiteboard as the shared context
- Maintain multiple conversational contexts
The design of MOO agents: Implications from a study on multi-modal collaborative problem solving - 21 MARCH 1997
Generated with Harlequin WebMaker