Human Computer Interaction
ERCIM News No.46, July 2001 [contents]

Towards a Generic Framework for Building Intelligent Collaborative Interface Agents

by George Vouros and Vangelis Kourakos-Mavromichalis

The project Intelligent Collaborative Interface Agents (ICIA) aims at the implementation of collaborative interface agents that act in conjunction with their human partners, increasing their competency in using a software application. 

The ICIA project has been initiated by the Intelligent Collaborative Systems (InCoSys) Group at the Department of Information and Communication Systems of the University of the Aegean, Samos, Greece. We are currently developing a collaborative anthropomorphic interface agent that acts as an active partner for children from 10 to 14 years, in the context of an educational software application.

Although user interfaces designed with well-known interface design strategies, such as the information visualization strategy [Ahlberg C., Shneiderman B.: ëVisual Information Seeking: Tight Coupling of Dynamic Query Filters with Starfield Displaysí. In Human Factors in Computing Systems: Proceedings of the CHI ë94 Conference. New York: ACM, 1994], provide facilities for the users to learn progressively how to perform on themóby experimenting and observing the effects of their actionsóit is the case that novice users, and particularly small children, need the help of an expert user in order to learn basic functionalities. In this case, users need the help of an instructor that would teach them to perform the needed tasks. To instruct users effectively, an instructor needs to understand usersí goals, ascribe to users mental states describing their knowledge, beliefs, capabilities, intentions and desires, and commit that he helpful behaviour will be exhibited towards achieving usersí goals when the need arises. Understanding user goals and committing to exhibiting helpful behaviour towards achieving these goals assures that the instructor shall not perform actions that are in conflict with usersí intended actions and desires.

In addition to interactive interfaces that provide facilities for users to learn progressively how to perform on them, this research proposes the use of intelligent collaborative agents. These agents aim to provide help to users taking into account their goals, intentions, information needs, preferences and characteristics. Such agents must be designed and implemented as active partners that look over the shoulder of users and collaborate with them towards achieving a common goal.

An approach to intelligent user interfaces that is close to the ICIA approach is that of COLLAGEN [Rich C., Sidner C., Lesh N.: ëCOLLAGEN: Applying Collaborative Discourse Theory to Human-Computer Interactioní. In AI Magazine, Special Issue on Intelligent User Interfaces, 2001 (to appear)]. COLLAGEN is complementary to the interface agents proposed in this paper since it can play the role of the middleware between an application, the user and a software agent that is actually collaborating with the user. Furthermore, COLLAGEN implies that the agent will always perform the next action that contributes to a goal, which is not appropriate for instructing the user on how to perform a task.

Key issues towards our goal are the following:

A major related issue is the ability of the interface agent to balance and intermix between reactive and social deliberative behaviour. In the first case, the agent takes the initiative to act by itself and presents information or performs on behalf of the user by reacting to user actions and to events happening in the interface. Deliberating socially, the agent collaborates with the user towards achieving a shared goal.

To implement a collaborative agent so as to address the above-mentioned key issues, we have adopted the SharedPlans model for collaborative planning activity proposed in [Grosz, B.J., Kraus, S.: ëCollaborative plans for complex group actioní. In: Artificial Intelligence 86, pp. 269-357, 1996]. SharedPlans is a general model that provides the basis for the design of collaborative agents. It adopts a mental state view of plans and enables collaborating agents to treat their mental states (ie, their beliefs, desires and intentions) in an integrated way.

To address the key issues mentioned above and comply with the generic principles and the stages of the collaborative activity [Wooldridge M., Jennings N.: ëThe Cooperative Problem Solving Processí. In Journal of Logic and Computation, 9(4):563-592, 1999] we have used the ICAGENT agent development framework. This is a generic framework for the implementation of intelligent agents that act in dynamic and unpredictable environments [Kourakos-Mavromichalis V., Vouros G.: ëBalancing Between Reactivity and Deliberation in the ICAGENT Frameworkí, In: Balancing Reactivity and Social Deliberation in Multi-agent Systems, LNCS, Springer-Verlag, 2001 (to appear)]. The ICAGENT framework provides advanced plan management abilities, has been developed based on the SharedPlans model for collaborative activity and provides the basis for the implementation of collaborative agents.

The collaborative interface agent recognizes the situations where users need help. This is currently done with specialized situation recognition rules.

The agent possesses knowledge on interface objects and domain tasks, ascribes mental attitudes to users, forms intentions to act and intentions to exhibit helpful behaviour, and utilizes instructional and presentation strategies for instructing users on how to achieve their goals.

Future Activities
Future activities concern:

Please contact:
George Vouros — University of the Aegean, Greece
Tel: +30 273 82226