The challenge for AI for the decade is to deal with the issues of interaction. The interaction between agents requires establishing, maintaining, and extending common ground. Common ground between human and computational systems can be aided by mediators that link to the capabilities and natural properties of task agents, people, and the world we live in. We need systems that are open and can grow; that allocate resources fairly to get jobs done without requiring either centralized planning or excessive communication. With agents with multiple goals, coordination of actions must proceed through mutual commitments. Finally, we must realize that not all problems of interest are AI problems, nor are results relevant to our research always to be found just in journals of our fields. In working with intelligent agents, it is useful to explore what is happening in such fields as cognitive psychology, anthropology, economics, neural modeling, and in the rest of computer science. The natural tendency of any science is to isolate itself as a community, and develop its own practices. But AI must cross the boundaries; to be intelligent is to be intelligent about something; success must come in the context of tasks from outside the field. People who are part of the systems must be moved as high as possible in the intellectual food chain, so that participation in the system helps them to be better in ways that they care about. We must not just look to build intelligent systems to solve specific problems; we need to learn how to help corporations and nations build intelligent organizations that support human values, and enhance the world we live in.
Bobrow, D. G. Dimensions of interaction: AAAI-90 presidentional address. AI Magazine. Spring 1991; 12 (3): 64-80.