Research Background & Goal

Past research on collective intelligence has dealt with wisdom formation in civil societies and nations, or the collective behaviors of creatures such as insects and bacteria. Recently, the proliferation of the Internet and World Wide Web has had an impact on the research of collective intelligence: activities best represented by Wikipedia are also recognized as collective intelligence. These research activities are aimed at accumulating "knowledge" and forming collective intelligence by "human" cooperation. The target, however, has been limited to relatively simple themes such as the compilation of an encyclopedia and predictions in presidential campaigns; the latter affords the participants easy access to related information in forming his/her prediction. If we try to deal with themes wherein rights to access related information is complicated, and forming individual opinion depends on others' opinion, then simple aggregation does not work and contributions from computer science are indispensable.

Researchers in computer science have studied computational models of collective decision making. This field is called "autonomous agents and multiagent systems." Research in this field started around 1980 and at present the research field has grown into the largest topic in artificial intelligence. The decision maker is called the agent and computational models of the cooperation, competition, negotiation, coordination, and organization of agents have been studied. Human organizations are classified into "team," "market," and "community." In a team, the members have a common goal; in a market, participants pursue their own goal; in a community, the members share common interests. So far, a variety of rational (or bounded rational) computational models have been actively studied for teams and markets. These computational models, therefore, dealt with the domain that makes it easy to automate the procedures such as cooperative problem solving and auction. However, we cannot ignore the human factor observed in human communities when dealing with collective intelligence formation. While accepting the use of partially automated procedures, a study of participatory systems, in which humans and agents learn from each other, is required. The following figure shows the advances in agent research.

Figure : Advances in Agent Research

Research on Semantic Web as the basic technology for collective knowledge structuring started in 2001. Semantic Web is an effort to make Web information, which only humans can understand at this moment, machine (agent) readable. Ontology description language OWL based on description logic was recommended by W3C in 2004. However, developing an ontology that always offers logical consistency is not easy in actual user fields. We take notice of the rapid acceptance of Web services. By using the framework of Web services knowledge can be freely combined with other knowledge if each knowledge is wrapped and presented as a web service. Since it is not easy to structure knowledge to support logical reasoning, the approach should be to structure and utilize knowledge services. Service-oriented knowledge structuring from user fields is an exciting direction for developing collective intelligence that supports both agents and society.

The above analysis of current research trends indicates the importance of collective intelligence created by the service-oriented integration of "humans," "agents," and "knowledge."

Our goal is to achieve a paradigm change in collective intelligence: from the current paradigm, wherein people manually collect and accumulate knowledge, to a paradigm consisting of "humans," "agents," and "knowledge," where collective intelligence is created and fostered by integrating services. In the new paradigm, human society and the corresponding multiagent systems learn from each other and complement each other. Modeling each human decision making process creates one agent, and simulations consisting of collective agents can predict collective intelligence formation. Knowledge is organized by operational semantics of services, not by its denotational semantics. By viewing humans, agents, and knowledge as services, we can develop participatory technologies to integrate them. The following figure shows our research goal.

Figure : Research Goal

• Ishida Toru. How to Create Super Intelligence, Ishida & Matsubara laboratory seminar, 2009. (pdf, 1.1MB)