Dual rationality and deliberative agents

Abstract

Human agents deliberate using models based on reason for only a minute proportion of the decisions that they make. In stark contrast, the deliberation of artificial agents is heavily dominated by formal models based on reason such as game theory, decision theory and logic - despite that fact that formal reasoning will not necessarily lead to superior real-world decisions. Further the Nobel Laureate Friedrich Hayek warns us of the 'fatal conceit' in controlling deliberative systems using models based on reason as the particular model chosen will then shape the system's future and either impede, or eventually destroy, the subtle evolutionary processes that are an integral part of human systems and institutions, and are crucial to their evolution and long-term survival. We describe an architecture for artificial agents that is founded on Hayek's two rationalities and supports the two forms of deliberation used by mankind. © 2010 Springer-Verlag London.