Economics as Distributed Computation Robert L. Axtell * Center on Social and Economic Dynamics, The Brookings Institution, 1775 Mas- sachusetts Avenue, NW Washington, DC 20036 raxtell@brookings.edu Summary: In human societies diverse people act purposively with powerful but limited cognitive processes, interacting directly with one another through techno- logically-facilitated and physically-mediated social networks. Agent-based com- putational modeling takes these features of humanity—behavioral heterogeneity, bounded rationality, network interactions—at face value, using modern object- oriented programming techniques to create agent populations that have a high de- gree of verisimilitude with actual populations. This contrasts with mathematical social science, where fantastic assumptions render models so cartoon-like as to beg credibility—stipulations like identical agents (or a single ‘representative’ agent), omniscient agents (who accurately speculate about other agents), Nash equilibrium (macro-equilibrium arising from agent-level equilibrium) and even the denial of direct agent-agent interaction (as in general equilibrium theory, where individuals interact only with a metaphorical auctioneer). There is a close connection between agent computing in the positive social sciences and distrib- uted computation in computer science, in which individual processors have het- erogeneous information that they compute with and then communicate to other processors. Successful distributed computation yields coherent computation across processors. When such distributed computations are executed by distinct software objects instead of physical processors we have distributed artificial intelligence. When the actions of each object can be interpreted as in its ‘self interest’ we then have multi-agent systems, an emerging sub-field of computer science. Viewing human society as a large-scale distributed system for the production of individual welfare leads naturally to agent computing. Indeed, it is argued that agents are the only way for social scientists to effectively harness exponential growth in compu- tational capabilities. Keywords: economics, distributed computation, multi-agent systems * Preliminary versions of this paper were presented at the U.S. National Academy of Sci- ences colloquium "Adaptive Agents, Intelligence and Emergent Human Organization: Capturing Complexity through Agent-Based Modeling" held at the Arnold and Mabel Backman Center in Irvine, California (October 2001), the Third Trento Summer School on Adaptive Economics, held at the Computable and Experimental Economics Labora- tory at the University of Trento, Italy (July 2002), and at the Agent-Based Approaches to Economic and Social Complex Systems (AESCS), Tokyo, Japan (August 2002).