Abstract
A "world model" is constructed where precedent searching is one of the primary driving mechanisms. The simulation assumes that nations in the system are utility maximizers but that they have relatively primitive decision mechanisms and are strongly influenced by their previous short-term successes and those of other states in the system. This model of foreign-policy decision making has been heavily influenced by recent artificial intelligence studies and simulations.
States in the simulation can follow one of three distinct strategies to maximize growth: imperialism, militarism, or trade. In each of these modes, a state can either increase or decrease its level of behavior, or it can switch modes. Decisions to switch are based on the success of the policy in increasing simulated GNP relative to earlier projections of how much GNP would increase. If a policy is clearly not working, a nation implements the reverse of it, if the policy is not producing major improvements, the nation randomly experiments or looks at the success of other nations in the system and follows whatever has worked for them. If a policy is clearly successful, it is continued.
The simulation was run using a system vaguely characteristic of the 19th-century world system, with 5 large, 5 medium, and 10 small nations. The resulting behavior is generally plausible, with bounded and fairly diverse activity depending on the random experimentation involved Because of the weak bounded rationality, the system does not lock on to a single pattern of behavior based on initial conditions, and so, for example, situations exist where medium powers eventually become stronger than the initial major powers. The most common pattern is one of a combination of trade links and imperialism, with about half the minor power being colonized and some exchange of colonies occurring through conflict. Keywords: artificial intelligence, foreign policy, global simulation, precedent-searching simulation, world modeling, simulation.
Get full access to this article
View all access options for this article.
