Brian Arthur

Brian Arthur is the economist who formulated the El Farol problem, a model of how people make choices when the outcome depends on how many others make the same choice. His computer experiment showed that decentralized agents, using diverse and fallible strategies rather than following a central planner, could produce an aggregate result that matched the group’s shared interest: over one hundred weeks, attendance averaged exactly the threshold at which the most people could have a good time without the bar becoming overcrowded.## The El Farol ProblemIn the early 1990s, Arthur asked whether a satisfying solution existed to a coordination problem inspired by El Farol, a bar in Santa Fe that sometimes became too crowded on nights featuring Irish music. He defined the stakes precisely: if the bar was less than 60 percent full, everyone there would have fun; if it exceeded 60 percent, no one would. People would therefore go only if they expected attendance to remain below that limit. Their individual decisions could not be evaluated in isolation: whether going was sensible depended on what everyone else expected and did.## Different Strategies, No Certain ForecastArthur assumed there was no mathematically deducible answer telling a potential visitor what to do. Instead, people would rely on different rules of thumb. Some might expect Friday’s attendance to match the previous Friday’s; others might use the last night they had personally visited, average attendance over several weeks, or predict the opposite of the preceding week. These approaches made the hypothetical crowd diverse, but no single strategy could guarantee a correct decision.## One Hundred Weeks of FluctuationArthur tested the problem in computer simulations, equipping agents with different strategies and letting them make attendance decisions over one hundred weeks. The resulting crowd size fluctuated sharply and irregularly. Sometimes the bar exceeded 60 percent capacity for three or four consecutive weeks; at other times it fell below the threshold on four of five weeks. Strategies that worked for a time eventually had to be discarded, because changing attendance undermined any dependable prediction.## The Collective ResultThose fluctuations had real costs within the model: on some nights the bar was too crowded for anyone to enjoy it, while on other nights people stayed home even though they would have had a good time. Yet across all one hundred weeks, attendance averaged exactly 60 percent. That was the level that let the maximum number of people have fun without making anyone endure an overcrowded bar. Arthur’s result is not that decentralized choices always work out well on each occasion, but that their aggregate can nevertheless align with what the group as a whole wants. The example makes coordination under uncertainty visible without appealing to a central authority that assigns people when they may go.