Sethares–Bell El Farol bar experiment

The Sethares–Bell El Farol bar experiment offered a strikingly simple answer to a problem of mutual dependence: people who judged their visits by their own experience, rather than by trying to predict everyone else, eventually produced steady attendance just below the bar’s capacity. The result was collectively intelligent, but it divided the group into frequent regulars and people who rarely enjoyed going, leaving no obvious measure by which to declare the outcome ideal.

A simpler strategy

Engineers Ann M. Bell and William A. Sethares approached the El Farol coordination problem differently from Brian Arthur. Rather than having would-be visitors use diverse strategies to anticipate the crowd, they gave all the bargoers the same straightforward rule: go if recent visits had been good; stay away if they had been bad.

The participants did not try to infer what the others were thinking. Nor did they know how many people had attended on nights when they themselves stayed away. Their usable information was local and personal: whether they had recently enjoyed El Farol. In this design, behavior was not coordinated by a shared forecast. Each person responded to an individual record of experience.

Attendance and its uneven bargain

After enough time for everyone to learn whether to return, weekly attendance settled just below 60 percent of the bar’s capacity. That was only slightly less effective than the result an ideal central planner would have achieved. Without sophisticated prediction, the group found a workable pattern by acting on separate experiences rather than trying to calculate the crowd in advance.

Stability came with a distributional cost. The group divided quickly into regulars, who went nearly every week, and people who went only rarely. Assuming everyone took equal pleasure in a visit, this arrangement was utilitarianly sound: more than half the group could go nearly every week, and the bar was only rarely crowded. Yet a sizable share of the group seldom went and usually had a bad time when they did. The aggregate result looked orderly; the experience was not shared evenly.

What counts as a solution

The experiment did not settle whether Bell and Sethares’s result was better than Arthur’s. Both approaches appeared surprisingly successful, but coordination problems offer no outside reference point that can make the answer easy. In the cattle-weight guessing example associated with Francis Galton, the crowd judged a reality independent of its guesses. At El Farol, attendance was different: every decision helped create the reality that later decisions had to confront.

That reflexive loop makes even getting the average attendance right seem remarkable. The experiment’s central implication is not that a single rule solves coordination, but that independence can help: people attending to their own experience may collectively reach a good result even when they do not deliberately coordinate their choices. Whether that result is good enough depends as well on who benefits and who is left out.