Decisions can be more democratic when individuals with no preset preference join a group
Odd as it sounds, adding wishy-washy members to a group can wrest control from a strongly opinionated minority and make collective decisions more democratic.
At least that’s what happened in an experiment with schooling fish and three kinds of computer simulations described in the Dec. 16 Science. “Quite counter-intuitive,” says study coauthor Iain Couzin of Princeton University. “What we’re trying to do with this paper is put out a new idea.”
Couzin is not arguing that there’s a benefit to a poorly informed electorate. But he does call for experiments to clarify the role that uninformed people with no opinion on a choice play in human consensus building.
The study “supports a growing body of evidence that larger groups are better decision makers than smaller groups,” says applied mathematician David Sumpter at the University of Uppsala in Sweden who studies collective behavior.
It also echoes economic research showing that having some fraction of uninformed traders in a market can reduce volatility, says Michael Kearns, a computer scientist at the University of Pennsylvania in Philadelphia who is interested in collective behavior.
The fish study grew out of computer simulations Couzin created that demonstrated the considerable power of opinionated minorities in otherwise indifferent groups, flocks or herds. When he mixed factions with different strengths of opinion in this simulation, as well as in two very different analyses of group behavior, he found hints of peculiar effects of uninformed parties.
For an experimental test of the odd effects, study coauthor Christos Ioannou, now at the University of Bristol in England, worked with small freshwater fish called golden shiners, which have a very strong tendency to stick together in schools. Like other animals such as bees, these fish can learn to associate food with some colors (yellow for the fish) much more readily than with other colors. Ioannu, who was kept in the dark about the predictions of the computer simulations, trained shiners to swim toward either yellow or blue marks in a tank for a treat. Even after the same training, those aimed toward yellow were more committed to their color. So these became the fish version of individuals with stronger opinions.
When the researchers mixed a minority (five) of these strongly yellow-seeking fish with a majority (six) of less passionate, blue-seeking fish, the whole group swam toward the minority yellow mark in more than 80 percent of the trials. When the researchers added five uninformed fish — with no training toward yellow or blue or any apparent tendency to swim toward either mark — the whole group ended up at the minority yellow target only half the time. Adding another five fish reduced the minority victories to below 40 percent.
But human groups have significantly different dynamics than laboratory fish, says Carl Bergstrom of the University of Washington in Seattle. The fish are just influencing each other locally instead of proselytizing intensely and sharing information globally, the way people do in modern democracies.
In his laboratory experiments with real people pushed to reach consensus, Kearns says, “when the minority wins, it tends to happen fast — it’s almost shock and awe,” he says. So he can imagine that adding neutral, perhaps vacillating parties could give a majority a chance to recognize and exert its force.
“Maybe the optimum state isn’t everybody being highly informed and having very, very strong political opinions,” Kearns speculates. Perhaps an ideal world would still need a little ignorance. “Maybe the role of these ignorant individuals, whether they be fish or American voters, is to provide a stabilizing, mediating effect,” he says. But whatever the ideal dose of ignorance may be, current levels clearly exceed it, he says. “I think we’re very, very far from the optimum.”
I.D. Couzin et al. Uninformed individuals promote democratic consensus in animal groups. Science, Vol. 334, December 16, 2011, p. 1578. doi:10.1126/science.1210280.