Politicians always vow to fight for what “the people” want, but knowing exactly what they want is like making sense out of the Tower of Babel. Voters are often deeply polarized, and even individuals wrestle over competing and contradictory goals.
Polls and voter referendums don’t solve the problem, because expressing support for a particular proposal isn’t the same as deciding how to allocate a limited budget between a slew of popular projects. In a city or nation with thousands or millions of voters, very few will have exactly the same priorities.
Now, however, researchers at Stanford School of Engineering are drawing on crowdsourcing and social algorithms to spur participatory democracy from New York City to Chicago to Vallejo, California.
Led by Ashish Goel, professor of management science and engineering, the Stanford Crowdsourced Democracy Team has helped 20 cities across the United States run real-world experiments in participatory budgeting.
In New York City, residents in five districts were given the chance to choose their top preferences from lists of proposed local projects. In a Brooklyn district, for example, people were asked to help allocate a capital budget of $2 million. The options included new Wi-Fi at an elementary school ($100,000); playground lighting at a park ($500,000); and new trees along residential streets ($260,000). In Cambridge, Massachusetts, residents could rank their preferences for a capital budget of $500,000. Their choices ranged from a new outdoor amphitheater for $350,000 to eight new bike-repair stations for a total of $12,000.
In each of these projects, the question the researchers grappled with was how to design algorithms that would get people to vote under constraints — not just on individual projects they like but on a whole budget. “The problem is not just that people don’t have the power to decide,” says Goel. “The problem is that you have to make them grapple with the same decisions as political leaders.”
It sounds simple, but it isn’t. How do you design a system that best reflects the combined preferences of thousands of people and scores of choices, without making the ordeal hopelessly time-consuming for citizens? How do you know if the results are indeed accurate?
Goel and his colleagues have developed and tested an array of algorithms that they believe address some of the toughest challenges.
The simplest approach, and the one that has been used the most so far, is a platform for participatory budgeting. Residents of a community are given a list of as many as 25 proposed capital spending projects and a specific budget. They are then asked to pick their five favorite projects, and those answers are then processed into a community-wide preference for the best mix of projects within the constraints of the budget.
The general strategy that performs best on a host of measures, says Goel, is “knapsack” voting. The knapsack concept, which has a long history in computer science, offers a way to have people look at the entire budget challenge. Instead of merely voting on the projects they like best, each person is asked to allocate an entire budget (or fill up the whole knapsack) in the way each thinks is best.
“In our experiments, when people are asked to solve a full complex budget problem, they have shown an ability to do so,” Goel says. “When users engage in knapsack voting, they act in ways that are consistent with our theoretical predictions. They start picking cheaper products. They take cost into account. They balance cost and value.”
Again, it all sounds simple. The challenge is in scaling up without producing mind-numbing complexity. Much of Goel’s theoretical work is about finding ways to get accurate readings without forcing people to wade through huge numbers of intricate choices. At some point, the quest for perfect accuracy would force individual voters to do almost as much work as the leaders they had elected to make those decisions.
The Stanford researchers have tested a number of ways to simplify the process and still preserve accuracy.
They found that it was possible to get a good picture of the community’s preferences — and identify clusters of minority views — by offering each voter a surprisingly small number of pairwise comparisons. That finding stemmed from a collaboration with government authorities in Finland, which ran a participatory democracy project to identify preferences for reforming off-road traffic laws.
The key principle was to give only a handful of randomly shuffled pairings to each person. With enough participants and with careful directing of the randomness, the researchers concluded, the result would be enough comparisons between enough pairs to provide a good measure of how all the options stacked up against each other.
To be sure, political leaders may not pay attention to what people want, regardless of how accurate the participatory budgeting might be. But Goel argues that the approach can reveal unexpectedly narrow differences — even in the bitter political polarization of Washington politics.
In experiments where participants were asked about the best ways to allocate the federal budget, he noted, people were in far less disagreement than the uncompromising partisan warfare in Congress might suggest.
“When people get into big discussion groups, they often descend into vitriol,’’ Goel says. “We wanted to know if you could design an algorithm that would provide the benefits of people talking to each other without the problems of talking in large groups.”
If communities can convey what they really want, he continues, they may be able to nudge leaders away from rancor and toward solutions.