Bowling Green, Kentucky is a small US city that recently completed an experiment to use AI for democracy. Can an online polling platform powered by machine learning capture what residents are looking for in their city?
In 2023, when Doug Gorman was elected leader of the county which includes Bowling Green and assumed office, the city was projected to double its size by 2050. However, there wasn’t a plan in place for how this growth would unfold. Gorman met with Sam Ford, a consultant from the locality who had worked on Pol.is’ surveying platform, which uses machine-learning to gather opinions of large groups. Ford says they “needed a plan” for the growth that is expected. Ford and Hennessy assembled a group of volunteers who had experience in eight different areas: economic development; talent; housing; public health; quality of life. They developed a plan that would use Pol.is as a tool to help create a 25-year city plan. The platform is one of many new technologies being used in Europe, and increasingly in the US, to ensure that local government is informed by public opinion.
In February, after a month-long advertising campaign, the Pol.is website was launched. Residents could anonymously submit their ideas (in less 140 characters) on what the 25-year strategy should include. They could also vote whether they agreed with or disagreed other ideas. The tool was available in the language of the participant, and moderators checked to ensure that the traffic came from the Bowling Green region.
During the month it was live, 7,890 people participated, and 2,000 submitted their own ideas. An AI-powered Google Jigsaw’s tool then analyzed the data in order to determine what people agreed on and disagreed upon.
According to experts on democracy technologies, who were not involved with the project, this level of participation – about 10% of the city residents – was impressive.
Archon Fung is the director of the Ash Center for Innovation and Democratic Governance, at the Harvard Kennedy School. “That’s a lot,” he says. He says that a local election could see a 25% participation rate, and all it takes is to fill out a ballot.
This is a more demanding form of participation, right?” “You’re voting or considering some substantive issues, and 2,000 people contribute ideas,” he says. “I think that’s quite a few people who are involved.”
In the Bowling Green experiment, the plans that attracted the most attention were hyperlocal. The most popular ideas were to increase the number of local doctors so that residents would not have to travel to Nashville for medical treatment, to encourage more restaurants and groceries to open in the city’s northern part, and to preserve historic buildings.
Other controversial ideas included the approval of recreational marijuana, adding gender identity and sexual orientation to the city’s anti-discrimination clause and providing more options for a private education. 2370 of the 3,940 unique suggestions received more than 80% support, including initiatives such as investing in stormwater infrastructure or expanding local opportunities for adults and children with autism.
Although the experiment was run by volunteers, they were not completely hands off. The moderators screened the ideas according to a policyand did not post redundant ideas. Ford claims that 51% of the ideas submitted were published and 31% were deemed redundant. Around 6% of the ideas were not published because they were completely off-topic, or contained a personal insult.
However, some researchers who are interested in the technologies that can improve democracy question whether this method of soliciting input is a reliable one to understand what communities want.
Self-selection is a problem. For example, certain types of people tend to attend in-person forums such as town halls. Fung says that research shows that people with higher education, homeowners and seniors are more likely to attend. It’s possible similar dynamics were at play when residents of Bowling Green decided to participate in this project.
James Fishkin, political scientist at Stanford, says that self-selection does not represent the opinions of the public. He developed a method he calls “deliberative polling” in which a sample of residents from a population is brought together over a weekend and paid $300 for their participation. They are then asked to deliberate with small groups. Some European governments use jury-style groups to make policy decisions.
It’s clear that these tools promise to move a community in a democratic direction. But we won’t be able to tell if Bowling Green’s experiment was successful until residents see how the city uses the ideas they raised.
Beth Simone Noveck is the director of a lab at Northeastern University that studies democracy and technologies. She says, “You can’t create policy based on tweets.” She points out that residents voted on 140-character ideas and now those need to be transformed into real policies.
She says that the next step is to have a conversation with residents about how to turn a 140-character proposal into a policy that can be implemented.
The organizers have already made the results of the survey publicand will be making recommendations to the Warren County Leadership later this year.
