Smarter Crowdsourcing: Countering Election Subversion

Misuse of the legal system and election-related violence are among the most serious threats facing American democracy. Working with Democracy Fund and The GovLab, Citizens Foundation applied the Smarter Crowdsourcing method to this problem: structured online deliberations that bring together carefully curated experts — election officials, researchers, legal scholars and practitioners — to turn their collective know-how into actionable recommendations.

What made this engagement different was Policy Synth, our open-source toolkit for combining human deliberation with AI agents. Policy Synth automated the creation of over a thousand different search queries — general, scientific, data-specific and news-related — to conduct comprehensive research into problems and their root causes, then generated, debated and ranked candidate solutions alongside the human experts. The result: research that would normally take months of analyst time was compressed into weeks, while every recommendation stayed grounded in expert judgment.

The first real-world test of human + AI policymaking

The election subversion project became the first major real-world case study of Smarter Crowdsourcing augmented with AI agents. The recommendations were delivered to Democracy Fund to support its grantmaking against election subversion ahead of the 2024 US elections.

We described the method and results in a peer-reviewed paper, Using Artificial Intelligence to Accelerate Collective Intelligence: Policy Synth and Smarter Crowdsourcing, by Róbert Bjarnason, Dane Gambrell and Joshua Lanthier-Welch, written for the ACM Collective Intelligence Conference 2024. Beth Simone Noveck and Róbert Bjarnason also wrote about the project in Fast Company: How AI could restore our faith in democracy.

Our position throughout: AI is a tool to enhance human intelligence and creativity, not replace it. Policy Synth keeps people in control of every judgment that matters, while AI removes the drudgery between them.