Online civic discourse is already hostile, and while artificial intelligence is part of the problem, the same technology can be part of the solution by helping people understand one another, surface useful information and make difficult conversations more productive.
A multi-year, $10 million “moonshot” grant from Laude Institute will help Cornell researchers realize the potential of AI, while giving online communities control over its role.
The interdisciplinary team of Cornell faculty and students, with expertise in AI, conversational behavior, computational systems and human-centered design, will develop tools and protocols to rebuild the foundations of trust online, where humans are no longer the only participants. The broader, long-term goal is a new communication ecosystem where AI strengthens public conversation rather than undermining it, researchers said.
“We are at a turning point in the history of human communication. AI is becoming an active participant in the conversations through which we form opinions, resolve disagreements and make collective decisions,” said Cristian Danescu-Niculescu-Mizil, Ph.D. ’12, professor of information science in the Cornell Ann S. Bowers College of Computing and Information Science and lead principal investigator on the project, “Rebuilding Trust in Civic Discourse: An Infrastructure for Auditable AI.” “We have a rare opportunity to shape what this transformation looks like, before the norms and infrastructure governing AI’s role in our conversations become entrenched.”
“This topic is incredibly challenging and extremely timely,” said co-principal investigator Jon Kleinberg ’93, the Tisch University Professor of Computer Science and Information Science in Cornell Bowers, in his remarks at the award ceremony on Oct. 29 in San Francisco. “We really feel the urgency here and feel this sense of mission and motivation to have AI make a positive change in some of the profound problems that we’re facing.”
Co-investigators are: David Rand ’04, professor in Cornell Bowers, the Cornell SC Johnson College of Business and the College of Arts and Sciences; Natalie Bazarova, M.S. ’05, Ph.D. ’09, associate vice provost for research and professor of communication in the College of Agriculture and Life Sciences; Robert Kleinberg ’97, professor of computer science in Cornell Bowers; and Mor Naaman, the Don and Mibs Follett Professor of Information Science at Cornell Tech, the Jacobs Technion-Cornell Institute and Cornell Bowers.
“Recent advances in AI pose a serious threat to civic discourse, enabling deepfakes, impersonation and manipulation,” Rand said. “But AI also has the potential to be a real positive force. In our experiments, conversations with AI can help correct misconceptions and mediate difficult conversations. Our project aims to unlock that positive potential in a way that communities can actually verify and trust.”
The team is one of four that received moonshot awards from Laude Institute to leverage AI to solve humanity’s biggest problems, out of eight finalists. In April, the Cornell team, which includes 11 student researchers, received an initial $250,000 seed grant from Laude Institute to create prototypes and a proposal for a shot at the $10 million grant. In the last six months, they developed a suite of tools:
- The Assistant Toolkit allows community members to build, configure and test AI tools that participate in group discussions or privately support individual users, reflecting their community’s needs and rules.
- ConvoCompass brings community-developed assistants, to help editors understand relevant policies and identify opportunities to use existing dispute-resolution processes.
- ConvoArena and Simulation Factory support studies with human participants and tests using simulated conversations to investigate how AI interventions affect discussions.
- The Auditing Benchmark provides a shared testing ground for developing and comparing independent auditors that check whether AI assistants follow their stated specifications.
“The seed grant period enabled us to refine our vision and build a strong foundation for our work with online communities through numerous interviews, co-design workshops and even actual tool deployment on their platforms,” Bazarova said. “We also used this time to assemble a highly motivated interdisciplinary team. I want to give a special shoutout to the student researchers for their dedication and outstanding work, as they have been a true powerhouse on this project.”
In addition to these tools, the project introduces a novel AI auditing layer that produces a transparent record of each agent’s interventions, enabling independent verification of whether an agent’s behavior aligns with its declared roles and standards. Together, these components establish a new foundation for online communication – allowing AI to actively participate in discussions while remaining continuously transparent and subject to independent verification.
“We are at a moment in human history in which we have more power as a species than we’ve ever had,” said Andy Konwinski, computer scientist and Laude Institute founder, at the award ceremony. “Please continue to work on the problems that you’re working on and take that to heart and inspire other academics and non-academics to work on those same species-level problems.”
Another team with Cornell connections also received a moonshot grant.
JP Pollak ’00, M.S. ’08, Ph.D. ’11, senior researcher-in-residence at Cornell Tech and assistant professor of clinical epidemiology and health services research at Weill Cornell Graduate School of Medical Sciences, is project co-lead of JupyterHealth, a collaboration between University of California, Berkeley, and the University of California, San Francisco.
Louis DiPietro is a writer for the Cornell Ann S. Bowers College of Computing and Information Science.
