A project led by Professor Tim Rogers has been chosen from 101 initial applications for one of nine Research Forward awards. Research Forward, a competition sponsored by the Office of the Vice Chancellor for Research (OVCR), stimulates and supports highly innovative and groundbreaking research at the University of Wisconsin–Madison.
The initiative is supported by the Wisconsin Alumni Research Foundation (WARF) and will provide funding for 1–2 years, depending on the needs and scope of the project.
“Previous rounds of Research Forward have moved campus forward in several critical research areas such as cancer treatment, pushing the frontiers of treating brain disorders and achieving direct carbon removal from the atmosphere,” says Cindy Czajkowski, interim vice chancellor for research. “As we are now seeing the transformative discoveries resulting from earlier rounds of Research Forward, it is exciting to build on those successes and we expect this year’s cohorts to continue this tradition where multi-disciplinary research leads to impactful innovations in new avenues of study that will forward the Wisconsin Idea.”
Rogers’s research project leverages AI to power more realistic simulated personas that can interact via natural language and are tuned to update their beliefs in ways that mimic real human participants — an AI Terrarium. The AI agents will be engineered and fine-tuned to capture patterns of real communication and belief change observed in representative human populations collected using panel surveys and behavioral experiments. The behaviors of interacting “digital twins” – AI models matched to human participants in the survey and experimental collections–will then be used to simulate and understand patterns of communication and persuasion in the real world.
Once developed, the AI Terrarium will enable low-cost testing of different message campaigns, news feeds, or interaction rules. Promising strategies will be validated in carefully controlled experiments using lab-based networked messaging platforms. The ability to simulate social media interactions with human fidelity would permit rapid-cycle testing of intervention strategies on a variety of health and political issues, ranging from vaccine skepticism to electoral misinformation, and allow piloting prior to costly real-world trials.