Researchers from the University of Washington, Stanford and Northeastern developed a web-based tool that can reduce partisan hostility on X without removing posts or requiring platform cooperation.
The tool uses AI to downrank posts showing antidemocratic attitudes or extreme partisan animosity, such as advocating violence or rejecting bipartisan cooperation. In a 10-day experiment with 1,200 participants during the 2024 election, those exposed to less of this content reported warmer feelings toward the opposing party, with effects seen across political affiliations. The approach builds on sociology research and leverages AI for nuanced feed reordering.
The team published findings in Science and released the tool’s code for others to explore interventions aimed at reducing polarization and improving outcomes like mental health and civic engagement.