Broad listening and collective input
Tools like Polis, Talk to the City, and Kouchou AI use language models and clustering to read free-form input from thousands of citizens, contributors, and beneficiaries. How should that input feed into allocation mechanisms, so that funding decisions track what affected communities actually say while staying incentive-compatible, legitimate, and auditable?
Other threads
Context
This thread anchors discussion at the Columbia 2026 workshop and is eligible for support through AI4PG 2026 grants.