The AI Vulnerability Database (AVID) project aims to help practitioners recognize, diagnose, and manage the risks of AI systems by building public interest open-source resources. However, calls for community participation in mitigating AI harms often fail to support the difficult implementation decisions of practitioners or are too specific. AVID proposes to develop community and multi-stakeholder interventions that shape AI practices at scale by designing a community-led editorial process, a community-led taxonomy of AI failures, a governance structure that enables participation, and a program of participatory workshops and user studies. The goal is to co-create resources with the many communities working to make AI less harmful.
Team
Borhane Blili-Hamelin, `19 PhD, Columbia University