Partnership on AI

An institution for cross-sector AI work

Partnership on AI was publicly established in September 2016 by Amazon, Facebook, Google, DeepMind, IBM and Microsoft. It now describes itself as an independent 501(c)(3) nonprofit with the formal name Partnership on AI to Benefit People and Society. Its corporate supporters supply grants and participate in work; that support is not recorded as corporate ownership. [1][2]

A resource for studying failures

PAI introduced the AI Incident Database in November 2020 as a searchable collection of reported harms and failures involving AI systems. Sean McGregor’s launch post, written as a representative of PAI partner XPRIZE Foundation, invited incident submissions and positioned the database as infrastructure for learning from past cases. Incident reports are evidence to examine, rather than automatic proof of a particular system’s legal liability. [3]

PAI’s current work spans safety-critical AI, labor and the economy, media and public policy. These convening and research roles help connect technical development with the institutions and people affected by deployed systems. They do not give the nonprofit regulatory authority over its partners. [2]