Improving verifiability in AI development
OpenAI + 57 orgs just published the playbook for auditing AI safety claims. Here's what verifiability actually means.
Why it matters
As AI systems move into production, the industry lacks standardized methods to prove safety and fairness. This multi-stakeholder report provides 10 concrete mechanisms for developers to demonstrate AI alignment and for policymakers to evaluate it—a foundational piece for trustworthy AI governance.
The key facts
8 to know58 co-authors across 30 organizations
Partners include Centre for the Future of Intelligence, Mila, Schwartz Reisman Institute, CASBS, CSET
10 mechanisms outlined for verifying AI safety, security, fairness, privacy claims
Tooling designed for three audiences: developers, users/policymakers, civil society
Published April 16, 2020
10 mechanisms for verifiability of AI safety, security, fairness, and privacy claims
Includes Centre for the Future of Intelligence, Mila, Schwartz Reisman Institute, CASBS, CSET
Framework designed for developers, policymakers, users, and civil society
Go to the source
OpenAI Blogopenai.com
Publisher excerpt: We’ve contributed to a multi-stakeholder report by 58 co-authors at 30 organizations, including the Centre for the Future of Intelligence, Mila, Schwartz Reisman Institute for Technology and Society, Center for Advanced Study in the Behavioral Sciences, and Center for Security and Emerging…