OpenAI Unveils Five AGI Principles…Altman Acknowledges Bigger Role

OpenAI unveiled Our Principles, a five-principle document for AGI development, on Sunday. The company stated it aims to prevent AI power from concentrating in the hands of a few, marking its clearest stance since the 2018 Charter.

The principles are five:

• Democratization: Entrust AI decision-making to democratic processes rather than to a lab.
• Empowerment: Give users broad autonomy, but act conservatively when harm is uncertain.
• Universal prosperity: Lower infrastructure costs and widen benefits through compute expansion, vertical integration, and global data center expansion.
• Resilience: Cooperate with companies, governments, and the ecosystem on biological weapons and cyber risks, and commit Foundation resources.
• Adaptability: The principles may be revised as technology changes, and the reasons for doing so will be disclosed.

Sam Altman acknowledged that OpenAI has grown far larger than it was at the time of the 2018 Charter. He wrote that in some cases resilience may need to be prioritized over empowerment, and he revisited the 2019 decision to delay the release of GPT-2 weights, explaining that the controversy led to the discovery of staged rollout, which later became the default release strategy.

This document is closer to a policy declaration one level above last year's Preparedness Framework. Amid tightening frontier AI regulation in the US and Europe, OpenAI is putting public standards front and center, citing cooperation with the Frontier Model Forum and AI safety research institutes in the US and UK.

Recently, surveillance controversy over Pentagon work and California's first state-level frontier AI safety law have both weighed on the company simultaneously. OpenAI appears to be consolidating its own public stance before regulators fully write the rules, and Altman added that the company may not get everything right, but it will learn and revise quickly.