OpenAI CEO Sam Altman has released a set of five guiding principles that also serve to rationalize the company's most controversial business decisions.
In a post titled "Our Principles," Altman argues that the future of superintelligence will either concentrate power in a few hands or distribute it widely. OpenAI claims it aims for the latter, but notably omits a pluralistic market of competing AI providers as an alternative. Instead, the focus is on access for end users, positioning OpenAI as a central provider while framing itself as a vehicle for decentralization.
The first principle, democratization, promises AI access for everyone and democratic decision-making. Yet it's unclear how this aligns with OpenAI's heavy lobbying spending. The principle also subtly criticizes rival Anthropic, which has faced friction with the Pentagon over its safety red lines.
The second principle, empowerment, grants users broad autonomy while pledging to minimize catastrophic harm. OpenAI says it will err on the side of caution when in doubt.
Universal prosperity, the third principle, justifies OpenAI's massive compute investments, vertical integration, and global data center buildout. Altman hints that new economic models may be needed to distribute AI-generated wealth.
Resilience, the fourth principle, calls for society-wide preparedness against AI risks. Iterative deployment—gradually rolling out capabilities—is just one part of this strategy. Altman acknowledges that OpenAI may need to collaborate with governments and other AGI projects on alignment and safety before proceeding further.
Finally, adaptability reserves the right to change course. The debate over releasing GPT-2's weights is cited as a case where early concerns proved overblown, leading to the iterative deployment approach.
The principles also serve as an indirect response to criticism over OpenAI's Pentagon deal. Altman concludes by inviting feedback, acknowledging that mistakes will be made and corrected.