Asilomar AI Principles

In the AI Story

The Principles represented the first broadly endorsed international framework for AI development, emerging from a moment when the field was sophisticated enough to require shared norms but not yet so competitive that coordination had become impossible. The conference produced genuine technical debate and reached consensus on principles that most participants initially expected to be controversial—a consensus that has held in letter if not always in practice.

The Principles' status in Tegmark's policy trajectory is aspirational. They are guidelines, not mandates. They describe what beneficial AI development would look like rather than creating structures to enforce that description. The 2017 landscape still permitted aspirational framing because capability lagged sufficiently behind concern that voluntary alignment seemed plausible. The subsequent years demonstrated the framework's limitations: capability advanced, compliance remained voluntary, and the gap between principle and practice widened.

The trajectory from Asilomar (2017) through the Pause Letter (2023) to the Statement on Superintelligence (2025) tracks Tegmark's assessment that guidelines, however well-crafted and broadly endorsed, cannot substitute for governance structures with enforcement authority. The progression is not a repudiation of Asilomar but a recognition of its insufficiency against the accelerating wisdom race.

Several Principles have aged particularly well as framings of the core challenges: Principle 10 (Value Alignment) specifies that highly autonomous AI systems should ensure their goals and behaviors remain aligned with human values throughout operation. Principle 15 (Shared Prosperity) addresses the distribution of AI's economic benefits. Principle 18 (AI Arms Race) warns against the arms-race dynamics that have since materialized. Principle 23 (Common Good) asserts that superintelligence should serve widely shared ethical ideals and the benefit of all humanity rather than one state or organization.

Origin

The Asilomar conference was convened by the Future of Life Institute in January 2017, building on earlier meetings including a 2015 Puerto Rico conference on AI safety. The conference format involved small-group drafting sessions followed by plenary debate, with principles requiring ninety percent consensus to be included. The final twenty-three principles were released with endorsements from over a thousand AI researchers and eleven hundred other signatories including physicists, philosophers, and technology leaders.

Key Ideas

Twenty-three principles. Covering research, ethics and values, and long-term issues including superintelligence.

Consensus threshold. Each principle required ninety percent agreement among conference participants.

Aspirational framing. Guidelines rather than mandates, reflecting 2017's less urgent landscape.

Historical echo. Named for the 1975 Asilomar Recombinant DNA Conference that set biotechnology norms.

Trajectory origin point. The baseline from which Tegmark's policy escalations have proceeded.