Definition
The Asilomar Guidelines are a landmark set of self-regulatory principles for scientific research. Originally established in 1975 for recombinant DNA technology, they serve as the primary historical precedent for scientists voluntarily halting dangerous research to establish safety protocols before proceeding. Stuart Russell uses this as a model for how the AI community might “own” the risks of superintelligence.
Why It Matters
They prove that scientists can take responsibility for the risks of their research before external catastrophes force a reaction. They serve as a vital blueprint for how the AI community might voluntarily govern superintelligence risks.
Core Concepts
- The DNA Moratorium (1973-1975): Biologists voluntarily paused research on splicing genes to assess risks, leading to the NIH Recombinant DNA Advisory Committee (RAC). This proved that scientists can refrain from a hazardous area (like human germline alteration).
- Asilomar AI Principles (2017): A modern set of 23 principles for AI development, including safety, transparency, and human values alignment.
- The “Whiteboard” Challenge: Russell notes that AI is harder to regulate than DNA because progress occurs on “whiteboards” and “tool AI” (e.g., game playing) naturally leads to “general AI” techniques.
- Relinquishment vs. Control: While the DNA ban on germline alteration has been largely successful, the “Standard Model” of AI makes control difficult because it ignores the inherent uncertainty of human goals.