2023 pause letter on giant AI experiments

The Future of Life Institute dates Pause Giant AI Experiments: An Open Letter to March 22, 2023. Its hosted PDF separately records May 5 as the PDF creation date; the two dates describe different stages.^[1]

The proposed pause

The appeal asks laboratories to suspend training beyond GPT-4 capability for at least six months, with public verification and government intervention if needed. It proposes independent safety review and governance measures, while allowing other AI development to continue. This documents a requested policy, not an implemented moratorium.^[1]

The PDF’s selected roster includes Stuart Russell and Yoshua Bengio (p.3). It is a sample rather than the complete signature list. Signing this letter does not establish endorsement of every later FLI position.^[1]

A signatory explains his reasons

In Science Friday’s April 7 interview with Ira Flatow, Russell discussed difficulty assuring the safety of opaque systems. He separated immediate risks from possible loss of control over more powerful future systems, qualifying the latter as not a known property of the then-current systems. This is his stated rationale; the interview does not independently verify its incident examples or forecasts.^[2]

Academic context and limits

Concrete Problems in AI Safety gives a narrower technical grounding: harmful behavior can arise from inadequate objectives, costly supervision or failures during learning. Its agenda focuses on accidents and does not establish a probability of extinction or validate a six-month pause.^[3] Connecting its mechanisms to this policy appeal is an editorial comparison, not evidence that the paper endorses the letter.

The later 2023 statement on AI risk makes a different public appeal: prioritize mitigation of extinction risk. Neither document is an empirical safety evaluation.