The Asilomar AI Principles
In January 2017 the Future of Life Institute published the 23 Asilomar AI Principles, developed at its Beneficial AI 2017 conference at Asilomar. The principles are grouped under research issues, ethics and values, and longer-term issues, and were opened for signature.
Why it matters
AI researchers jointly signed a list in which privacy and transparency stand beside race avoidance, value alignment, recursive self-improvement and superintelligence. Six years later the pause letter rests on the twentieth principle: advanced AI could be a profound change in the history of life on Earth and should be planned for accordingly.
According to FLI's post, a statement was kept only if at least 90% of the conference's attendees agreed with it. Among the principles: 5, avoid a race that cuts corners on safety standards; 10, value alignment of highly autonomous systems; 16, human control over delegating decisions; 18, avoid an arms race in lethal autonomous weapons; 22, strict safety measures for systems that recursively self-improve; 23, superintelligence only in the service of widely shared ethical ideals. The record gives no count of signatories: in the January 2017 archived copy the counter and lists are drawn by script and empty. No document read gave the dates of the conference.