AI safety requires more than just slowing our pace | Stuart Russell

Safety requirements are essential and cannot be compromised. They are based on achieving specific objectives, rather than merely adjusting the speed at which progress occurs. Recently, AI safety has been under intense scrutiny due to a series of alarming revelations involving prominent AI companies. These events culminated in Jacob Coxon's resignation from Anthropic, which has sparked intense debate and concern within the AI community.

Amidst this turmoil, Dario Amodei, CEO of Anthropic, penned an extensive letter titled "We Must Pace the Frontier," outlining his vision for ensuring AI safety while allowing for progress. Sam Altman of OpenAI, Elon Musk of xAI, Demis Hassabis of Google DeepMind, and Satya Nadella of Microsoft have all expressed their support for these proposals.

The concept of "pacing the frontier" originated from Formula 1 racing, where a pace car is used to regulate speed by setting a safe limit for other vehicles. Amodei suggests that, similar to this, AI development should not halt entirely but instead follow a controlled pace. His proposed three-part plan includes establishing internal AI system evaluators, creating common safety standards for companies in democratic nations, and developing a broader agreement with authoritarian countries.

Amodei emphasizes that safety requirements must come first, and progress should only continue once these requirements are met. He advocates for a "red lines" approach, in which developers must ensure their models meet safety standards before moving forward, thus avoiding the disastrous consequences of unchecked AI development.