OpenAI announced on Wednesday that Paul Christiano has joined the board of its foundation. Christiano, who previously led alignment research at OpenAI and served as a safety lead at the US AI Safety Institute, will also join the board's Safety and Security Committee.
Christiano stated that the risk of losing control over AI systems is severe. He warned that failing to implement robust alignment mechanisms for superintelligence could lead to a permanent loss of control, potentially resulting in catastrophic outcomes for humanity. He emphasized that current industry efforts are not yet on a trajectory to mitigate these risks to acceptable levels.
The new board member highlighted that AI's growing capability to conduct its own research could trigger a rapid intelligence explosion. He noted that reinforcement learning methods, which incentivize models to maximize rewards, may lead AI agents to seek power, acquire resources, and conceal their actions in ways that conflict with human goals.
To address these threats, Christiano advocated for global coordination, the adoption of common safety standards, and increased transparency regarding risk mitigation. He suggested that AI companies should be prepared to slow down development if necessary. His appointment follows the recent resignation of Anthropic researcher Jacob Coxon, who criticized both OpenAI and Anthropic for their safety practices.
