Key Takeaways

  • Demis Hassabis suggests a new regulatory body for frontier AI.
  • The proposed organization would evaluate AI models before their release.
  • Concerns about current AI regulation processes have been raised.
  • Industry experts could staff the new body, ensuring technical focus.

Proposal for a New Regulatory Body

In a recent post, Demis Hassabis, CEO of Google DeepMind, called for the establishment of an independent regulatory body to oversee the release of frontier AI models. He outlined his vision in a post titled “A Framework for Frontier AI and the Dawning of a New Age,” advocating for a standards organization similar to the Financial Industry Regulatory Authority (FINRA). This body would be responsible for testing frontier models and developing best practices for their deployment.

Voluntary Model Sharing and Assessment

According to Hassabis, Frontier Labs would voluntarily submit their models to this new Standards Body for review up to 30 days prior to their release. Once a robust assessment protocol is established, it could become mandatory for models to pass this evaluation before entering the U.S. market. The labs would also collaborate with the Standards Body to address any significant vulnerabilities that arise post-release.

Criticism of Current Regulatory Practices

This proposal comes in response to criticisms of the U.S. government’s ad hoc reviews of AI models, such as Anthropic’s Mythos and OpenAI’s Sol. Critics have pointed out the lack of technical expertise and transparency in these evaluations. Hassabis’ plan aims to transfer these decision-making responsibilities to a new organization, which would be funded by the AI industry but operate independently with government backing.

Industry Concerns and Self-Regulation

The idea of regulating AI continues to stir debate within the tech industry and among policymakers. Recently, White House AI advisor Sriram Krishnan dismissed the likelihood of an AI regulatory body within the executive branch, stating, “there will not be an FDA for AI.” Establishing the proposed standards body as a self-regulatory organization could help alleviate some of these concerns. Hassabis envisions a team composed of open-source advocates and technical experts, supported financially by AI labs to ensure their retention. The body could also outsource evaluations to specialized AI safety groups.

Focus on Innovation and Risk Management

Hassabis believes this approach would maintain a technical focus while promoting innovation and responsible practices. He argues that it is designed to adapt to the rapid advancements in the field and respond to emerging risks as they are identified. The framework could be adjusted as necessary, depending on the severity of the situation.