Google and Google DeepMind have launched the DeepMind Institute to facilitate discussions on artificial general intelligence. The organization is led by directors Shane Legg, James Manyika, and Demis Hassabis, with Legg acting as managing editor.
The institute aims to highlight diverse perspectives on AGI from Google, DeepMind, and the global research community. Its first publication features four essays addressing economic policies, model reasoning, human flourishing, and evaluation frameworks for frontier AI.
Researchers Rohin Shah and Anca Dragan discuss the declining transparency of AI models. They suggest that developers and regulators should address safety trade-offs by limiting opaque sequential computation or requiring proof that less transparent systems remain monitorable.
Demis Hassabis proposes the creation of a U.S.-led standards body to evaluate advanced AI models. Under his plan, developers would initially submit models for review 30 days before release, with the process potentially becoming mandatory for deployment.
The proposed body would eventually implement independent, undisclosed tests to prevent model tailoring. Hassabis suggests that this framework could include a coordinated slowdown of AI development if safety concerns escalate.
