Google has launched the DeepMind Institute to facilitate discussions surrounding Artificial General Intelligence (AGI). The organization is led by Shane Legg, James Manyika, and Demis Hassabis, with Legg serving as the managing editor.
The institute's initial release includes four research papers covering economic policy, human prosperity, model evaluation methodologies, and the maintenance of human-readable reasoning paths in AI systems.
Researchers emphasize the need for transparency in AI models. They propose limiting 'opaque sequential depth' and requiring developers to demonstrate that less transparent systems remain under effective human control.
Demis Hassabis proposed establishing an AI standardization body. This entity would conduct pre-release testing of advanced models, with developers submitting systems 30 days before deployment. Eventually, the body would implement independent, closed-source testing protocols.
The proposed regulatory framework suggests that if safety concerns escalate, the standardization body could mandate a slowdown in the development of advanced AI systems to ensure compliance with safety standards.
