Facebook
Britain's News Portal
Around The Clock
BREAKING
Loading latest headlines…

DeepMind Institute launched to broaden AGI debate

Google and Google DeepMind researchers have launched the DeepMind Institute to foster public discussion on artificial general intelligence (AGI).

  • The DeepMind Institute was launched on Wednesday by Google and Google DeepMind researchers.
  • Its aim is to surface differing views on AGI from Google, Google DeepMind, and the global research community.
  • The inaugural collection of four essays covers topics including economic policies, human-readable model reasoning, and evaluating frontier AI models.

Google and Google DeepMind researchers launched the DeepMind Institute on Wednesday, 16 September 2026, to advance discussions around artificial general intelligence (AGI). Directors of the new institute include DeepMind co-founder Shane Legg, Google executive James Manyika, and Google DeepMind chair Demis Hassabis, with Legg also serving as managing editor.

The institute aims to bring forward diverse perspectives on AGI from within Google, Google DeepMind, and the wider global research community. An announcement stated that these views "will not always agree, and they will likely change their minds, as more data and information comes to light at the fast-moving frontier."

The initial collection of four essays addresses various subjects, such as economic policies for managing potential AGI disruption, the preservation of human-readable model reasoning, principles for human flourishing, and a framework for evaluating frontier AI models.

One essay, authored by DeepMind safety researchers Rohin Shah and Anca Dragan, suggests that the decreasing transparency of AI models is not inevitable. They argue that developers and regulators should directly address safety trade-offs as new architectures make powerful models harder to monitor. This could involve limiting "opaque serial depth" or requiring developers to demonstrate that less transparent systems remain monitorable.

In another essay, Demis Hassabis proposes a US-led standards body for frontier AI to evaluate advanced AI models. Under this framework, developers would initially submit models voluntarily for review up to 30 days before their release. Hassabis suggests that passing these tests could eventually become a requirement for deploying frontier models in the United States, once the evaluation system proves effective.

Why this matters: The launch of the DeepMind Institute and its initial essays signal a shift in the AI safety debate towards concrete proposals for disclosure, external scrutiny, and potential coordinated slowdowns in development.

Related Articles

Get the news that matters.

Join thousands of readers getting the best of British news straight to their inbox.