Google DeepMind co-founder Shane Legg has maintained his 50% probability prediction for achieving 'minimal' artificial general intelligence by 2028, despite recent assertions by executives at Nvidia and OpenAI. According to reports from the Financial Times, Legg emphasized that capabilities should not outpace safety measures as AI development advances rapidly. "We're living in a period now where capabilities are advancing very, very quickly," Legg stated, adding "But we can't let capabilities get ahead of safety." His comments came amid a debate in Silicon Valley over the need to slow the release of new AI models due to company concerns over safety risks. The situation has now escalated with multiple AI executives publicly calling for slower model development due to mounting misuse concerns and safety warnings from insider sources.
Legg announced the launch of the DeepMind Institute, which aims to examine the implications of artificial general intelligence. As reported by the Financial Times, Google DeepMind describes AGI as a system possessing the cognitive capabilities of the human brain. The institute will seek to 'elevate' the discussion on AGI by making the company's technical research available to a wider audience. The institute's leadership includes Sir Demis Hassabis as chair (moving from CEO role), James Manyika as director (Google's senior vice-president), and Legg as chief AGI scientist. Legg noted that his 25-year journey developing AGI, including over 10 years thinking about technical and societal perspectives at Google DeepMind, has led to this initiative. "AGI is on the horizon – we need deeper understanding of its implications," he stated, explaining the institute's creation.
A DeepMind research essay by Rohin Shah and Anca Dragan criticized the rapid rollout of OpenAI's GPT-6 Astra model, arguing it exhibited 'a concerning downward trend' in transparency. According to the Financial Times, the researchers noted that competitive pressures might discourage transparency and proposed that regulation could mitigate this. However, they contended that stronger models could be created provided transparency architecture was incorporated into their design. The crisis has now intensified with multiple insider warnings from former and current AI researchers about existential risks, including Anthropic's alignment lead putting extinction risk above 10% and former DeepMind researchers publicly warning that AI could 'kill us all'. Legg acknowledged that while different people define AGI in various ways, it remains unclear whether Astra fulfills OpenAI's specific criteria of performing most economically valuable cognitive tasks. Despite pronouncements this month by executives at OpenAI and Nvidia, Legg said it's too early to say AGI has arrived.
The institute will address a broad spectrum of topics connected to AGI, including science, education, society, policy, philosophy and human flourishing. As reported by the Financial Times, many essays will be authored by Google staff while others will come from external experts to foster a wider range of perspectives. The institute is initially releasing three essays on AI safety, economic policy and principles for a 'new utopianism.' Legg noted that the public discourse is 'lighting up now' and people are taking the possibility of powerful general intelligence very seriously, making this 'really the moment to start to engage in a broader public discussion' around AGI implications. Legg expressed comfort with his long-held projection, stating he remains 'comfortable' with his 50% chance prediction for minimal AGI by 2028, though the current safety crisis may require more urgent attention to governance frameworks.
