Shane Legg, co-founder of Google DeepMind, warns that AI development speed must not outpace safety measures. He disagrees with claims from OpenAI president Greg Brockman and Nvidia CEO Jensen Huang that the era of Artificial General Intelligence (AGI) has already arrived.
Legg maintains his 2028 prediction, estimating a 50% probability of achieving 'minimal AGI' by that year. He argues that current models, including OpenAI's Astra, have not yet met the company's own criteria for AGI, which defines it as the ability to perform most economically valuable cognitive tasks.
DeepMind has launched the DeepMind Institute to study the deployment of AGI and its societal impacts. Led by Shane Legg, Demis Hassabis, and James Manyika, the institute aims to translate technical research into public-facing content and facilitate broader discussions on AI safety and policy.
The institute's inaugural publications include a critique by researchers Rohin Shah and Anca Dragan regarding the transparency of GPT-6 Astra. They suggest that competitive pressures may reduce transparency, and advocate for regulatory intervention to ensure safety mechanisms are integrated during the design phase.
