Skip to main content

DeepMind's Legg Pushes Back on AGI Hype: Astra Isn't There Yet

DeepMind's Legg Pushes Back on AGI Hype: Astra Isn't There Yet

A war of words has erupted over whether artificial general intelligence is already here. Nvidia CEO Jensen Huang recently congratulated OpenAI on its new Astra model and declared AGI has arrived. Not so fast, says Shane Legg, co-founder and chief AGI scientist at Google DeepMind.

In a new interview, Legg pushed back hard on the celebration. He pointed out that OpenAI's own bylaws define AGI as "a highly autonomous system that outperforms humans at most economically valuable work." Astra can do plenty of impressive things, but whether it truly handles most economically valuable cognitive tasks remains an open question.

Legg has been thinking about AGI longer than almost anyone. He still sticks to his long-held prediction: about a 50% chance we reach "minimum AGI" before 2028. DeepMind's bar is even higher—a system with all the cognitive abilities of the human brain.

Safety Can't Be an Afterthought

Legg didn't just throw cold water on the hype. He also sounded an alarm. AI capabilities are racing ahead, but safety mechanisms and control frameworks are lagging behind. Without enough understanding and risk management, he warned, we could find ourselves facing far more powerful systems with no preparation at all.

To tackle that gap, DeepMind announced the creation of the DeepMind Institute. Legg will lead it alongside DeepMind chairman Demis Hassabis and Google senior vice president James Manyika. The institute will study how rapid AGI development affects scientific research, education, social structures, and human well-being.

So the next time someone tells you AGI is already here, remember Legg's caution. The race is real, but the finish line? We're not there yet.

Key Points

  • Shane Legg disputes Jensen Huang's claim that OpenAI's Astra has achieved AGI.
  • OpenAI's own definition requires outperforming humans on most economically valuable tasks—Astra hasn't proven that.
  • Legg maintains a 50% chance of minimum AGI before 2028, with DeepMind setting an even higher bar.
  • DeepMind launched a new institute to study AGI's societal impacts and ensure safety keeps pace.