DeepMind CEO: AI comparable to the human brain will emerge in 5-10 years

On March 17 local time, DeepMind, Google's artificial intelligence research lab, held a press conference at its London office, where CEO Demis Hassabis made shocking statements, claiming that artificial general intelligence (AGI, AI that is as smart as or even smarter than humans) will begin to emerge in the next five to ten years:

“I believe today’s AI systems are still very passive, and there are many things they cannot accomplish. But in the next five to ten years, many of these capabilities will begin to manifest, and we will start to move towards what is known as artificial general intelligence.”

Hassabis defined AGI as “a system that can exhibit all the complex capabilities of humans.” His prediction is much more aggressive than the conservative estimates of most experts, suggesting that the so-called "AI singularity" may arrive several years earlier than expected.

CNBC noted that Hassabis's prediction is particularly noteworthy because DeepMind, as the core force behind Google's AI research, has already demonstrated its accurate grasp of the trajectory of AI development through groundbreaking achievements in projects like AlphaGo and AlphaFold.

What is needed to achieve AGI?

Hassabis believes that the main challenge in achieving AGI lies in enabling current AI systems to truly understand the contextual information of the real world. Although there are already AIs capable of autonomously breaking down problems and completing tasks, such as breakthroughs in the complex strategy board game Go, applying this technology to the real world remains fraught with difficulties. Hassabis stated:

“The question is, how quickly can we scale planning concepts and autonomous behaviors—including planning and reasoning capabilities—into the real world, while integrating ‘world models,’ which are models that can understand the surrounding environment?

In the past few years, we have made good progress in world models. The current question is how to best combine these world models with planning algorithms?”

Hassabis and Google Cloud CEO Thomas Kurian stated that the so-called “multi-agent” AI systems are rapidly developing behind the scenes, becoming an important technological advancement. Kurian remarked:

“When you think about communication between agents, we are exploring ways to enable agents to express themselves… For example, what are their skills? What tools do they use? These are key factors that allow you to ask questions to the agents, and once such an interface is established, different agents can communicate with each other.”