Google Deepmind CEO Demis Hassabis expects the first forms of Artificial General Intelligence (AGI) to emerge within five to ten years, though significant technical challenges remain.
Speaking at Deepmind’s London office (via CNBC), Hassabis emphasized that current AI systems, while impressive in specific tasks, still lack many essential capabilities.
“These systems are very impressive at certain things. But there are other things they can’t do yet, and we’ve still got quite a lot of research work to go before that,” Hassabis explained, defining AGI as “a system that’s able to exhibit all the complicated capabilities that humans can.”
Building real-world understanding
The path to AGI requires developing AI systems that truly understand the physical world, Hassabis says. While researchers have successfully created autonomous planning and problem-solving capabilities for contained environments like the game of Go, transferring these abilities to real-world scenarios presents a much greater challenge.
Ad
“The question is, how fast can we generalize the planning ideas and agentic kind of behaviors, planning and reasoning, and then generalize that over to working in the real world, on top of things like world models — models that are able to understand the world around us,” Hassabis says.
Changing AGI Perspectives
According to the Deepmind CEO, researchers have made good progress on world models recently, but finding the optimal way to combine them with planning algorithms remains a key challenge. Meta’s leading AI researcher Yann LeCun shares Hassabis’s more cautious outlook and is pursuing his own development approach.
Hassabis’s statement aligns with his earlier position from August 2024 that current AI capabilities are often overestimated, while the technology’s long-term potential remains underestimated. On median, users of the Metaculus forecasting platform expect AGI to arrive by 2030.
The AI industry increasingly questions whether AGI will arrive through simply scaling up existing technologies. According to a recent survey, the majority of AI researchers don’t believe AGI can be achieved with LLMs.
OpenAI has recently moved away from its earlier position that emergent capabilities in ever larger AI models would lead to sudden breakthroughs in AGI. The company now describes the transition to AGI as a gradual process of continuous evolution.
Microsoft CEO Satya Nadella has dismissed AGI predictions as “nonsensical benchmark hacking,” suggesting that AI development should focus instead on delivering concrete economic benefits.