DeepMind CEO reveals 3 areas where AGI falls short of human intelligence

True artificial general intelligence (AGI) is on the horizon, but current systems still lag behind human intelligence, said Google DeepMind CEO Demis Hassabis at an AI summit in New Delhi. 

AGI refers to hypothetical machine intelligence capable of reasoning and problem-solving in ways it wasn’t explicitly trained for.

Hassabis identified three key limitations in today’s AGI systems. First is continual learning: current systems are largely static, unable to learn online from new experiences or adapt to changing contexts. 

“What you’d like is for those systems to continually learn, maybe personalise to the situation and tasks at hand,” he explained.

Second, he noted a lack of long-term planning. While AGI can handle short-term tasks, it struggles to plan over months or years in the way humans can.

Finally, Hassabis highlighted inconsistency. AGI systems may excel in certain areas—such as solving complex problems in the International Math Olympiad—yet still make simple mistakes on elementary tasks. 

“A true general intelligence system shouldn’t have that kind of jaggedness,” he said, noting that human experts rarely make such errors in their specialty.

Hassabis previously estimated that true AGI could arrive within five to ten years. DeepMind, founded in 2010 and acquired by Google in 2014, is also behind the development of Google’s Gemini AI.

The AI Summit in India, running this week, has attracted major tech leaders including OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, Google CEO Sundar Pichai, and Meta’s chief AI officer Alexandr Wang.