DeepMind CEO: AGI still lags in three areas

Asked at an AI summit in New Delhi whether current AGI systems can match human intelligence, DeepMind CEO Demis Hassabis replied, "I don't think we are there yet." AGI is a hypothetical form of machine intelligence that can reason like people and solve problems using methods it was not trained in.

He named three shortcomings. First, "continual learning"—today's systems are frozen based on the training they received before implementation and do not continually learn online from experience, personalise to context, or adapt to specific tasks. Second, long-term planning: "They can plan over the short term, but over the longer term, the way that we can plan over years, they don't really have that capability at the moment." Third, consistency: systems can excel in some areas yet stumble on elementary problems, producing a jagged performance profile that a true general intelligence should not show.

Hassabis told "60 Minutes" last year that true AGI would arrive in five to 10 years.

India, New Delhi

agi, demis hassabis, deepmind, continual learning, long-term planning, consistency, new delhi, 60 minutes, online learning, task adaptation