DeepMind CEO: AGI Still Falls Short of Human Intelligence

Despite rapid advances in AI, Demis Hassabis, DeepMind's CEO, emphasised at the AI summit in New Delhi that AGI remains far from achieving human-level reasoning.

While current systems can perform remarkable tasks, the CEO says they continue to face limitations in three key areas: continuous learning, long-term planning and consistency across tasks.

AGI is defined as machine learning capable of reasoning and problem-solving in context it hasn't been trained for. When asked if AI could match human intelligence today, Demis answered: "I don't think we are there yet."

Demis emphasised the challenge of continual learning. He said: "What you'd like is for those systems to continually learn online from experience, to learn from the context they're in, maybe personalise to the situation and the tasks that you have for them."

He illustrated the gap between human and AI creativity with an example from physics: "One way we maybe would test that is you could imagine training a foundation model with a knowledge cutoff of something like 1911 and then see if it could come up with general relativity like Einstein did in 1915.

"But that would be a good test for AGI and I think today's systems clearly would not be capable of doing that."

Long-term planning and consistency across tasks

Demis noted that AI models are limited to planning horizons. "They can plan over the short term, but over the long term, the way that we can play over years, they don't really have that capability at the moment."

While current systems can solve immediate problems effectively, they lack the ability to project strategies over months or years, he says, which is a core component of human intelligence.

Demis outlines that another major gap is inconsistency. "So, for example, today's systems can get gold medals in the International Math Olympiad, really hard problems, but sometimes can still make mistakes on elementary maths if you pose the question in a certain way," he said.

"A true general intelligence system shouldn't have that kind of jaggedness."

AI as a co-researcher

Demis sees AGI first acting as a tool to augment human research. He predicts a "new golden era for scientific discovery" in the coming decade. "The next phase is going to be incredible for human experts and scientists," he said, adding that "having a tool like AI will really help scientists learn about and be able to understand and process all of that information from multiple different domains".

He described how general AI systems like Google Gemini would collaborate with specialised models like AlphaFold: "If Gemini wanted to or needed to understand the structure of a protein, I think it would be better for it to call AlphaFold as a tool than put all that protein information into the main system."

Demis also addressed risks. Beyond the potential misuse by individuals or nations, he warned about technical and societal challenges: "There's a societal challenge that may actually end up being the harder problem than the technical ones."

Near-term concerns include cyber and biosecurity, and establishing international standards for safe AI deployment, the CEO outlines.

Despite the hurdles, Demis remains cautiously optimistic about AGI's trajectory. "I would say cautious optimism, if the best minds work towards that, I think we'll solve the technical risks," he said.

He also highlighted the global potential of AI: "The generation that grows up native with that technology will end up doing some sort of incredible things that we can only dream of right now."

While AGI has not yet reached human-level intelligence, Demis' remarks underscore both the promise of AI in augmenting human capability and the prudence needed to manage its risks responsibly.