Google DeepMind CEO: The next Einstein could be an AI

Contents

It is the eternal promise of the industry and the fig leaf to justify burning billions of dollars and vast amounts of energy until then: the search for AGI. Artificial General Intelligence (Künstliche allgemeine Intelligenz) is considered the holy grail of AI researchers. When software learns to learn without human help, unimaginable progress is said to be possible. Some also fear that AI will then turn against humans. But increasingly, people consider this future vision a mirage and the path to it still very long.

Demis Hassabis, head of Google's AI division, DeepMind, believes AGI is achievable. But only in five to ten years. In an interview with Alex Kantrowitz of Big Technology, he didn't spare his competitor OpenAI and its CEO, Sam Altman. He warns that AGI should not be used as a marketing term. General artificial intelligence is a system that can exhibit all the cognitive abilities that humans have, he says, “And I mean all.”

What AGI must be able to do

But where is the benefit for humanity? This form of AI would only have it when it helps humanity achieve new breakthroughs, Hassabis explains. It's not enough to solve a mathematical equation or a conjecture. Groundbreaking conjectures are needed, a new Einstein. Or in the artistic field, a Picasso, or a Mozart. And with abilities and a pace that would be hardly or not at all possible for humans.

And although he holds the view that the capabilities of today's AI models have not yet been fully recognized and exploited, Hassabis is convinced that they are still far from AGI. Today's AI has a “goldfish brain.” It can search the internet, but this knowledge does not change the model and is forgotten after the session. A superintelligence would go even further; it could integrate other systems like weather satellites or think in 14 dimensions – things that no human would be capable of.

Several breakthroughs needed

AI still needs several breakthroughs on the path to AGI: in addition to continuous learning, these are more efficient context windows and long-term planning. While the human brain processes only what is important with selective attention, AI treats all information in the context equally. This is inefficient and expensive.

In the stage conversation in Davos, Hassabis also addressed whether Google, like OpenAI, plans to integrate advertising into its chatbot. Hassabis praised the start-up Anthropic, in which Google has already invested billions of dollars. Their development tool, Claude Code, is very successful. Google itself wants to better showcase the capabilities of its AI model Gemini with the newly released IDE Antigravity.

Google's Smart Glass Plans

Hassabis specifically commented on smart glasses: Google is working with partners such as Warby Parker, Gentle Monster, and Samsung on a new generation of AI-powered glasses, which are expected to be launched on the market “perhaps by summer.” Unlike the failed Google Glass more than ten years ago, both the form factor is now mature and, crucially, the “killer app” is available: a universal digital assistant that helps with everyday tasks hands-free. Hassabis himself is personally involved in the project.

(mki)