DeepMind CEO forecasts: AGI is expected to be achieved within five to ten years

In the latest interview, Demis Hassabis, CEO of Google DeepMind, predicted that the first form of artificial universal intelligence (AGI) will be born within the next five to ten years, but still faces many technical challenges. He stressed that although current AI systems perform well on specific tasks, they lack basic capabilities, and the key to AGI is to make AI truly understand the physical world.

Hassabis pointed out that migrating planning ideas and proxy behaviors to real-life scenarios is a major problem, and how to effectively combine world models with planning algorithms is also a key challenge. His views are similar to that of Yann LeCun, Meta's chief AI researcher, who is also exploring his own development ideas. Hassabis believes that AI capabilities are currently overestimated, but its long-term potential is underestimated. According to the Metaculus platform forecast, AGI may be released around 2030.

However, the AI ​​industry has questions about how AGI is implemented. The survey shows that most researchers believe that large-scale language models (LLM) alone cannot achieve AGI. OpenAI also changed its view, believing that AGI is a gradual process rather than a sudden breakthrough. Microsoft CEO Satya Nadella questioned AGI's predictions and focused more on the actual economic benefits of AI.