Demis Hassabis outlines the timeline for the development of AGI at a recent briefing in London.
Demis Hassabis, CEO of Google DeepMind, announced that he believes artificial general intelligence (AGI) could emerge within the next five to ten years.
Speaking during a briefing in London, Hassabis stated that while current AI systems are still largely passive and limited in their abilities, significant advancements are on the horizon.
Hassabis defined AGI as a system capable of exhibiting complex abilities akin to human intelligence.
He indicated that artificial superintelligence (ASI), which is expected to surpass human intelligence, would follow AGI, although he acknowledged that the exact timeline for such breakthroughs remains uncertain.
One major challenge in achieving AGI, according to Hassabis, is enhancing AI systems' capacity to understand real-world context.
Although AI has successfully tackled problems and completed tasks in controlled environments, its application in more dynamic, real-world situations presents greater difficulties.
He emphasized the importance of generalizing planning and reasoning concepts for effective real-world application, moving towards AI models that can comprehend and interact with the world around them.
Hassabis noted significant progress has been made in developing world models in recent years, and he highlighted ongoing efforts to combine these with planning algorithms.
Multi-agent AI systems, which can cooperate and compete in environments, are being explored as a promising technological advancement in this area.
As an example, he referenced DeepMind's work on training AI agents to play the strategic game 'Strechkraft,' where agents operate in a competitive and collaborative manner.
This approach is seen as a path towards more advanced AI capabilities.
Contrasting Hassabis's predictions, last year, Robin Li, CEO of the Chinese tech giant Baidu, estimated that more than ten years would be required for AGI development, dismissing shorter timelines proposed by some peers.
Conversely, Dario Amodei, CEO of the AI startup Anthropic, indicated at the World Economic Forum in Davos that a form of AI superior to most humans in almost all tasks could appear within two to three years.
Cisco's Vice President of Digital Communications Technology, Jetu Patel, posited that AGI could potentially materialize this year, outlining three developmental stages of AI: basic AI, AGI, and superintelligence.
Patel suggested that we may see significant evidence of AGI development in the current year, while he estimated that superintelligence could be expected within a few years.
Furthermore,
Elon Musk, CEO of
Tesla, previously forecasted that AGI might be realized by 2026, and Sam Altman, CEO of OpenAI, indicated that such systems could be developed in the near future.