‘Catastrophic outcomes’ could accompany rise of advanced AI – Google DeepMind CEO
The path towards creating artificial general intelligence (AGI) could involve “catastrophic outcomes” such as cyberattacks on energy or water infrastructure, Google DeepMind CEO Demis Hassabis has warned. He suggested that AGI could arrive within the next decade.
Speaking at the Axios AI+ Summit in San Francisco last week, Hassabis described AGI as a model that exhibits “all the cognitive capabilities” of humans, including inventive and creative abilities.
He argued that current large language models remain “jagged intelligences” with gaps in reasoning, long-term planning, and continual learning. However, he suggested that AGI could soon become a reality with continued scaling and “one or two more big breakthroughs.”
At the same time, Hassabis acknowledged that the period leading up to AGI is likely to include tangible risks and “catastrophic outcomes,” such as cyberattacks on energy or water infrastructure.
“That’s probably almost already happening now… maybe not with very sophisticated AI yet,” he said, calling this the “most obvious vulnerable vector.” He added that bad actors, autonomous agents, and systems that “deviate” from intended goals all require serious mitigation. “It’s non-zero,” he said of the possibility that advanced systems could “jump the guardrail.”
Hassabis’ concerns echo broader warnings across the tech industry. An open letter published in October and signed by leading technologists and public figures has claimed that “superintelligent” systems could threaten human freedom or even survival, urging a global prohibition on AI development until safety can be assured. Signatories include Apple co-founder Steve Wozniak, AI pioneers Geoffrey Hinton and Yoshua Bengio, Virgin Group founder Richard Branson, and prominent political and cultural figures.
Others have taken a more optimistic view. Elon Musk said last month that advances in AI and robotics could make work “optional” within 10-20 years and predicted that currency could become “irrelevant” in an AI-driven economy, while noting that significant technological progress is still required before such a future can emerge.
