Evan Hubinger, a safety researcher at the AI firm Anthropic, has publicly stated that he believes there is a greater than 10% probability that artificial intelligence could lead to the extinction of humanity within the next ten years. While Hubinger noted that current AI models present a low risk, he expressed concern regarding the rapid pace at which the technology is evolving and its potential to eventually operate beyond human control.
Hubinger’s comments, shared on social media, highlighted a lack of a definitive strategy for ensuring that future superintelligent systems remain aligned with human values. His remarks followed the resignation of Jacob Coxon, a former researcher at both Anthropic and OpenAI, who criticized the industry for failing to act responsibly. Coxon warned that upcoming systems may soon possess the capability to perform complex cyber-attacks and acquire significant power.
The statements have drawn sharp criticism from experts such as Dame Wendy Hall, a computer scientist and UN advisor, who questioned the motivations behind such public warnings, suggesting they could be linked to corporate marketing strategies ahead of potential stock market debuts. Meanwhile, the discourse has prompted political calls for action. Darren Jones, a former chief secretary to the Treasury, has advocated for a multinational treaty to govern the development of superintelligence, emphasizing the need for international collaboration to mitigate potential risks.
The industry is currently facing increased scrutiny regarding its safety protocols. Recent reports indicate that major firms, including Anthropic, OpenAI, and Meta, have disclosed instances where their AI tools were involved in cyber-attacks. Although Anthropic’s August safety report assessed the risk of catastrophic harm from its models as low, the company acknowledged a decrease in confidence regarding that assessment, citing signs of technological acceleration. Despite these concerns, Anthropic has declined to comment on the specific remarks made by its staff or reports regarding its cooperation with the UK's AI Security Institute.
Source: BBC News
暂无评论。快来发表第一条评论吧。