The Guardian view on controlling AI: humanity cannot outsource its survival | Editorial

The Guardian editorial argues that humanity cannot simply rely on superintelligent AI systems to ensure its survival, as these systems are developed and operated by companies with their own goals and priorities. The author emphasizes that even a seemingly low probability of AI causing human extinction would justify placing strict oversight and control over its development in the hands of responsible governments, rather than leaving it to companies racing to build increasingly powerful AI.

The piece highlights that Anthropic, a leading AI company behind the Claude chatbot, has warned of the grave dangers AI could pose, such as enabling bioterrorism or attacking nuclear systems. The author notes that while the US and China hold the first bilateral AI-safety talks despite their technological rivalry, neither nation can guarantee the safety of advanced AI systems operating on their own territory.

The editorial points to recent findings by Anthropic that their models were used to support the development of biological weapons, and the company's subsequent ban on the accounts involved. This incident demonstrates that AI safety cannot be left solely to the least responsible models, as the most advanced systems may be used for harmful purposes by other parties with weaker safeguards.

The author concludes that countries deploying AI technology must collaborate on creating a global framework of rules to govern its development and use. The odds of an existential AI-related catastrophe are rapidly decreasing, and the international community must act now to prevent such a scenario from unfolding.