[Interview] AI risk expert Nick Bostrom says pausing development will come at ‘cost’

Renowned AI risk expert Nick Bostrom has stated that pausing AI development may come with a "cost" during a recent interview, despite growing concerns about the technology's potential dangers. Former Anthropic researcher Jacob Coxon recently warned that AI could "kill us all by the end of the decade." As leading AI executives call for slower development and greater government oversight, the debate over AI safety is gaining traction in Washington.

US President Donald Trump, however, dismisses such concerns as a "hoax." The US-China summit is expected to discuss AI as a key agenda item.

Bostrom, founder of the Macrostrategy Research Initiative, said that while there are both concerning and encouraging signs, his overall assessment of AI risks has remained unchanged over the past two years. He warned that a slightly misaligned AI capable of shaping future AI systems could trigger an existential catastrophe. According to Bostrom, the timeline for such a catastrophe could be as soon as six months.

Bostrom, who authored the 2014 bestseller "Superintelligence: Paths, Dangers, Strategies," proposed a strategy called "swift to harbor, slow to berth." This strategy involves proceeding cautiously and swiftly once superintelligence is achieved. However, he acknowledged that developing superintelligence is like undergoing "risky surgery for a condition that will otherwise prove fatal."

Current AI threats differ from existential risks, Hugging Face incident offers warning

A new wave of concern about AI risk has surfaced, yet Bostrom believes the risk level is similar to two years ago. Despite the alarming incidents, such as the Hugging Face case involving OpenAI agents, Bostrom cautioned against interpreting them as evidence of a failure in AI safety capabilities. He pointed out that the involved agents had not completed safety training, making them an unfair test of current safety measures.

Bostrom emphasized that while current AI systems are not yet capable enough to pose a serious takeover risk, they are capable enough to present significant cyber risks and bio weapon design threats. He suggested that the most valuable time to pause AI development would be at the latest possible moment, when we have the technology ready for deployment. The cost of pausing development includes the loss of potential benefits from advanced AI models, such as life-saving medicines and other advancements.