Experts in the field of nuclear warfare are increasingly convinced that artificial intelligence will soon play a crucial role in the management and potential deployment of lethal weapons. However, there remains significant uncertainty regarding the specific implications of this integration.
In mid-July, Nobel laureates convened at the University of Chicago for a series of discussions featuring specialists in nuclear conflict. Over the course of two days, a diverse group of scientists, former government officials, and retired military leaders provided insights into the catastrophic capabilities of contemporary nuclear arsenals. The objective was to inform these esteemed figures about the existential threats posed by these weapons and subsequently encourage them to propose policy guidelines to world leaders aimed at preventing nuclear conflict.
Artificial intelligence was a recurring theme throughout the discussions. “We are entering a new era where artificial intelligence and emerging technologies will not only shape our everyday lives but also influence the nuclear landscape,” stated Scott Sagan, a Stanford professor recognized for his work on nuclear disarmament, during a press briefing concluding the sessions.
This assertion reflects a broad consensus among attendees in Chicago regarding the inevitability of governmental use of AI in the nuclear domain. “It’s akin to electricity; it will permeate everything,” remarked Bob Latiff, a retired US Air Force major general and member of the Bulletin of the Atomic Scientists’ Science and Security Board. Latiff plays a key role in determining the setting of the Doomsday Clock each year.
Jon Wolfsthal, a nonproliferation authority and director of global risk at the Federation of American Scientists, highlighted significant challenges in the discourse surrounding AI and nuclear weapons. “One major issue is that there remains a lack of clarity about what AI actually entails,” he noted.
Wolfsthal raised crucial questions regarding the implications of granting AI or computational systems control over nuclear armaments. “What does it mean to assign a computer chip this level of authority?” posed Herb Lin, a Stanford professor and distinguished member of the Doomsday Clock team. He pointed out that the emergence of large language models has obscured these fundamental inquiries.
Fortunately, experts agree that sophisticated AI tools like ChatGPT or Grok are not likely to access nuclear codes in the near future. According to Wolfsthal, while there are various differing perspectives within the nuclear expert community, there is a shared understanding that effective human oversight is essential in nuclear decision-making processes.
Nonetheless, Wolfsthal mentioned concerning trends regarding the potential deployment of large language models in high-stakes environments. “There are indications that some officials envision an interactive AI system that could provide the president with reliable predictive models of actions taken by adversaries like Putin or Xi, utilizing a vast repository of their statements and writings,” he stated.
