How AI is changing warfare, and why its use in nuclear decisions raises concerns

The military use of artificial intelligence (AI) is advancing rapidly across the globe, according to experts. In the United States, the Department of Defense has expressed a policy of actively utilizing AI for tasks such as setting attack targets, operating drones, and performing various other functions. China and Russia are also believed to be incorporating AI-equipped drones into their military operations.

Japan's Defense of Japan 2026 white paper, approved by the Cabinet on August 4, acknowledges the role of AI in new ways of warfare. It highlights that AI can facilitate faster and more accurate decision-making compared to opponents.

The question remains: is it safe to entrust decision-making to AI? Japan's Defense Ministry asserts that AI will serve as a support tool, but there is an ongoing discussion about the extent of AI's role. A paper released in February by King's College London provides a chilling insight into the matter. Professor Kenneth Payne, from the university, conducted one-on-one wargame simulations using three AI models under various conditions.

In over 90% of the 21 games, low-yield nuclear weapons, known as "tactical nuclear weapons," were employed. Additionally, there were three instances that resulted in full-scale nuclear war. The AI did not opt for peaceful resolutions or surrender.

The paper suggests that while AI can learn human psychology and strategy, the taboo against nuclear use, which has persisted since 1945, is not adequately reflected in its reasoning. Professor Payne emphasized the importance of understanding how advanced AI models emulate or fail to mimic human strategic logic. This knowledge is crucial as AI increasingly influences strategic outcomes in the modern world.