Peter Schweighofer

#Wargame #simulations can’t predict the future, but they can explore outcomes. These findings aren’t encouraging. Putting complex, nuanced political/military decisions in the hands of AI is just asking for annihilation. I thought there was a sci-fi film series based on this premise.... 1/2

https://www.newscientist.com/article/2516885-ais-cant-stop-recommending-nuclear-strikes-in-war-game-simulations/

AIs can’t stop recommending nuclear strikes in war game simulations

Leading AIs from OpenAI, Anthropic and Google opted to use nuclear weapons in simulated war games in 95 per cent of cases

New Scientist
Feb 26 at 1:32pmWeb
Show threadPeter SchweighoferFeb 26

The Paxsims blog offers additional context for this issue, including a link to the rather comprehensive paper on the study the news article reviewed, a video about the pape author’s views on AI, strategy, and war, and, in a sadly disappointing turn, ChatGPT’s own assessment (AI: DR). 2/2

https://paxsims.wordpress.com/2026/02/25/payne-ai-and-simulated-nuclear-crises/

Payne: AI and simulated nuclear crises

A recent paper by Kenneth Payne (King’s College London) on the strategic behaviour of various LLM AI models in a wargame is receiving considerable attention. Today’s leading AI models engage …

PAXsims
Show threadcybergothsFeb 26
@PSchweig86 But didn’t #Wargames (1983) cover this earlier than the series of films(1984 onwards…)?
Show threadPeter SchweighoferFeb 26
@cybergoths Quite true. And in fact the study author mentions it in the video on the Paxsims blog. But in that case the "AI" backed down, whereas in the other series the AI rapidly escalated into launching hte nukes. Obviously Wargames did a more Hollywood cerebral dive into the issue than Terminator....