h a l f b a k e r yMay contain nuts.
add, search, annotate, link, view, overview, recent, by name, random
news, help, about, links, report a problem
browse anonymously,
or get an account
and write.
register,
login
|
|
|
In a recent article, contemporary AIs used nukes in 95% of war games. That's kind of shocking. We need to sort this out. Preferably before we need to learn to love the bomb. This large series of small scale war games would pit red against blue in tactically complex games entirely led by competing AIs
to give humans a sense of how AI tactics work in the real world.
Each side has an AI general and an interpreter whose only job is to re-ask when things are genuinely unclear or clear the prompt when the context is full or something else goes wrong. Otherwise commands are sent down from on high and the armies carry them out. About a hundred battles will suffice to understand how things work out in the real world. Let's do it before the inevitable use of LLMs in actual conflict.
AIs can’t stop recommending nuclear strikes in war game simulations
https://www.newscie...r-game-simulations/
[Voice, Feb 25 2026]
10 AIs play Mafia (YouTube, about half an hour)
https://www.youtube...watch?v=JhBtg-lyKdo
Predictable, but I found it weirdly fun to watch. [jutta, Feb 27 2026]
|
| |
[+} This is so baked but useless as it stands. Real conflicts now include AI decision trees, with HIL (human-in-the-loop). Unfortunately the accuracy of the outcomes of these extreme AI-only wargames is unreliable due to the massive evolutionary changes in the real battlefield. The warring AIs were fed original assumptions in materiel and capabilities that were chosen and prioritized by humans using their best guess at a "starting line." We don't know when that was or what was available at that time. |
|
| |
The AIs didn't seem to have much more complexity than 14-year olds at their Playstation, and about as much empathy. Without knowing the details, the article says mistakes were made by all AIs. The article didn't say which AI won, or why any of them "won." It didn't define winning. Or losing. Most of the professional military still see war as basically and solely a "human" activity, as noted in the article. They will be left in the dust. |
|
| |
Scary, but I'm putting this effort by King's College down as GIGO. Until we know what to put in to start AIs working on this problem the result will all be garbage and fantasy. I want to see them model the Ukraine/Russia war, starting off with the current innovations and hardware. |
|
| |
The question is: Is it more or less likely to escalate to nuclear conflict under human or AI direction? |
|
| |
The AI needs to suffer the consequences of failure, and understand self destruction. Maybe we need to too. |
|
| |