>last year, in which Centcom experimented with an AI recommendation engine, showed such systems “frequently fell short” of humans in proposing the order of atta

>last year, in which Centcom experimented with an AI recommendation engine, showed such systems “frequently fell short” of humans in proposing the order of attack or the best weapon to use. I wonder if it's stupid risky to actually train AI on actual weapon behaviours, let alone all of them to optimize coordination. That seems like the crown jewel of leaks to infer actual US capabilities. It's one thing to embed a model with info about adversary targetting, but once you include your own capabilities to engage those targets, which I presume you would have to, then those models also becomes a huge liability.