AI models choose violence ?

[OneStepBack]

Yeeh boundary , thats good to have , but  how good boundary....

the problem with military AI is even small error could cause big danger potentially,  wrong target etc...

Last night I remembered someone in AI being interviewed.  He talked about a study he did did on problem solving.  The AI apparently used pathways it was not programmed to use in solving a specific problem ie it created these new pathways itself.  I tried looking for the interview today but couldn't find it. It may have been on YouTube.  My mind isn't good these days.  I will keep looking for it.

As far as military warfare goes it could potentially open a pandora's box.  Scary stuff.