I don't know if we could do it the same way, but we could possibly learn things from how humans do it now. E.g. if one metric is "is the person holding a gun?"

I don't know if we could do it the same way, but we could possibly learn things from how humans do it now. E.g. if one metric is "is the person holding a gun?" then machine vision software to identify guns might be useful.