> They can. A neurotypical person's brain could spontaneously generate a violent psychotic episode due to a stroke or adrenergic tumor or somesuch. Just like any area of the sky could spontaneously throw a lightning strike at you at any time. It's low probability, certainly, but murder is a low-probability event to begin with.
I think if we eliminated all murders except these, we would be in excellent shape. My point is these are not the ones worth focusing on, because we don't have good tools to deal with "random, history free, psychotic break."
> The important point is that a model with "non-deterministic" people in it has more predictive power, epidemiologically, than a model where it's impossible to become a murderer without "warning signs."
Well, if your probabilistic model has no "warning signs", then how does it provide any information at all? If you don't have a method of using information to differentiate the probabilities when given a person/group of people/location etc. then you have no predictive power at all, except for the average murder per capita.
> building education toward methods of "de-escalation" for psychotic episodes, crimes of passion, etc.
De-escalation of psychotic episodes is an impossibly hard thing to teach without protracted work with a mental health professional. In addition, I seriously doubt that there would be any effectiveness when taught to people who have not experienced psychosis. Teaching this to everybody would be inhibited not only by cost, but by the fact that there is not likely enough people in a society that would be good enough therapists to do this on a large scale.