I take issue with the viewpoint that behavior is best understood by treating all involved minds as if they were perfect Bayesian reasoners motivated only by increasing the frequency of their alleles in the gene pool. Evolution doesn't work like that. Evolved minds are just a pile of kludges that were good enough.
Instead of the way you've been thinking about this, suppose instead that empathy is a straightforward consequence of mirror neurons, which essentially try to model the mental state of others based on their behavior. That might be most useful when dealing with the others in the ancestral social group, but it might also be useful to figure out that that bear over there is probably just protecting her cub, while the tiger last week was probably just hungry, for example. Of course, it would have to be possible to override empathy fairly easily, even in the context of one's own social group, but in the absence of any particular reason to do so empathy might in large part guide one's behavior. Stick people in an environment where food comes from the store, nothing wants to kill and eat you and the animals you encounter most are domesticated or mostly acclimated to humans and one can see how empathy might be applied to all kinds of things it wouldn't be under other circumstances.
This is just a half-baked hypothesis by an amateur, but you get the idea. Brains end up just being good enough, not optimal. If anything, it's a wonder that ours aren't even worse than they are at general purpose computation.
Instead of the way you've been thinking about this, suppose instead that empathy is a straightforward consequence of mirror neurons, which essentially try to model the mental state of others based on their behavior. That might be most useful when dealing with the others in the ancestral social group, but it might also be useful to figure out that that bear over there is probably just protecting her cub, while the tiger last week was probably just hungry, for example. Of course, it would have to be possible to override empathy fairly easily, even in the context of one's own social group, but in the absence of any particular reason to do so empathy might in large part guide one's behavior. Stick people in an environment where food comes from the store, nothing wants to kill and eat you and the animals you encounter most are domesticated or mostly acclimated to humans and one can see how empathy might be applied to all kinds of things it wouldn't be under other circumstances.
This is just a half-baked hypothesis by an amateur, but you get the idea. Brains end up just being good enough, not optimal. If anything, it's a wonder that ours aren't even worse than they are at general purpose computation.