Well, if we are to model taste changes we would have to, actually, model time - something that we have not, yet, done in this thread. The formal extension is straightforward enough: each good should be indexed by time, so that consumption bundles become consumption streams - and the preference is defined over the space of such streams, not single-period bundles. In this way we can compare streams involving various amount of spinach along your lifetime. If you, at birth, know how your preferences will change in the future, that would have been enough - but, of course, that is unrealistic. So we need to introduce uncertainty into the model as well. That is a little harder - I will get to it, eventually. But, once certain technical difficulties are dealt with, we may be talking of "preference shocks" and individual response to expectation of those.
Taking expectations, whatever that means is, in fact, conceptually a rather complicated thing for an economist. I will, hopefully, get to it at some point. Expected value of a stream of income, though, is fairly straightforward - as long as you understand under which exact conditions the concept makes sense - I will get to that as well, if I continue long enough
