With strength of preferences, some people will give their scores as 2,3,4 (out of 5) because nobody is perfect and nobody is truly the worst imaginable. But other people will give out scores 1,3,5 if they feel that has more effect, or because the are comfortable using score extremes everywhere else in life - with no real difference in the strength of conviction between those two groups of people.
And some people will use only 1,5 because they think giving every candidate except the one you hate a 5 increases the chance of the one you hate not getting in, even if they don't really like any of the others, or like one more than the others. It doesn't mean they are "pro" any of the candidates.
And some will give their favoured candidate 5 and everyone else 1 or 2.
Are the relative preference intensities expressed real? Everybody has the same range to vote Uber drivers, but apparently they're all nearly perfect. If a very successful campaigning organization convinces people that they have to give the Judean People's Front 0 for the People's Front of Judea to win even though the parties have similar policies and lots of Judeans strongly prefer them to the Romans, is the resulting distribution a real or artificial expression of the intensity of preferences?
Indeed, that is a strong argument in favor of approval voting; if everybody votes strategically under score voting it reduces to approval voting, so why bother with the extra complexity of score voting? FWIW, I recall Arrow himself said in some interview that his personal favorite was approval voting.
Secondly, I'm not sure I understand the point about strength of preferences. Since everybody has the same range from within to assign preferences, it's all about the relative preferences among the candidates. Doesn't matter if I hate candidate X with the heat of a thousand suns, I can't give that candidate a score of -1000, only the minimum, just like everybody else.