In many cases, when one is looking for a solution to a problem and they find a question, the problem presented in the question is not necessarily the exact same problem they're trying to solve. Yet, it's close enough that the answers still apply.
Then, there are other cases where one stumbles upon a question which indicates (based on the wording, etc.) that they're in the exact same situation that the OP was in when they wrote the question (obligatory xkcd). In a subset of those cases, what the asker is trying to do might not be recommended and/or they might receive (good) answers addressing the bigger picture or the more general problem. For example, someone might ask:
How do I Foo the Bar?
And they get (again, good) answers like this:
You shouldn't really Foo the Bar; that's bad practice. Consider bla bla bla instead. [insert more explanation here]
In some situations, the OP's problem is so specific that they do really want to Foo the Bar and they don't care about anything else. So, when they receive an answer that tells them how to do that, they accept it, but it doesn't get as many upvotes as the other answer(s) (for good reasons).
Another example (specific to Stack Overflow and technical sites) is when the accepted answer only applies to an ancient version of a piece of software.
In both of the cases above, if someone finds themselves in the very same situation as the OP, they would really want to know which answer worked for them (which can mostly be inferred from the fact that the answer is accepted).
That's why unpinning the accepted answer --while leaving everything else as is-- renders the "accept" feature almost completely useless. Don't get me wrong, I'm not against unpinning it; on the contrary, I believe that the old (still current on most sites) system where the outdated answer or the answer that is too specific / suggesting something against good practice is listed as the top one is not a good system. However, there needs to be a way for users to somehow "jump" to the accepted answer if they choose to do so.