There's an important difference you are missing here, though, which is that even if you don't know what specific type context a function is being used in, you still have some very precise information about what the function may do. For example, you may not know what 'mzero' is going to do in this type context, but you do know that it has some specific properties (e.g. mzero >>= f === mzero) and can extrapolate from there.

You'll have similar problems in dynamic dispatch systems: what does this object do? It could do so many things. In Haskell, we demand that the object always follow a certain set of rules. This is very useful.