Because it doesn't exist in a vacuum, it exists inside an agent, that is inside the world. And the agent has to achieve a goal, to maximize utility, by learning to act intelligently. This creates a complex value system for states and actions. The values are used to select our actions, and we experience them as "qualia".
So colors and pleasures are just sensory experiences accompanied by their action-values. In the end it's all a game, and qualia is generated by the game. We're mistakenly searching for qualia in the brain, while it was all around us, in the game itself. The brain is but one part of it.
Why wouldn't it? Not everyone shares your underinformed intuitions. Predictions about light are colors, predictions about reward are pleasure, etc.
Could you point to something related to predictive coding and control networks explaining anything phenomenological (or give an argument)? From what I can tell the two are completely orthogonal.
Edit: I should add that my suspicion is you are mistaking a system which is isomorphic to relationships between phenomenological entities with something which could account for the presence of the entities themselves.
I would recommend checking out the comment here from user 'theoh' on the "relations are what matters" perspective.
Predictive coding in the interoceptive and control networks of the brain, actually. Sure, it doesn't explain everything, but it pretty well dispels the intuition that makes p-zombies conceivable and explains the vast majority of the empirical and phenomenological facts to be explained.