Friston's characterizations of qualia being knowledge about internal states is how I treat them too. But it still doesn't explain why the knowledge "feels like" something instead of the "information access" being realized by possessing simply blind dispositions to act and talk in a certain manner (including use of "qualia" language serving as an information bottleneck to simplify complex internal dynamics and allow simpler communication within intra-mind components and among inter-minds).

Ultimately I don't think there is anything mysterious. The formal functional characteristics that we find in mathematical forms are ultimately realized by physical concrete things with physical properties (whatever "physicality" may be -- which may turn out to be idealistic in essence). One of the properties of certain physical configurations may be qualitativity. Qualitativity need not be "essential" for realization of cognitve forms (or access to internal state information), but simply one "way" the realization happens in biological (and possibly non-biological) entities. If similar forms of realization can happen in sillicon is then upto a theory of consciousness that account for which physical configurations and what exactly leads to qualitative dispositions as opposed to non-feel dispositions.

I also don't think Bishop's answers to need of phenomenality answers anything. Why should uncertainty reduction require phenomenality per se? You can just well have a higher-order decision mechanism that experiments, makes discrete actions in the world, intervenes and so on to reduce search space without any appeal to phenomenality. I think it's possible that how our "phenomenal consciousness" works contingently at a certain level, but it doesn't mean there could not have been alternatives without phenomenality to realize some uncertainty-reduction functionalism (especially if we can make a mathematical model of this process).