The biggest problem with the paper is that the "conscious agents" are never really identified. Hoffman says that, at this point, the idea of a "conscious agent" is entirely abstract with respect to possible objective, measurable entities. He likens it to a Turing Machine. So, the case may be that conscious agents are not conscious in the human experiential sense, and that many conscious agents could possibly constitute a single person, only some of which correspond to commonplace awareness. In other words, classical free will could still be an illusion, and the universe probably doesn't care about you.
The theory is in its preliminary stages, to be sure, but Hoffman draws his confidence from the rigid mathematical formulations spelled out in Observer Mechanics: A Formal Theory of Perception (the technical edition of Visual Intelligence), which, he claims, can be used to derive many of the equations of quantum mechanics.
With regards to the interface theory being a rehash of Kant's noumenon/phenomenon distinction, I'm not sure if that's supposed to be a dismissal or what. There's a lot to be gained by taking philosophically sound arguments and putting them into modern, technical terms. The entire argument found in Observer Mechanics and Visual Intelligence rests on the basic language of Kant's distinction, but the evidence and fine points don't necessarily fall out of the Kant's original work.
Now, I'm not saying that I agree with every one of Hoffman's points and accept conscious realism; I just think it deserves a fair shake. I've tried debating with him several times and it's not easy (I always walk away thinking that I should have majored in math), but the experience is worth it.