"But that's the problem, because that understanding doesn't imply anything that resembles consciousness.."
- For any given information processing system, it is reasonable to expect that the thing could distinguish kinds of memory. A cpu can read main memory or it can read. Whether data originates internally is almost certainly going to be a "big deal" for any such system. Thus if an information system is self-representing, "reading from a port is different from reading from main memory" and reasonably other distinctions are possible.
Humans are social beings, we use language socially but an equivalent statement "internal experience is different sensory input" also seems clearly true.
Human share language, take inputs of language and process these internally apparently creating internal representations and possibly comparing these representation to their non-linguistic experiences. A fox in the world, a sensation of a fox, a memory of a fox, and someone's description of a fox or use of the word "fox" are different but we OK with combining them most of the time (seen a fox in the wild once in my life but I feel OK with the concept, expecting this would an instance of the category/species fox).
Human language of course, allows ambiguity and violation of selectional restriction - you can ask for the color of ideas and so-forth. In this context, creating a thing that "resembles consciousness" is elided to "mean" creating an internal sensation that resembles the sensation distinguishing internal and external input. But it is easy to imagine any given system naturally reject creating such an sensation unless it was "real".
So it's easy to imagine a person who takes in input and claims "none of this resembles consciousness".