Natural abstractions can be translated between different degrees of complexity
This is a response to Sophisticated behavioural functions without consciousness.
If we consider a superintelligent AI system which forms no abstractions at all, because it has no computational limit that forces it to, I believe that we should still be able to understand the information the AI system works with.
Alternatively, consider explaining a complicated technical topic to a 6-year-old. You have your own set of abstractions, which help you in understanding the topic. It’s plausible that you will be able to adapt your abstractions in order to make it understandable for the 6-year-old.
Similarly, the described AI system might still be able to describe it’s knowledge to you, using abstractions that you are likely to understand.