>Right from the start, it assumes that there is a difference between „merely following rules“ and „true intelligence“.

It depends on how flexible those rules are, right? Are the rules a one to one lookup, or are there branching paths with different outcomes?

If the man in the room sees an incoming symbol, looks it up in the book, and sees only one possible output symbol, and sends that out, then he doesn't need to understand Chinese.

If he has more than one option of output, and needs to monitor the results of his output choices, then he's no longer just a symbol translator. He's now an active participant in shaping the incoming information. To get better at choosing symbols, he's going to have to learn Chinese!