Okay. I've played a few more rounds, glanced at the "How 20Q Learns" section, and recalled various things that I've read or heard elsewhere (e.g.) about AI, and I've concluded that most of my complaints about the gameplay of 20Q are in fact examples of differences in how computers and humans (or this human, at least) "think" and in what their goals are.

20Q doesn't want or need justified true beliefs to as high a degree of exactitude, subtlety, and certainty as reasonably possible. It "wants" to build an organic dataset that will enable it to "guess" correctly within 20 questions for common objects and 30 questions for obscure ones most of the time, in spite of contradictions in the input pool. I, who have little faith in the wisdom of hoi polloi (lucky I live in a democracy, eh? Long live the philosopher king!), am disinclined to value such input. So I interpret any mechanism to maximize the functionality of an imperfect and contradictory input stream as a design flaw.

What does it say about me that I take it personally when a computer game asks me "stupid" questions or reveals that other anonymous players have taught it wrong information?

Don't answer that. I don't want to know.

(I still think Amazon oughta fix their author ambiguity issue, though.)