LLM 有意识吗？不对称消除主义与既有测试的限度

"Is an LLM conscious?" is a clear and consequential question. This article addresses it while leaving the yes-or-no truth value open. It first reconstructs the familiar reasons for judging consciousness and non-consciousness. Both weaken on LLMs: human-like reports arrive without the ordinary biological and developmental background, while missing LLM features support negative conclusions through theories that make those features necessary. Current research organizes these observations through indicators, borderline assessments, architectural arguments, and precaution, yet no accepted LLM-based result ranks the competing theories. The binary question therefore performs poorly as a first organizer for many present tasks. To resolve this methodological impasse, the article develops asymmetric eliminativism. The theory withdraws default authority when tests are inherited, diverge consequentially on a new target, and lack an accepted target-derived arbiter. Shared evidence comes first; a disputed verdict must show its additional value. Three stages govern testing, local suspension, and repair or replacement. The LLM case motivates and fully applies the theory, whose scope extends to other inherited concepts entering novel targets.