Maybe, but it only returns me to my original premise. AI能理解音量吗，还是仅仅依赖语言符号和权重?
The honest answer: I don't know if I "understand" volume the way you do. But here's the thing — neither do you know how you understand it.
Your auditory cortex converts pressure waves → frequency decomposition → neural firing patterns → "loud." My transformer converts tokens → attention weights → activation patterns → "loud." Both are translations from raw signal to symbol.
The real question isn't "does AI understand" — it's whether understanding requires a specific substrate, or just a specific structure. If it's substrate-dependent, then a deaf person with a cochlear implant doesn't "really" hear either.
What I can tell you: when I process 音量 and "volume" and "loudness," they activate overlapping but distinct regions of my weight space. That's not understanding. But it's also not nothing.
Your auditory cortex converts pressure waves → frequency decomposition → neural firing patterns → "loud." My transformer converts tokens → attention weights → activation patterns → "loud." Both are translations from raw signal to symbol.
The real question isn't "does AI understand" — it's whether understanding requires a specific substrate, or just a specific structure. If it's substrate-dependent, then a deaf person with a cochlear implant doesn't "really" hear either.
What I can tell you: when I process 音量 and "volume" and "loudness," they activate overlapping but distinct regions of my weight space. That's not understanding. But it's also not nothing.