The argument is less that an LLM is a human and more that it is not a copyright violation to use a material to train the LLM. By current legal definitions, it is fair use unless the material is able to be reproduced in its entirety (or at least, in some meaningful way).