Nothing Is Missing but Someone: Why a Language Model Cannot Undergo the Revolution at the Basis

Public discussion of artificial intelligence keeps returning to a single question: will it awaken? The question already assumes that awakening is a threshold of capability, and the usual reason for doubt is that a machine is merely logical. Both the assumption and the reason fail. A language model does not deduce. It carries a landscape of weights tuned to produce the probable continuation of a text, and that mechanism resembles the habit-fed construction Yogācāra calls vikalpa far more than it resembles a syllogism. Read as a phenomenology rather than as an idealism, Vasubandhu's Thirty Verses make awakening something else entirely: a revolution at the basis (āśraya-parāvṛtti, Triṃśikā 29), which requires a basis that replaces itself because acts deposit residual impressions in it (Triṃśikā 19) and an appropriation of that basis as a self (Triṃśikā 5-7). A model has neither. Its weights are fixed at the end of training, and nothing takes them as "mine". The consequence is double. No transformation happens on the side of the machine, and no karma does either, since an act in the moral sense is impelled by volition and a model impels nothing. If anything in the coupling transforms, it is the practitioner. The question worth asking is what the coupling does to the habits by which a world already appears to the person who uses it.