This article argues that, if Yogācāra is read rigorously as a theory of how experience is constituted, how error becomes concretized, and how consciousness is transformed through perfuming, then the relation between modern artificial intelligence (AI) and Yogācāra is not merely metaphorical. Rather, selected mid-level Yogācāra claims exhibit analyzable, limited structural correspondences with contemporary AI. The article does not presuppose that Yogācāra must be non-idealist, nor does it use AI to establish Yogācāra metaphysics; it brackets that debate and compares functional roles, dynamics of updating, and system-level locations. Centered on the Saṃdhinirmocana- sūtra, Mahāyānasaṃgraha, Mahāyānasaṃgrahabhāṣya, Viṃśatikā, Triṃśikā, and Cheng weishi lun, and read together with contemporary Yogācāra scholarship, the article shows that Yogācāra is concerned with representational mediation, subliminal dispositions, self-grasping, and dependent arising. It further argues that selected structures in contemporary AI, including representation learning, foundation models, world models, retrieval-augmented generation, continual learning, agent architectures, and alignment research, can be understood as functionally parallel to selected mid-level Yogācāra analyses, without implying historical dependence or doctrinal identity. On that basis, the article proposes six Yogācāra-inspired implications for future AI: embodied multimodal world models, memory systems integrating parametric and episodic traces, explicit but bounded self-models, dream-like offline simulation, socially co-conditioned multi-agent cognition, and a deep- alignment orientation that is formally comparable, though not identical, to āśraya-parāvṛtti.

Keywords:
Yogācāra, ālayavijñāna, artificial intelligence, world models, self-model