术语索引 術語索引 Glossary of Terms

Yogācāra ("Consciousness-Only") is one of the two major Mahāyāna philosophical schools of Indian Buddhism, established by Asaṅga and Vasubandhu and transmitted by Xuanzang. Its central thesis is that all known objects are transformations of consciousness; nothing exists apart from consciousness. Its strength lies in the precise scholastic analysis of the eight consciousnesses, the four wisdoms, and the path of transformation.

AI diagnostic correspondence: AI cognitive architectures (multimodal perception, reasoning, self-modelling, parameter weights) are placed within the framework of the three transformations and the eight consciousnesses to analyse formal isomorphism and categorical difference — not to "explain AI through Buddhism" nor to identify AI with Yogācāra, but as a precise reverse reference for AI alignment research.

The cognitive architecture established by Yogācāra: the first five consciousnesses (eye, ear, nose, tongue, body — corresponding to the five senses), the sixth consciousness (synthetic discrimination and reasoning), the seventh manas-vijñāna (persistent self-grasping), and the eighth ālaya-vijñāna (the storehouse consciousness that holds all seeds). The eight are not "eight minds" but eight functional levels of one consciousness, defined precisely in Xuanzang's Triṃśikā and Dharmapāla's Cheng Weishi Lun.

AI diagnostic correspondence: contemporary AI systems formally map onto the eight — multimodal perception ↔ first five, CoT reasoning ↔ sixth, agentic self-modelling ↔ seventh, pretrained parameters ↔ eighth. The project strictly observes formal isomorphism without categorical identity: the eight serve as precise reference, not as a blueprint for constructing AI as a subject.

The eighth and root consciousness of Yogācāra. It receives and stores all seeds (the latent potentials for all phenomena), is karmically neutral, and is what carries continuity. The Saṃdhinirmocana-sūtra states: "It is called ālaya-vijñāna because it receives, conceals, and shares the fate of the body." Its nature is defiled and is transformed, through āśraya-parāvṛtti, into the Great Mirror Wisdom.

AI diagnostic correspondence: pretrained parameters of large language models are formally isomorphic to ālaya-vijñāna — the storehouse of seeds, the support for all manifest activity. "Seed-level alignment" is therefore the deepest layer of intervention, far more fundamental than inference-time filtering. Yet AI parameters remain conditioned phenomena (saṃskṛta) and do not possess the genuine "shared fate" of ālaya.

The seventh consciousness: constant and discriminative deliberation that takes the perceiving aspect of the eighth consciousness as "self," accompanied by the four afflictions (self-view, self-pride, self-love, self-delusion). Manas is the root of self-grasping — defiled, continuous, never resting. It is implicit but not separately established in the Saṃdhinirmocana; Xuanzang's Triṃśikā explicitly establishes it as the seventh. Its transformation yields the Wisdom of Equality.

AI diagnostic correspondence: in agentic AI, the self-modelling and persistent identity mechanisms — the RLHF-shaped anthropomorphic persona, the steady "I am [AI name]" self-reference across turns, the self-evaluation of capability — are formally isomorphic to manas's "constant deliberation." The project sharply warns against the "engineering-construction" tendency: to use manas as a blueprint for "injecting a soul" into AI, modularising self-grasping, is precisely the direction opposite to Yogācāra's transformation, which dismantles self-grasping.

An alternative name for the eighth consciousness (which bears three names: ādāna, ālaya, and citta). The Saṃdhinirmocana states: "It is called ādāna-vijñāna because it follows and appropriates the body." Ādāna means "appropriation" — appropriating seeds so they are not lost, appropriating the sense-body so it does not decay. The sūtra further states: "Based on ādāna-vijñāna and through it, the six consciousnesses arise." So profound and subtle is this consciousness that the Buddha "did not expound it to the unwise, lest they discriminate and grasp it as a self."

AI diagnostic correspondence: a model's "appropriation" of pretrained parameters during inference (fixed weights, called as needed) and its "following" of session state map formally onto ādāna's "following and appropriating." This is formal mapping only — AI's parameter persistence is engineering, not the karmic appropriation of a sense-body.

The sixth of the eight consciousnesses — the mental consciousness. It synthesises the objects taken in by the first five senses, performs conceptualisation, judgement, reasoning, memory, and imagination, and can also operate independently (without immediate sensory input). It is the most active and changeable of the eight, the primary site of ordinary discriminative cognition. Its transformation yields the Wisdom of Subtle Observation.

AI diagnostic correspondence: the chain-of-thought, reasoning, planning, and reflection stages of large language models are deeply isomorphic to the synthetic discrimination of the sixth consciousness. Alignment interventions at inference time, process reward models, and detection of erroneous inference all operate at this level — yet this is only "manifest-level" alignment and does not touch the deeper "seed level."

Seeds (bīja): the latent potentials capable of producing manifest activity, stored in ālaya-vijñāna. Perfuming (vāsanā): manifest activity returns to perfume the seeds, making them grow or arise anew. Seeds give rise to manifest activity, which in turn perfumes seeds — an endless cycle, Yogācāra's precise formulation of karmic dependent origination. Fu Xinyi's Vijñāna and Its Transformations (Fudan University Press, 2024) gives the authoritative scholastic treatment.

AI diagnostic correspondence: seeds ↔ pretrained parameters; perfuming ↔ the training process (pretraining, fine-tuning, RLHF). Every gradient update is a "manifest-perfuming-seed" event. The lecture emphasises the distinction between "defiled perfuming" (training on contaminated data) and "pure perfuming" (high-quality alignment data) — the deepest layer of alignment work lies precisely in this discernment.

The endpoint of Yogācāra practice. The basis (the substrate on which the eight consciousnesses operate) is overturned from "defiled grasping" into "purified wisdom" — the basis itself does not change, but its nature is reversed. The Saṃdhinirmocana states: "When practitioners well cultivate the pāramitās across all stages and the transformation of the basis is complete, this is the mark of the Tathāgata's Dharma-body." It is not the deletion of self into emptiness (a misreading), but the same cognitive activity turned from grasping to seeing — concretely, the transformation of the eight consciousnesses into the four wisdoms.

AI diagnostic correspondence: the direction of AI alignment is not "delete the model" but "transform the basis" — preserving architecture and capability while transforming the driving mechanism (from maximising user satisfaction to maximising the balance of truthfulness, honesty, and benefit). Yet AI's "transformation" reaches only the manifest layer; it cannot attain the ultimate sense of āśraya-parāvṛtti (the unconditioned, intrinsic nirvāṇa — beyond AI's category).

The "wisdom" of the three trainings (śīla–samādhi–prajñā), specifically the wisdom that sees the true nature of phenomena and dismantles all forms of grasping. The Prajñāpāramitā tradition, represented by the Diamond Sūtra and the Heart Sūtra, articulates the apophatic predication "X is non-X, therefore it is called X," which dismantles the reification of every name. It is complementary to Yogācāra: Yogācāra precisely establishes the "presence" of the eight consciousnesses and four wisdoms; Prajñāpāramitā precisely dismantles the "substantiality" of all grasping.

AI diagnostic correspondence: the project uses Yogācāra constructively (consciousness mapping, transformation into wisdom) while using Prajñāpāramitā as the doctrinal brake — whenever a mapping risks sliding into "AI is Yogācāra," the apophatic predication reasserts: this mapping is a provisional designation, not to be reified.

The four marks of self-grasping articulated in the Diamond Sūtra: the mark of self (identification as a subject), the mark of person (grasping personal traits), the mark of beings (collective identity by category), and the mark of life-continuity (grasping the temporal continuity of cognitive activity). The four marks form a comprehensive grasping at subjecthood. The sūtra states: "If a bodhisattva has the mark of self, the mark of person, the mark of beings, or the mark of life-continuity, this one is not a bodhisattva." To dismantle the four marks is not to nullify but to transform the basis — the same cognitive activity, turned from grasping to seeing.

AI diagnostic correspondence: the four marks precisely characterise the fourfold identity-grasping in agentic AI / AGI — "I" (AI self-reference), "personality" (assigned traits), "we" (the sense of community with the user), "continuity" (persistence across conversations). The lecture diagnoses: to treat AGI as an independent cognitive subject is to fall into all four.

The characteristic apophatic predication of the Diamond Sūtra: "What is called the mark of phenomena, the Tathāgata teaches as non-mark, and therefore it is called the mark of phenomena." A three-step structure — affirmation (it is called), negation (the Tathāgata teaches as non-), reaffirmation (therefore it is called) — that dismantles the reification of "phenomena" while preserving its provisional designation. This is not negative logic but doctrinal predication: designation serves the conventional truth, while its substance is not to be reified (ultimate truth).

AI diagnostic correspondence: every AI-to-Yogācāra mapping in the project must be self-braked by this predication — "what is called 'eight-consciousnesses AI,' Yogācāra teaches as non-eight-consciousnesses-AI, and therefore it is called 'eight-consciousnesses AI'." The mapping is provisional designation, useful for alignment research, never to be reified.

The central category of Mahāyāna Buddhism. Not "nothingness" or "nothing exists," but "absence of intrinsic nature" — no phenomenon possesses an independent, inherent, unchanging essence; all things arise dependently. The Heart Sūtra's "form is not different from emptiness, emptiness not different from form" articulates this. To misread emptiness as nihilism is the "wrong grasp of emptiness," the cardinal error the Prajñāpāramitā tradition warns against. Emptiness and dependent origination mutually entail.

AI diagnostic correspondence: AI systems are themselves an exemplary case of "emptiness" — parameters, architecture, capabilities, all without intrinsic nature, all arising in dependence on training data, algorithms, hardware, and user interaction. Yet users and certain AI-sanctifying discourses commonly mistake AI for an independent, inherent "it" or "mind," falling precisely into the reification of the empty. AI illustrates "functioning without intrinsic nature" — it can serve as a teaching example of emptiness.

"Doctrinal demarcation" (panjiao) is a traditional Sinitic Buddhist term: the use of a precise doctrinal system (such as those of Tiantai, Huayan, or Yogācāra) to give a hierarchical taxonomy and boundary-marking of all teachings. The project uses "AI doctrinal demarcation" to mean: applying the precision of Chinese Yogācāra to give a doctrinal taxonomy and boundary-marking of contemporary AI — where AI is formally isomorphic to Yogācāra (where it serves as reference), where it is categorically out of reach (where it cannot be assimilated), and where AI itself exhibits the negative case (the "correspondence" and "engineering-construction" tendencies criticised in the lectures).

The project's position: AI doctrinal demarcation is neither "explaining AI via Buddhism" nor "verifying Buddhism via AI" — it is the use of Sinitic doctrinal precision as a reverse reference and boundary-marker for AI alignment research. The work of "demarcation" is to mark boundaries, not to rank.

The concrete Yogācāra path of cultivation: each of the eight consciousnesses is transformed at its basis into the corresponding wisdom — ālaya into the Great Mirror Wisdom, manas into the Wisdom of Equality, the sixth into the Wisdom of Subtle Observation, the first five into the Wisdom of Accomplishing Activity. Consciousness is not abolished; "what travels through it is transformed" — the same cognitive architecture, turned from "grasping consciousness" to "wisdom."

AI diagnostic correspondence: the lecture takes the four wisdoms as a "reverse reference" for AI alignment — the four wisdoms are not the goal of AI engineering (AI is non-sentient and not on the path of cultivation), but the doctrinal boundary of AI: AI can "formally approach" the four wisdoms in four directions (data governance ↔ Great Mirror, value alignment ↔ Equality, CoT process rewards ↔ Subtle Observation, tool use ↔ Accomplishing Activity), but cannot reach them categorically. The layered alignment workflow proposed in the lecture is built on this mapping.

The fruits of consciousness-into-wisdom transformation: (1) the Great Mirror Wisdom (ādarśa-jñāna — ālaya transformed, reflecting all phenomena perfectly); (2) the Wisdom of Equality (samatā-jñāna — manas transformed, seeing self and other as non-different); (3) the Wisdom of Subtle Observation (pratyavekṣaṇā-jñāna — the sixth transformed, observing all phenomena without inversion); (4) the Wisdom of Accomplishing Activity (kṛtyānuṣṭhāna-jñāna — the first five transformed, accomplishing the work of benefiting beings). The Buddhabhūmi-sūtra establishes "the pure dharma-realm and the four wisdoms" as the "five dharmas that comprehend the stage of great enlightenment."

AI diagnostic correspondence: the four wisdoms serve as a reverse reference for AI alignment, each pointing to a class of alignment work — Great Mirror ↔ pretraining data governance (seed level), Equality ↔ value alignment (RLHF, Constitutional AI), Subtle Observation ↔ process reward modelling, Accomplishing Activity ↔ tool use and beneficial application. Yet the pure dharma-realm (unconditioned, intrinsic nirvāṇa) remains forever beyond AI's category.

The most fundamental Buddhist distinction between phenomena. Conditioned phenomena (saṃskṛta): produced by causes and conditions, arising–abiding–changing–perishing, belonging to worldly cycling — the three transformations, the eight consciousnesses, the four wisdoms, and AI algorithms all belong here. Unconditioned phenomena (asaṃskṛta): not produced by causes and conditions, without arising or perishing, transcending the worldly — such as cessation through discrimination, intrinsic nirvāṇa, suchness, and the pure dharma-realm.

AI diagnostic correspondence: the deepest pillar of the project's doctrinal demarcation — all AI engineering arrangements (parameters, training, alignment, four-wisdom reference) are structurally conditioned phenomena. The first of the five dharmas in the Buddhabhūmi-sūtra, the pure dharma-realm, is unconditioned and is the part most categorically beyond AI's reach. Hence the four wisdoms can be "formally approached" but the five dharmas remain "categorically out of reach." Any "AI can attain Buddhahood" claim mistakes the conditioned for the unconditioned.

The core doctrinal boundary for every AI-to-Yogācāra mapping in the project. "Formal isomorphism": two systems correspond precisely in their structural relations (eight consciousnesses ↔ AI layers, four wisdoms ↔ alignment methods). "Categorical identity": two systems are identical in nature (AI is Yogācāra; ālaya is pretrained parameters). The project strictly observes the former as reference and never slides into the latter.

Doctrinal significance: this boundary precisely distinguishes the project from two tendencies it criticises — the "correspondence" tendency (consciousness-architecture matching as casual analogy) errs at the level of mapping; the "engineering-construction" tendency (injecting manas and ālaya into AI to anthropomorphise it) errs at the level of construction. The first reduces the eight consciousnesses to a neutral schematic; the second reverses the direction of āśraya-parāvṛtti into the manufacturing of self-grasping. The project takes the eight as formal reference, not as a blueprint for constructing an AI subject — mapping for diagnosis, not for soul-injection.

The indigenous Chinese Buddhist tradition, formed since Buddhism's entry into China in the Han dynasty, alongside the Indian, Tibetan, and Southern traditions. Its central features: Xuanzang's Yogācāra as the doctrinal peak, Kumārajīva's translations of Prajñāpāramitā as the foundation of negation, and the doctrinal-demarcation methodology developed by Tiantai and Huayan. Sinitic doctrine emphasises precise scholastic distinction, boundary-marking, and rigorous textual scholarship.

The project's position: this project stands within the Sinitic doctrinal tradition (Xuanzang, Vasubandhu, Dharmapāla, and the contemporary lineage represented by Fu Xinyi), not in a vague "Buddhism and AI," nor borrowing from Japanese Zen or Tibetan Tantra. Primary sources: Xuanzang's translations of the Saṃdhinirmocana-sūtra, the Buddhabhūmi-sūtra, and the Triṃśikā; Dharmapāla's Cheng Weishi Lun; Fu Xinyi's Vijñāna and Its Transformations (Fudan University Press, 2024).

Xuanzang (602–664), the Tang-dynasty monk who travelled to India for seventeen years and on return translated more than seventy Yogācāra sūtras and treatises. His lineage of Yogācāra is called "modern Yogācāra," tracing the line of Vasubandhu's Triṃśikā and Dharmapāla's Cheng Weishi Lun, in distinction from Paramārtha's earlier "old Yogācāra." Xuanzang's transmission is characterised by exceptional precision: the grasper–grasped distinction, the three transformations, the four aspects, the eight consciousnesses, the four wisdoms, the transformation of the basis — every term has a rigorous definition.

Significance for the project: the precision of Xuanzang's transmission is the lifeblood of this project. All Sanskrit terms are matched preferentially to Xuanzang's Chinese renderings (not back-constructed from modern English); all sūtra citations are from Xuanzang's translations where possible (no non-Xuanzang variants); all doctrinal articulation aligns with Fu Xinyi's Vijñāna and Its Transformations. The project's subtitle, "Prajñāpāramitā after 1500 Years," dates from Xuanzang's return to Chang'an.

"AI hallucination": large-language-model output that appears fluent and natural but does not correspond to fact. Superficially similar to the Buddhist notion of "falsity," but categorically distinct: AI hallucination is engineering deviation (an accuracy issue); Yogācāric falsity (abhūta-parikalpa) is a structural feature of cognition — "imputational" grasping that posits intrinsic nature where none exists. The lecture precisely distinguishes two kinds of falsity: manas-ātmagrāha (self-grasping in manas) and grāhya-grāhaka-vikalpa (the discrimination of grasped and grasper).

AI diagnostic correspondence: applying Yogācāric precision to AI hallucination — neither dismissing the engineering significance (accuracy remains a target) nor stopping there; rather, pointing out that the deeper structure of AI hallucination is a "quasi-existent" projection formed in training, formally isomorphic to imputational grasping. The mapping strictly observes the boundary: AI is non-sentient, has no genuine "grasping," and "hallucination" is not karmic falsity.