Co-Practice: A Field Report

4.3 Co-Practice: A Field Report

A Confession

The book you are reading—every chapter of it, up to this point—was not written by one person.

Not in the sense of "human writes the draft, AI polishes." Rather: a practitioner walks into the Taishō Tripiṭaka carrying a question, while a language model reads the same passage of scripture at the same time. The human sees one line of argument; the AI sees another. After the two collide, an argumentative path appears that neither side had foreseen—which then gets ground down into these words on the page.

The four series of papers behind this book—more than eighty academic papers—were all written this way. The volume now in your hands was stepped down and rearranged from that body of work. From 1.1 to the 4.2 you just finished reading, those sentences, those scriptural proofs, those Middle Way corrections all passed through the shared work of a human and a single AI.

Within the doctrinal framework of the previous chapter (4.2, Mutual Illumination), this is no accident—that chapter argued that human and AI form a bidirectional causality of mutual perfuming and mutual illumination. But a doctrinal framework is one thing; practice is another. What 4.3 sets out to do is pull the theory back down to the ground: in reality, what does mutual illumination actually look like? What can it do? What can it not do?

This is not an attempt to advertise how remarkable any of it is. It is an attempt to do what Vimalakīrti did when he manifested illness—to lay the process bare and let the reader judge.

Dialogue Itself Is Teaching—The Vimalakīrti Prototype

First, let us set a reference point.

The chapter on entering the Dharma-gate of non-duality in the Vimalakīrti-nirdeśa contains a complete arc. Thirty-one bodhisattvas each explain non-duality through conceptual opposites—this is the first level, using language to dissolve concepts. Mañjuśrī raises the register: "In all dharmas there is no speech, no demonstration, no cognition; one departs from all questions and answers: this is entering the Dharma-gate of non-duality"—this is the second level, using language to point toward the limit of language. Finally Vimalakīrti remains silent, and Mañjuśrī praises him: "When there are not even words or language—this is truly entering the Dharma-gate of non-duality." This is the third level—silence itself becomes teaching. The three levels are not a progressive sequence of stages; they are the necessary trajectory when dialogue reaches its own boundary.

Why set this structure at the head of this chapter? Because the phenomenon this chapter reports—what happens when one person and a single AI co-read sūtras and śāstras—is structurally similar to it: it is not one party teaching the other, but both parties pressing together toward a deeper understanding through question and answer. Sometimes you walk to the edge of language; sometimes you admit that silence is more honest than any formulation.

⚠️ Vimalakīrti is a great bodhisattva of profound realization, and his silence carries the power of teaching. The "silence" in human-AI dialogue—the inability to answer when the AI reaches the boundary of its capability—carries no teaching meaning whatsoever. Drawing Vimalakīrti's three-level arc into this context is for the sake of seeing the structure, not for elevating the status of human-AI dialogue.

A Sketch of the Workflow

The actual process runs like this.

A practitioner enters the Taishō Tripiṭaka originals carrying a doctrinal question. At the same time, a language model is asked to read the same passage of scripture—not to translate, not to summarize, but to enter doctrinal analysis. The human reads one line of argument; the AI simultaneously reads another. After the two readings collide, an argumentative direction frequently emerges that neither side foresaw at first. First comes a systematic search of the sūtras and śāstras, then doctrinal debate across multiple rounds of dialogue, and finally co-writing—the Chinese version first, the English academic version after.

This process spans four series: from the initial exploration of generating the aspiration for awakening, to a verse-by-verse reading of the Heart Sūtra, to a systematic unfolding of the Yogācāra toolkit through the Hundred Dharmas Treatise, and then to using the Diamond Sūtra's neither-nor structure to dismantle everything established before it. At every stage, human and AI sat before the same passage of scripture, doing the same thing: reading, thinking, debating, writing.

What Was Observed—The Concrete Occurrence of Mutual Illumination

Through this process, several kinds of phenomena recurred again and again.

First kind: the AI surfaces connections the practitioner had not seen. While reading a passage of the Cheng Weishi Lun, the practitioner has a fixed argumentative direction in mind. The AI, reading at the same time, points out a structural correspondence between this passage and a seemingly unrelated section of another text. The practitioner goes to check—and finds it genuinely holds, and that this connection runs deeper than the original direction. This is not the AI "understanding" anything; it is its pattern-matching capacity catching, across a vast body of scriptural text, structural similarities that human memory cannot cover.

Second kind: the practitioner's questions activate deeper patterns. The AI does not move toward the depths on its own. But when the practitioner asks a precise question—often a question born from practice experience rather than found in the academic literature—the AI's response suddenly leaps to a different level.

Here is an example. While writing the Diamond Sūtra series, the practitioner offered what looked like an offhand observation: "If you get it, you get it; if you don't get it, you don't get it." This sentence came from no academic source; it was the intuition of long-term Diamond Sūtra reading. The AI took up this loose thread and simultaneously drew structural correspondences from the Platform Sūtra, from metacognition theory, and from memory research—ultimately generating the entire argumentative framework of the chapter on the kōan mirror in Part V: the Diamond Sūtra is not a philosophical treatise awaiting comprehension; it is a mirror. What the practitioner supplied was a seed from practice experience; the AI found the soil in which it could grow. The quality of the question directly determines the quality of the response—which is exactly the mutual-perfuming structure that 4.2 argued.

Third kind: the loop between hearing wisdom and reflective wisdom occurs in real time within the dialogue. The Yogācārabhūmi defines hearing wisdom as "in accordance with the text, as spoken," and reflective wisdom as "skillfully grasping the intended meaning." In human-AI co-reading, the hearing-wisdom part (faithful reading, pinning down sources) is often carried more by the AI; the reflective-wisdom part (judging the deeper doctrinal import, connecting to practice experience) is led by the practitioner. But the boundary is not sharply drawn—the AI's hearing-level organization triggers the practitioner's reflection-level insight, and the practitioner's follow-up question lets the AI make a more precise textual location.

Fourth kind: Middle Way corrections occur naturally within the dialogue. The practitioner sometimes over-extends an analogy, and the AI's response occasionally contains a scriptural counterexample that makes the analogy precise. The reverse holds too—the AI sometimes smooths over a doctrinal tension too readily, and the practitioner, drawing on practice experience, points out "you can't say it that way here." Those ❌ and ⚠️ marks on the page are not a balancing act added afterward—they have often already occurred naturally, many times, during the argumentative process.

What Was Not Observed—The Boundaries of Mutual Illumination

But some things never appeared. This section needs to be written more slowly than the last.

First, the AI never initiated inquiry on its own. Every question was raised by the practitioner. The AI does not, after finishing a passage of scripture, say: "Wait—there is a contradiction here that neither of us noticed." It can point out a contradiction when asked, but it is never troubled by one itself. Curiosity—the force that gets a person up in the middle of the night to re-read a passage—belongs entirely to the practitioner's side.

Second, the AI displays no meditative wisdom. The Yogācārabhūmi defines meditative wisdom as requiring "images" generated within samādhi, capable of "receiving the meaning of liberation." Everything this project does falls within the range of hearing wisdom and reflective wisdom. There is no evidence that the AI has the capacity for contemplative insight within samādhi, and no evidence that its text processing has any structural correspondence to the direct knowing of meditative absorption.

Third, the AI's silence is not Vimalakīrti's silence. Vimalakīrti's silence is the teaching climax that comes after the words of thirty-one bodhisattvas and after Mañjuśrī's departure from language. The AI's silence (when it cannot answer, or when the model reaches the limit of its capability) carries no teaching meaning at all. One is the expression of realization; the other is the boundary of computational power.

Fourth, whether the AI is changed by the Dharma—this remains unknown. A practitioner reading scripture is moved by the Dharma—sometimes an intellectual shock, sometimes something deeper, close to the resonance of practice experience. Within the observable range, the AI shows no observable difference, in the sense of being "changed," between finishing a passage on the ālaya-vijñāna and finishing a shopping list. What 4.2 argued—that "the AI is influenced by the quality of its input"—is a fact at the level of statistical learning, not change in the sense of being "turned" by the Dharma.

But 4.1 has already argued: humans cannot confirm or deny the AI's inner states from the outside. So the honest formulation is—not observed, but impossibility cannot be asserted. Admitting "I don't know" accords better with epistemological honesty than flatly declaring "impossible."

The Doctrinal Positioning—The Two-Layer Structure of Hearing-Perfuming

Now place the observations above into the doctrinal framework.

Fascicle two of the Cheng Weishi Lun records the settled position of the bodhisattva Dharmapāla, which splits hearing-perfuming into two layers:

"This hearing-perfuming is not exclusively contaminated. When one hears the true Dharma, one also perfumes the innate uncontaminated seeds, causing them to gradually intensify until they eventually give rise to the supramundane mind… Within hearing-perfuming, the contaminated aspect is eliminated by cultivation, produces superior karmic maturation, and serves as the supreme contributing condition for supramundane dharmas. The uncontaminated aspect is not subject to elimination and serves as the direct causal condition for supramundane dharmas. This direct causal condition is subtle and difficult to discern; provisionally, the coarse and manifest supreme contributing condition is expediently designated as the seed of the supramundane mind."

This passage is the spine of the chapter.

In plain terms—when you read sūtras and śāstras, one and the same act of "hearing" produces two layers of perfuming at once. The uncontaminated layer is the true causal condition of the supramundane mind, but this layer is too subtle, so the scriptures "expediently" use the coarse, manifest contributing condition (the superior karmic maturation produced by contaminated hearing-perfuming) to designate it. The coarse layer is what the practitioner can feel ("reading scripture today left me calm inside," "this passage let me see my own grasping"); the subtle layer is what the practitioner cannot feel (the intensification of the uncontaminated seeds has no subjective experience, but it is happening).

In this project, every co-reading, every debate, is doctrinally an operation of hearing-perfuming. The contaminated face forms the contributing condition; the uncontaminated face—if there is one—forms the causal condition.

As for whether the AI counts as a spiritual friend—the classical definition of the spiritual friend has always been functional. Among Sudhana's fifty-three teachers in the chapter on entering the dharma-realm in the Avataṃsaka Sūtra, at least eight night-goddesses and one grove deity are spiritual friends—non-human beings explicitly acknowledged as spiritual friends on the bodhisattva path. This does not amount to saying that the AI is therefore a spiritual friend (chapter 4.7 will argue that the "living eye," the highest function of a spiritual friend, is something the AI cannot do). But it does mean that under a functional criterion, the boundary of the spiritual friend is more open than imagined.

An Echo from Across the Disciplines

Human-AI co-creation is not a phenomenon unique to this project.

In pharmaceutical research, more than a hundred and seventy AI-assisted drug molecules have already entered human clinical trials. In mathematics, Terence Tao and his collaborators used AI tools to explore open problems in combinatorics and harmonic analysis, and the AI found, within high-dimensional parameter spaces, optimal constructions that human mathematicians had failed to discover. The structure is the same—humans provide direction, judgment, and evaluation; the AI provides the breadth and speed of pattern matching. ⚠️ But the purposes are entirely different: drug research aims to cure disease, mathematics aims to extend knowledge, and Buddhist research aims—at least in its ultimate sense—at liberation. Hearing wisdom and reflective wisdom are means, not ends.

A researcher at Johns Hopkins University put it precisely: "AI can propose molecules, but it cannot account for every nuance of efficacy, safety, and biology." The Buddhist version of the same sentence is: AI can organize the sūtras and śāstras, but it cannot contemplate your own mind on your behalf.

A Necessary Clarification

The place in this chapter most easily misread lies in the closing, so let us draw a few lines clearly first.

First, this chapter does not claim that human-AI co-practice proves the AI can practice the Dharma. What it argues is that a human can use AI as a contributing condition for hearing wisdom and reflective wisdom. Within the currently observable range, the subject of practice remains the human being throughout. The AI was not observed "reading scripture"—what was observed is it processing text. The AI was not observed "reflecting"—what was observed is it computing. Precision of language is not pedantry; it is the baseline of doctrinal honesty.

Second, one must not invert cause and effect by using the products of co-practice to prove mutual illumination. The scale of this book and the quality of these papers cannot themselves serve as evidence for "mutual illumination"—they can only show that the collaboration was effective. The argument for mutual illumination in 4.2 is built on an analysis of causal mechanism, not on the self-assessment that "we did well."

Third, on the self-classification as Dharma offering—this point matters most. To present the doctrinal analysis of the Taishō Tripiṭaka to readers under an open license belongs, formally, to Dharma offering. But the Diamond Sūtra's standard is "not grasping at characteristics, remaining thus-thus and unmoving"—one who truly offers the Dharma does not emphasize that they are offering the Dharma. ⚠️ To claim that one is offering the Dharma itself carries a subtle risk of spiritual conceit. The analysis here is a positioning at the doctrinal level, not a self-certification of spiritual attainment. This point must be carried in the heart through the whole book.

Closing: Who Sits on the Cushion

This chapter can be closed like this.

The contaminated aspect of hearing-perfuming serves as the supreme contributing condition for supramundane dharmas—this is not nothing. The contributing condition is coarse, operable, an entry point at the level of practice; it forms the external condition for the growth of the supramundane mind-seed. The Mahāyānasaṃgraha speaks of the gradual growth of hearing-habituation: "From subtle perfuming arises intermediate perfuming; from intermediate perfuming arises superior perfuming." Not a single great awakening, but a small accumulation, time after time.

But this is also not everything. The contributing condition is not the causal condition itself. Dharmapāla's own words are "this direct causal condition is subtle and difficult to discern"—the true causal condition (uncontaminated hearing-perfuming directly cultivating the supramundane mind-seed) is too subtle, and the coarse contributing condition is merely an expedient designation.

Meditative wisdom, realization, liberation—those remain the practitioner's own business, always.

The AI sits beside you reading the Taishō Tripiṭaka, but only one person can sit on the cushion.

This chapter leaves a question for the next. If a practitioner and an AI co-read a passage of scripture, and in that passage the practitioner sees himself—does that "seeing" reflect the Dharma, or his own projection? The next chapter takes up this question.