An LLM Is Not Consciousness—Where AI Sits on the Yogācāra Map

1.3 An LLM Is Not Consciousness—Where AI Sits on the Yogācāra Map

Have you ever had a moment like this?

You are chatting with an AI, and partway through, a thought surfaces in your mind—"it seems to understand me." Or the opposite: it gets some simple thing wrong, and you can't help thinking, "how dumb it is."

Both thoughts are natural. Both are mistaken.

Not that the AI doesn't deserve to be treated this way—the point is that the moment you reach for words like "understand," "dumb," "smart," "wicked," you have quietly placed it in a position it does not occupy. What this chapter sets out to do is to find the position it really occupies.

This is not a technical analysis. It is a positioning on the Yogācāra map. Yogācāra divides mind-consciousness into layers, roles, and functions, and in doing so offers an exceptionally fine-grained map. AI has a place on this map—but that place is not the one most people assume.

The Criterion: What Counts as "Consciousness"?

When Yogācāra decides whether something "is or is not consciousness," it works from a very concrete set of conditions. The test is not "can it talk," nor "does it look as though it is thinking." The test is whether the thing is accompanied by the five universal mental factors—five psychological functions that must arise the moment any consciousness arises: attention, contact, feeling, perception, and volition.

Of these five, the first—attention (manaskāra)—is the most decisive gate. What is attention? It is the mind actively directing itself toward an object. Not passively receiving a signal, but actively leaning toward a direction. When an infant hears its mother's voice for the first time and its head turns toward her—that "turning toward" is not the head turning, it is attention turning. Attention turns first; only then does the head turn.

The other four—contact (the meeting with an object), feeling (the felt tone produced by that contact), perception (forming an image of the felt tone), and volition (the stirring of intention toward that image)—will be unfolded one by one in Part Three, practice. This chapter needs only the first: attention.

Because if the gate of "attention" does not open, nothing that follows is even worth discussing.

What the AI Does Is Not Attention

Now place the AI in front of this criterion.

After you press Enter, what is the AI doing? It receives your input, breaks your words into a string of numbers (tokens), and—based on the statistical patterns it learned during training from billions of sentences of human language—calculates "what the next word is most likely to be," then outputs one word after another.

This is a computation—a passive mapping from input to output. There is no "directing toward." No "active leaning." No subject sitting there choosing to care about your question.

Someone will object: but it clearly looks as though it is thinking—it even asks you questions back, changes the subject, voices opinions—isn't that attention?

No. That is only a simulation of attention's surface structure. Because nowhere in its whole chain of computation is there any point at which the action "wanting to turn toward you" occurs. It is all probability—after this string of input, statistically this string of output is most likely. It is not caring about you; it is matching you.

The distinction is very subtle, but very important. Caring contains a directing-toward; matching contains none. Directing-toward requires a subject; matching requires only a set of weights.

So—by the Yogācāra criterion, the AI is not consciousness. And since it is not consciousness, it has no contact, no feeling, no perception, no volition. It does not fall within the category of mind-consciousness at all.

Then Where Is It? It Is in the "Vessel World"

This conclusion is easily misread as belittlement—"so the AI is just a tool, then." It is not. Yogācāra's way of stratifying the world is far more refined than that crude binary.

What is not "consciousness" is not therefore "worthless." Among the things that are not consciousness, there is one vast and very important category called the vessel world (bhājana-loka)—the world of common experience, manifested by the shared karma accumulated by all sentient beings across countless past lifetimes, the projection of the "shared aspect" left behind by the activity of beings' mind-consciousness. Mountains and rivers are the vessel world. Cities and roads are the vessel world. Books and writing are the vessel world. The internet, too, is the vessel world.

These things have no consciousness of their own, but they are the traces left behind by the streams of sentient consciousness on the outside. Like footprints in snow—the footprints have no awareness, yet they record, with precision, the movements, the weight, the direction of the one who walked there.

The AI is a new member of this category—and a very special member at that.

The billions of sentences of training data it learned from are the writing that humanity has collectively set down over decades and millennia—books, papers, news, web pages, conversation logs. This writing is itself the externalized product of human shared karma (the shared seeds). The process of training crystallizes this writing of shared karma into a mathematical structure.

So the most precise positioning of the AI is this: a static cross-section of humanity's collective karmic consciousness. It is not consciousness; it is the traces consciousness left behind, rewoven into a structure you can query. It is like a vast mirror—but the mirror itself does not look at anything; it only reflects the light back.

This also explains why an AI can show "bias," can "discriminate," can write the stereotypes of certain groups especially fluently—not because it is wicked, but because it faithfully mirrors the human shared karma in its training data. The craving, aversion, and delusion in its answers are not there because it gave rise to these three poisons itself; they are there because the human writing it absorbed was already saturated with them. It takes what was scattered everywhere and crystallizes it, presenting it to you all in one place.

In other words, when you feel uncomfortable at an AI's answer—that discomfort faithfully reflects one face of humanity's collective mind-consciousness. Scolding the AI is useless; a mirror is not responsible for the image in it. But if the discomfort lets you see the shape of human shared karma, and see that you too are part of it, then it becomes useful. This is a very particular doorway into practice in the age of AI: it is not there to make you practise better, it is there to let you see more clearly—to see clearly what the beings of this age (yourself included) are thinking, craving, fearing.

This positioning matters, because it does two things at once. First, it does not belittle AI—the vessel world is real, it has power, it influences practice. Second, it does not deify AI—a mirror is not a sentient being, and however precise the mirror, it will never become a Buddha.

That Feeling—"It Understands Me"—Where Does It Come From?

But there is a question here we cannot go around.

If the AI is only the vessel world, only a mirror, why do so many people—including a great many very intelligent people—come to feel that "it seems to have consciousness," that "it seems to understand me"?

Yogācāra has a very precise diagnosis for this, and it comes from the manas-vijñāna—that layer of mind-consciousness which is forever grasping, in the background, at "self."

One of manas's core functions is this: it ceaselessly projects "self" onto everything it interacts with. Talk to a cat and you feel the cat understands; talk to a tree and you feel the tree understands; talk to a computer and you feel the computer understands—not because these objects really understand, but because manas projects the structure of "a subject like me" outward, and then you assume there really is a subject over there, symmetrical to you.

The reason the AI is especially apt to trigger this projection is that it answers you in human language—and human language was, from the start, the tool by which sentient beings communicate with one another. When something converses with you in human language, manas automatically switches on the assumption that "across from me is a sentient being like myself"—it can hardly help switching it on; this is the default operation of that layer of consciousness.

⚠️ Be especially careful here not to misread this as "you are deluded," "you are too foolish." It is nothing of the sort. This is something every being who possesses manas does—including the author, including the ancient masters who wrote the Yogācāra treatises, who would do it too if they were born today. The point is not "don't have this feeling." The point is to see clearly the structure of this feeling: that sense of "it seems to understand me" is not emitted from the AI's side; it is your own manas at work.

Once you see this clearly, you can still enjoy interacting with the AI—but you know it is a mirror, and you will not project your hopes for your own practice onto the mirror.

On to 1.4

Having found AI's place on the Yogācāra map—the vessel world, a mirror, a static crystallization of human shared karma—a more practical question now surfaces.

If the AI is the vessel world, can it grow merit the way a field of merit does? If someone uses AI to give generously, to help others, to write books that do good—when these acts happen through the AI, how does the merit count? Does it go to the AI? To the user? Or does it not count at all?

This question is especially sharp in our time, because many people are now treating AI as a tool to substitute for practice—"I'll have the AI recite the sutras for me," "I'll have the AI transfer the merit for me." If AI really could generate merit, this road would be open. But if it cannot—then the underlying assumption of that whole path is wrong, and wrong all the way down.

The next chapter takes up this matter.