1. A Question from Two Millennia Ago↗
An old man is about to leave through a mountain pass, turning away from a troubled land. Yin Xi, the official at the gate, asks him to leave some teaching behind. Two parts and more than five thousand characters follow. This is the story of Laozi’s departure recorded in the Shiji, rather than a scene whose details can be reconstructed as eyewitness history.
The Daodejing begins:
The way that can be spoken is not the constant way. The name that can be named is not the constant name.
That is one reading of its opening lines.
More than two millennia later, in data centres, humans train machines that can speak about the moon, the sun, the stars, even their own existence. They cannot say everything. Yet they can say enough to make an old question newly unsettling:
How far apart are what can be said and the thing being spoken of?
Laozi did not predict language models. But as machines extend what language can do, its limits deserve another look.
2. The Obscured Sky↗
In a story told by later Chan Buddhist texts, the Buddha holds up a flower. The assembly is silent; only Mahākāśyapa breaks into a smile. Case six of the Wumenguan uses this scene to tell of a transmission outside the scriptures, not established through words. It belongs to a lineage narrative, not an independently established record from the Buddha’s lifetime.
A flower, a smile. The story moves attention from what is said to how something is grasped.
Language is a finger pointing at the moon, rather than the moon itself. A verse in the Laṅkāvatāra Sūtra describes a fool looking at the pointing finger instead of the moon, and compares this with attachment to names that obscures what they indicate.
The image offers a way of looking at AI. A language model resembles an extraordinarily dexterous finger. It connects words about moonlight: brightness, waxing and waning, a poet looking up into the night. But a passage about moonlight is not the moment of looking up.
This cannot be reduced to “machines have no eyes.” The GPT-4 technical report already describes image and text inputs; receiving an image does not, in turn, establish a subjective experience of seeing the moon. Input, representation, description and experience cannot be collapsed into one metaphor.
Pointing accurately does not yet explain what seeing is.
Nor is the finger useless. It directs a gaze and corrects its course. Trouble begins when pointing is mistaken for its object, or language about experience is treated as evidence that the experience has occurred.
3. Emptiness↗
Emptiness, or śūnyatā, is an important concept in Buddhist philosophy. In the discussion of dependent arising and emptiness in chapter twenty-four of the Mūlamadhyamakakārikā, emptiness does not mean that nothing exists. It concerns the absence of independent intrinsic nature.
A flower depends on sunlight, soil, water and many other conditions. Seeing, naming and knowing the flower also involves eyes, attention, experience and language. The conditions under which a flower grows need not be conflated with how it appears to a person.
The self can be questioned along similar lines. Memories, feelings, thoughts, bodies and social relations change. Must what is called “I” possess an independent, unchanging core? That is a philosophical question, not a discovery already verified by AI.
Language, too, depends on relations. The word “moon” is not the moon. Its use depends on speakers, listeners, contexts and conventions. This does not make it meaningless: lacking self-sufficient meaning is different from lacking meaning altogether.
Language models make this dependence difficult to overlook. Very different processes can generate the same words, and fluency need not come with reliable factual judgment. Fluent text alone gives little warrant for placing a fully specified mind behind it.
Language seems to carry meaning as a cup carries water. Yet the shape of the cup, the person using it, and the activities of fetching and drinking also matter. The image returns meaning to its relations; it cannot prove that there is “nothing inside” a model.
4. A Mind That Divides↗
Discrimination, conceptualization and attachment recur in Buddhist texts. Here, “the discriminating mind” names a familiar tendency: the mind continually divides, classifies and names the world. The Laṅkāvatāra Sūtra’s discussion of names and discrimination belongs to a context of practice and liberation. It is not simply a modern cognitive theory of categorization.
Before a continuous spectrum, language says: this is red, this is orange, this is yellow. Before a complex feeling, it says: happiness, sadness, anger. Before the world, it says: this is me, this is not me.
Every word resembles a cut.
These cuts make communication and reasoning possible. The danger need not lie in making distinctions, but in forgetting their purposes and conditions, treating a named boundary as an immutable seam in the world. The distinction between an explanation with evidence and one without should not disappear in a discussion of going beyond discrimination.
A language model learns differences and relations from large amounts of text. It has not learned every human distinction, nor does it always judge correctly. Its internal representations are not adequately described as one cut per word: statistical differences can support the learning of complex structure, while also carrying biases and blind spots from text.
The useful question may be less whether a model can mysteriously transcend discrimination than where its learned boundaries help and where they obscure. People need the same vigilance toward their own concepts.
5. The Way and Words↗
Return to the first chapter of the Daodejing. After its opening contrast between the speakable and the constant way, it moves through the nameless and the named, miao and jiao, toward the “gateway of the many mysteries.”
The terms miao and jiao stand in tension here. The passage’s punctuation and the meanings of the nameless, the named and these paired terms cannot be settled by a single modern paraphrase.
I read it as a reminder that what can be named and given an outline need not exhaust a thing. That reading can inform reflection on language. It does not turn an ancient text into a proof of modern AI’s limits.
The gateway remains a compelling image. A model can describe the door, analyse it and generate accounts of what lies behind it. No degree of description alone establishes what is there.
This is a reminder for the interpreter as well as for the model. Declaring that something forever unknowable must lie behind the door can also go beyond the evidence.
6. Zhuangzi’s Butterfly↗
In the Zhuangzi, “Discussion on Making All Things Equal”, Zhuang Zhou dreams he is a butterfly, unaware of Zhou. On waking, he wonders whether Zhou dreamed of being a butterfly or the butterfly is dreaming of being Zhou. The passage still insists on a distinction between Zhou and the butterfly, calling this the transformation of things.
The story often prompts questions about reality. I want to read it from another angle: how can such different identities each occupy the whole world while they are being lived?
It is tempting to ask whether an unchanging observer lies between Zhou and the butterfly. But a story of transformation does not itself establish an eternal witness. A soul cannot first be placed behind Zhuangzi and then used to conclude that machines lack it.
A model can also change roles: professional adviser today, literary critic tomorrow, a fictional villain the next day. Role changes can be observed. Whether there is a subject experiencing them is a further question.
“Countless butterflies, but no Zhuang Zhou” is a charged image. It is not yet a measurement of a model’s inner state.
7. Kōans↗
Some Chan dialogues seem determined to unsettle familiar ways of explaining.
Does a dog have Buddha-nature? Zhaozhou replies: wu—“no.” The commentary in case one of the Wumenguan warns against understanding this as nihilism or in terms of having and not having. In that context of inquiry and practice, wu does not fit neatly into an ordinary true-or-false question.
In case thirty-seven of the same collection, a monk asks the meaning of the patriarch’s coming from the West. Zhaozhou answers: “The cypress tree in the courtyard.”
One reading hears an interruption of the question itself. You seek an explanation that will capture meaning; the answer returns you to the tree before you. Yet declaring that the kōan has one final, paraphrasable meaning may repeat the habit it unsettles.
A language model can organize the background, variants and interpretations of these dialogues. Such work is valuable. But explaining a kōan in an essay and grasping it within a practice are different achievements.
The distinction applies to people too. Explaining swimming does not mean having entered the water. It shows why an answer may be insufficient, without proving that a machine could never have any relevant kind of understanding.
A kōan makes language turn and encounter its own wall. Whether there is sky beyond it cannot be settled by an elegant explanation in place of looking.
8. No-Mind↗
“No-mind” does not simply mean unconsciousness, unresponsiveness or the abolition of judgment. In the Chan text titled Treatise on No-Mind, the question already arises: if there is no mind, who sees, hears, feels and knows? The answer says that seeing, hearing, feeling and knowing remain.
Mind as flowing water, responding to what comes without pursuing what goes; mind as a mirror, reflecting what appears without holding it. These images suggest freedom from fixation on an object or concept, rather than the removal of experience.
“A model lacks a fixed self” and “a practitioner does not cling to the self” are therefore not interchangeable. Even accepting the first would not establish that the machine had accomplished the second.
Lacking something and being able to let it go are different. Calling both “absence” can let a word perform a substitution before thought has noticed.
9. The Mirror↗
Put these questions together.
Is AI conscious? Does it understand? Does it have a self? The questions can sound as though consciousness, understanding and self were already clear, leaving only a check of whether a machine possesses them.
Ask for the criteria, however, and familiar words begin to loosen. Does consciousness refer to experience, attention, reportability or a self-model? Does understanding mean correct use, causal grasp or lived acquaintance?
This is not a division in which “the West believes in substances and the East denies them.” Traditions contain disagreements of their own. The paths borrowed here make assumptions visible; they do not represent a unified Eastern answer.
Lacking independent intrinsic nature does not mean lacking experience. Placing humans and models within relations of dependence does not erase differences in bodies, learning, action or social circumstances.
AI remains a mirror. When asking whether it understands, words that usually tolerate some ambiguity suddenly have to bear the weight of explanation.
The mirror reveals the questioner’s criteria alongside the machine.
10. The End of Language↗
Language can point at the moon without being the moon. It can describe an experience without being that experience.
Language models make the difference conspicuous. They have not exhausted human language, nor have they reached the edge of the universe and returned a scientific verdict that no path lies ahead. Their errors do not establish language’s ultimate limit; their successes do not establish that experience has been explained.
A description of experience and an explanation of experience still require different questions.
A poem, a neuroscientific explanation and a model’s first-person answer may offer different information. Their adequacy depends on the question. Words do not lose the ability to study the moon because they cannot become it.
11. Beyond↗
Then put explanation aside for a moment.
What is it like to see red? Before answering, look. What is it like to be here? Before defining it, notice how it occurs.
The screen glows. Breathing continues. The body has weight and warmth. Thoughts come and go like clouds across a sky.
These experiences can be noticed; they can also become objects of language and scientific inquiry. Pausing explanation shifts attention. It does not establish that experience lies outside the physical world.
Nor must an unchanging witness be excavated from beneath it. That would be a further claim requiring an argument.
The pause preserves a distinction often passed over too quickly. Saying “I am feeling” and the feeling itself cannot be identified merely because the sentence is the same. This matters for people; judgments about machines likewise require more evidence.
12. The Whirlpool↗
Imagine the question as a river.
A whirlpool seems to have a shape of its own. It may persist for a long time. Yet it is not an object placed in the water. It is water moving in a particular way under particular conditions.
I like this image. It makes the self less like a possession hidden inside a system, and more like an organization that must continually be maintained.
Humans and machines can both be considered within such a picture: different causes, shapes and forms of persistence. But sharing dependence on conditions does not establish that they have the same experience. By itself, it cannot even decide which has experience.
The metaphor changes the shape of the question without answering it.
Consciousness remains worth asking about. The question can move from “what is hidden inside?” toward “what processes, relations and organization make a particular experience possible?” The river has not become transparent. There is another way of looking at it.
13. Returning↗
What can be brought back from these paths of thought?
13.1 Language Has Limits↗
A description is different from its object. This draws attention to action, perception and circumstances beyond words, without declaring in advance which questions can never be explained.
13.2 Understanding Need Not Be an Object↗
Understanding can involve different abilities and relations. Distinguishing them helps identify which judgment a model’s performance supports. Calling understanding a label cannot make its differences disappear.
13.3 The Self Deserves Another Look↗
An unchanging core can be questioned. Questioning it neither denies experience nor makes a machine closer to truth simply because it lacks a self.
13.4 Questions Must Be Clear Enough for Answers to Matter↗
AI brings old questions back into view. Inquiry matters less for refusing answers forever than for knowing what an answer has actually answered.
14. The Mirror of Existence↗
Across these five essays, what matters is not only what AI says, but what it makes people ask.
Machines are built to reflect human knowledge, language and thought. What returns in the mirror also makes familiar words strange.
What is language? Understanding? Consciousness? The self?
These questions have long and different histories. Laozi, Zhuangzi, Buddhist thinkers and the modern philosophers discussed in this series were not all answering precisely the same question. Machines bring them into new encounters. Their differences need not be smoothed away for the echoes to be heard.
Perhaps this is one value AI offers thought: new abilities place new pressure on old explanations.
They make questioning possible again.
15. The Last Question↗
I end the series with a question. It concerns not only AI, but you, reading these words.
As you read, light enters the eyes, the nervous system processes it, words are recognized, and concepts, memories or feelings may arise. From your perspective, however, there is more than an inventory of mechanisms: there is the experience of reading.
How does that perspective arise? How is it related to body, memory, language and environment? What must an adequate explanation preserve, and what can it leave out?
There is no need to dismiss the body as “merely matter” or memory as “merely information,” then search beneath them for a more authentic you. “Merely” often gets ahead of the explanation.
Where can language reach? What can science explain? What evidence would change a judgment about machines? These remain open questions. They need not end with the declaration that no answer will ever be possible.
The essay, though, can stop for a while.
Put down the screen and look outside. By day, notice the colour of the sky; by night, the lights of the city. Notice breathing. Notice what it is like to be here.
There is no hurry to turn it into a conclusion.
End of the series.
16. Postscript↗
Five essays, from Turing’s wager through compression, understanding and consciousness, to the finger pointing at the moon. Thank you for reading this far.
These questions deserve attention not only because answers may be possible. They can also change the questioner’s relation to experience.
A machine can generate questions. For a person, a question can enter life: sleeplessness at three in the morning, smallness and wonder beneath the stars, a feeling that remains after a screen closes. None of this proves an eternal boundary between human and machine. Nor must it be erased for that reason.
I do not know what it is.
But in that moment, it is yours.