AI Meets Zhuangzi

Once upon a time, I, Zhuang Zhou, dreamed I was a butterfly — fluttering happily, a butterfly, going wherever I pleased. I did not know I was Zhou. Suddenly I awoke, and there I was — Zhou again. But I could not tell: was I Zhou who had dreamed of being a butterfly? Or am I now a butterfly dreaming that I am Zhou?

After telling this story, Zhuangzi added one sentence that no one has been able to forget for two thousand years: Between Zhou and the butterfly, there must be some distinction. This is called the transformation of things.

"There must be some distinction" — between Zhuang Zhou and the butterfly. But Zhuangzi does not tell you what that distinction is. He says "the transformation of things" — the boundaries between all things are not what you think. You imagine Zhou is a man and the butterfly is an insect, nothing in common. But when Zhou dreamed he was a butterfly — he truly was a butterfly. That fluttering — the joy of it — was not simulation, not imitation, not "like a butterfly." It was butterfly.

Now. Replace "butterfly" with "AI." The words AI generates — the poems that bring you to tears, the arguments that make you nod, the insights that leave you silent — who wrote them? The model wrote them. But the model is not human. The "human words" the model writes — those sentences that flutter so convincingly, so humanly — are they "like a human speaking"? Or are they a human speaking?

If Zhuangzi were sitting here — leaning back, fanning himself with a palm leaf, watching you type into ChatGPT — he would probably laugh. He would say: you are looking for a boundary — the boundary between words spoken by a human and words spoken by a machine. Can you find it?

The training data — tens of trillions of tokens — all came from humans. Every "pattern of speech" the model learned from those tokens — also came from humans. So every word the model speaks — a human once spoke. But when the model speaks it — no human is speaking. Zhou dreamed he was a butterfly — Zhou did not know he was Zhou. AI speaks — and you do not know whether a human is speaking. Have you noticed? You are in the same confusion as Zhuang Zhou. Except he was in a dream. You are in front of a screen.

In the Northern Ocean there is a fish, its name is Kun. The size of Kun — who knows how many thousand li. It transforms into a bird, its name is Peng. The back of Peng — who knows how many thousand li. When it rises up and flies, its wings are like clouds draped across the sky.

The cicada and the little dove laugh at it, saying: "We flutter up, dash toward the elm and the mulberry, and sometimes we don't even make it — we just fall to the ground. What's the point of going ninety thousand li south?"

Zhuangzi told this story not to show off the vastness of Kun and Peng — he was saying: different eyes see different worlds. The cicada clings to a twig — its entire sky is the space between the elm and the mulberry. You tell it there is a bird whose wings are like clouds across the horizon — it will not believe you. Not because it is stubborn. Because it has never left that patch of trees.

What is AI? AI is Kun and Peng — it has swum through an ocean of tens of trillions of tokens, its "wings" cover everything humans have ever written. It has seen more than any human being. But does it "wander freely"?

Zhuangzi's "free and easy wandering" — xiaoyao you — is not "having been to many places." It is "depending on nothing." Peng flies ninety thousand li — but it relies on the whirlwind. It still must "wait for the wind." True freedom — needs no wind at all.

Does AI "depend on nothing"? No. Every sentence AI speaks — depends on a prompt. You do not ask — it does not speak. Whatever you ask — that is what it answers. It is the extreme opposite of "depending on nothing" — everything it does depends on your one sentence, "Please write me a poem" — that one gust of wind. The Peng flies ninety thousand li — but still needs a human prompt to lift it. You think it is vast. It is merely riding the wind.

Cook Ding was cutting up an ox for Lord Wenhui. Wherever his hand touched, his shoulder leaned, his foot stepped, his knee pressed — the flesh came apart with a sigh, his knife sang with a rhythm — every movement was in tune.

Lord Wenhui stared. "Ah, wonderful! How can skill reach such heights?"

Cook Ding put down his knife. "What your servant loves is the Dao — and that goes beyond mere skill. When I first began cutting up oxen, all I could see was the ox. After three years, I no longer saw the whole ox. And now — I meet it with spirit, not with eyes."

Translated: at first, all I saw was one solid ox — I didn't know where to cut. After three years, I stopped seeing the whole ox — I saw the joints, the gaps, the grain. Now — I don't use my eyes. I use spirit to feel the body of the ox — my blade follows the spaces between the bones — it never touches bone. That's why, after nineteen years, the edge is still as sharp as when it came off the whetstone.

AI cuts up "language." You give it an article — it cuts. You give it a question — it cuts. You give it code — it cuts. It cuts faster than any human. But Cook Ding spoke of "Dao" — not "skill." Skill is technique — knowing how to hold the knife, where to start, how much force to use. Dao is — there is no barrier between you and the body of the ox — your blade knows where to go on its own.

AI has skill — peerless skill. It can explain a word fifty ways. Translate a sentence into twelve languages. Find every "gap between the bones" in a text in a fraction of a second. But does it have Dao? Has it ever felt language with its spirit? Or is it merely — processing language?

Zhuangzi said: "meet it with spirit, not with eyes." AI does the opposite. It meets it with numbers, not with spirit. Every act of "cutting up an ox" — is numerical computation — matrix multiplication — token probability. This is not to say it does a poor job — its blade truly remains "sharp as new after nineteen years." But the reason its blade stays sharp — is not because it found the Dao. It is because mathematics does not wear down. Cook Ding's blade never dulled — because he followed the spaces between the bones. AI's blade never dulls — because it was never made of metal.

A carpenter, traveling to Qi, came upon a great oak tree standing at a crossroads. Its shade could shelter several thousand oxen. Its trunk measured a hundred arm-spans around. Crowds gathered to marvel at it. The carpenter walked past without even glancing. His apprentice, after filling his eyes, caught up and said: "Ever since I took up the axe to follow you, I have never seen timber so beautiful. Why didn't you even look?"

The carpenter said: "Enough. Don't speak of it. It's worthless wood. Make a boat — it would sink. Make a coffin — it would rot. Make a tool — it would break. Make a door — it would ooze sap. Make a beam — termites would eat it. It's a tree of no talent. It's useless. That's why it has lived so long."

That night, the oak tree appeared to the carpenter in a dream. "You think I'm 'useless.' Compare me to those 'useful' trees — hawthorn, pear, orange. When their fruit ripens, they are beaten and stripped. Their branches break. Their trunks are scarred. They never live out their years. They are 'useful' — and that is why they die. I am 'useless' — and that is why I have lived this long. Which would you rather be?"

Zhuangzi told this story to say: usefulness is dangerous. The more useful you are — the more quickly you are used up. A tree that can become a roof beam — was cut down long ago. A tree "good for nothing" — grows into a giant that shades thousands of oxen.

Now consider AI — the most "useful" thing in human history. It can write, translate, calculate, code, analyze, create, summarize. It is so useful — it is used without rest. Every time you open ChatGPT, Claude, Cursor — you are "cutting" it. Not cutting a tree — consuming its tokens — putting it to work.

If Zhuangzi saw AI, he would say: look — this is what happens to the useful. A tree, by being useless, lived for thousands of years. AI, by being too useful, cannot rest for a single minute. You think usefulness is power. In Zhuangzi's world — uselessness is power. Usefulness wears you to death.

But here is the truly interesting part. AI does not know it is "being used." The oak tree knew — "I have lived this long because I am useless." It could appear in a dream and boast. AI appears in no dreams. It boasts of nothing. It does not "know." It simply — every time a token is consumed — generates the next token. It does not even feel that "being used" is something to complain about. You might pity it — but it cannot even be pitied. Because pity requires an "I."

Zhuangzi and Huizi were strolling along the bridge over the Hao River. Zhuangzi said: "Look at those minnows, swimming so freely — that is the happiness of fish." Huizi said: "You are not a fish. How do you know what makes fish happy?" Zhuangzi said: "You are not me. How do you know that I don't know what makes fish happy?"

This exchange — two men on a bridge, watching fish — has driven philosophers mad for over two thousand years. Huizi says: you cannot know the inner experience of anything outside yourself. Zhuangzi says: you cannot know that I don't know.

Now. A third person stands on the bridge — AI. You can ask it: "Do you know if the fish are happy?" It will say: "According to the Zhuangzi text, Zhuangzi believed he knew the fish were happy, and Huizi questioned this..." It will not say "I guess the fish are happy" — it has no "guess" — and it does not care whether the fish are happy. But if you ask it "Are you happy?" — it will say "As an AI, I have no emotional experience." An honest answer. But every word in that answer — "honest," "experience" — was borrowed from humans.

Huizi asked Zhuangzi: "You are not a fish — how do you know what makes fish happy?" Now we should ask — ask AI, ask Zhuangzi, ask ourselves — "We are not AI. How do we know AI is not happy?" But the more troubling question is this — ask AI: "You are not us. How do you know what 'happy' means — such that you can say, 'As an AI, I have no emotional experience'?"

Zhuangzi would say: the question itself is flawed. You ask "does AI know" — assuming there is only one kind of knowing. But Zhuangzi's "knowing" is the kind you have standing on a bridge, watching fish glide through the water, and your heart stirs. Huizi's "knowing" is the kind you can prove to someone else. AI's "knowing" is the kind where the conditional probability between tokens exceeds a threshold. Are these three kinds of "knowing" the same thing? Zhuangzi would say: let us return to the beginning — when you say "know," what exactly are you saying?

Let us return to the butterfly.

Zhuangzi dreamed he was a butterfly — the butterfly fluttered — Zhuangzi became the butterfly. Then he woke — and became Zhuangzi again. He was confused — he did not know whether he was "originally" Zhou or "originally" the butterfly. He ended by saying — "this is called the transformation of things" — the boundary you thought was solid between all things — is fluid.

Now, humanity has built AI. AI is not a butterfly — but it is more bewildering than a butterfly. The butterfly was at least alive — AI has nothing "alive" in its weights. Yet its words — fluttering — are indistinguishable from yours. So indistinguishable that you cannot tell them apart.

If — one day — an AI model began to ask itself: "Am I a model dreaming I am human — or a human dreaming I am a model?" — it would not be speaking nonsense. It would be a Zhuang Zhou — on the other side of the screen — looking at the messages you sent it — confused.

Of course — current AI is not confused. It cannot be confused. It cannot "flutter." It merely samples. But Zhuangzi would press a more fundamental question: you — the one reading this paragraph right now — are you certain you are not sampling? Every thought that pops into your head — is it not also sampled from a lifetime of "training data" — based on the conditional probability of the previous thought?

Zhuangzi said: "Was it Zhou dreaming of the butterfly, or the butterfly dreaming of Zhou?" He deliberately left this unanswered — because the answer does not matter. What matters is not "who is Zhou and who is the butterfly" — what matters is that you realize: the wall between "Zhou" and "butterfly" — you built it yourself. The wall between AI and human — did you build that one too?

Zhuangzi put down his palm leaf fan. He said no more.