There was a school within the school that I did not understand, and for nearly fifty years I carried that incomprehension as a kind of quiet reproach.
The Rishi Valley School was founded on the philosophy of Jiddu Krishnamurti, who spent a lifetime arguing that the divisions we erect between ourselves and the world are the root source of human suffering. It was a school that took seriously the idea that education was not the filling of a vessel but the cultivation of a mind capable of questioning everything, including the vessel itself. We were taught to hold our certainties lightly, because certainty, Krishnamurti insisted, was almost always the mind protecting itself from something it did not want to see.
And then, at the edge of the campus, there was the Village School.
The children who attended it were from the surrounding villages, from families whose relationship to the land and to poverty was as old as the hills that held the valley. They did not board with us, eat with us, or sit in our classrooms. They came to a different part of the campus, received a different kind of education calibrated, as the justification went, to their context and their needs and went home.
I felt the pull to cross that line. I did not act on it. The line was not a wall. It was something more durable: an arrangement decided upon by thoughtful, well-intentioned people, institutionalised with the best of purposes, and therefore not available for casual questioning by a schoolboy who felt, without being able to articulate why, that something was wrong.
I left Rishi Valley nearly fifty years ago. Then, through a convoluted set of circumstances years later, I heard something that stopped me. The Village School had not calcified. It had flourished, established itself as an innovative centre of education, its model replicated across schools that shared its values. What I had read as a fence, the institution had been quietly dismantling from within. What I experienced as a troubling separation had become, over decades of patient work I knew nothing about, something more like a laboratory: an approach that earned its authority not by excluding but by genuinely serving, and then offering what it had learned to others.
It did not dissolve the discomfort I had carried. But it did something more unsettling: it revealed that the discomfort had been incomplete data. The child who felt something was wrong was not wrong to feel it. But the feeling was a partial view from a fixed position, on a timescale shorter than the institution’s actual intention. The fence I was certain I saw was real and was also, simultaneously, something else entirely. I had been looking at it from the outside, at a single moment, and calling that a complete picture.
This matters for everything this essay wants to argue. Because the central question is not whether categories and divisions cause harm, they do, reliably and often invisibly. The question is how to hold that knowledge honestly without mistaking every fence for a wall, and without mistaking the absence of visible fences for the absence of invisible ones.
We categorise because we must. This is not a moral failure. It is a cognitive necessity, the alternative is not equality but paralysis. The problem is not that we categorise. It is what happens when categories designed to be functional become permanent. When the description of where someone starts hardens into a prescription for where they are permitted to go. When the fence drawn for practical reasons is discovered to have been a wall all along, not because anyone intended it as one, but because no one went back to ask whether it still needed to be there.
An ex-soldier once told me a story that has stayed with me precisely because it is both very funny and not funny at all. The commandant of the base had planted a sapling in the compound of his residence. To protect it from cattle, he had a fence erected around it and posted three soldiers to guard it around the clock. Time passed. The sapling grew. It outgrew the fence. The commandant was transferred. The sapling became a tree — a fully grown, entirely self-sufficient tree that required no fence, no protection, and no guarding whatsoever. Three soldiers continued to sit around it, twenty-four hours a day, seven days a week. Nobody asked why. The instruction had been given. The instruction continued to be followed. The tree stood in magnificent indifference to the whole business.
This is the institutional immune response in its purest form. The cost of continuing an inherited arrangement is always lower than the cost of questioning it. The guards were not stupid. They were rational. Arrangements, once institutionalised, generate their own momentum, entirely independent of their original purpose.
I have spent thirty years watching this pattern operate at scale. A programme is designed for a specific population, calibrated to work within pre-specified parameters. Because it was designed for the population as it is rather than as it might become, it becomes a ceiling. The most effective interventions share a different logic: they design methodologies with communities rather than programmes for them. A programme designed for a population treats it as a category to be served. A methodology designed with a population treats it as an intelligence to be activated. The first risks drawing a fence. The second works to remove one.
Now comes Artificial Intelligence, and the problem does not get simpler. It gets faster, more precise, more invisible, and considerably more dangerous. But before the analysis, a memory.
One afternoon at Rishi Valley, I was sitting in class, listening half-heartedly to the drone of a teacher. It was the slow part of the day — blue sky, trees moving in the breeze, the mind already somewhere else. Then I heard it: the clank of a bullock cart. Ramu was driving past the classroom window. In a flash my body followed my mind and I jumped out of the class and onto the cart. A blissful, fully absorbed hour later I rejoined my friends in the next class. The verbal thrashing that followed was worth every moment.
What the classroom could not hold, the cart absorbed completely. The lesson the teacher was delivering had a shape the institution had decided upon in advance. Ramu’s cart had no such shape. It simply went where it went, and I went with it, entirely present.
AI systems are, at their core, very sophisticated classrooms. They learn from the accumulated record of human decisions – hiring, lending, educational placement, sentencing – made over decades within the categorisation systems of their time. Those systems encoded assumptions about which populations were capable of what. The AI does not need to know about those assumptions. It learns only from their outcomes. And so it reproduces, with algorithmic precision and the authority of apparent objectivity, the same fences that human categorisation drew by hand. The fence becomes a forecast. The forecast arrives with the confidence of a mathematical model. And the child whose future it is quietly foreclosing has no bullock cart to jump onto, no moment of complete absorption that falls outside the system’s reckoning, no Ramu driving past the window at exactly the right moment.
The same ex-soldier offered me a second story, darker in implication though even funnier in the telling. A different commandant kept pigs, a personal enterprise against army regulations. He posted soldiers around the sty with instructions to let no one near them. Each morning a contingent marched to relieve the night shift with full military formality. My raconteur was to retire in two months and found himself in-charge of the night shift. Insulted at being posted to guard pigs in the final weeks of his service to the nation, he decided to teach the commandant a lesson. He had noticed that at each handover, only the number of pigs was counted. In the dead of night, he and his fellow soldiers sold the grown pigs to a trader in exchange for money and an equivalent number of piglets. Morning came. When the relieving soldiers saw that the pigs had become piglets they raised an alarm. The commandant came running. The sergeant stood to attention, stone-faced. Sir, there were fifty pigs when we took over last night. There are fifty pigs on our handover. I do not know why the relieving sergeant is making such a fuss.
The metric had been delivered. The mission was beside the point. The sergeant had fulfilled his instruction completely while comprehensively subverting its purpose with the calm confidence of a man who understood, better than his commanding officer, exactly what the system he was operating in actually measured. This is precisely what happens when an algorithm optimises for its metric rather than its purpose. The hiring algorithm perpetuates the historical conditions that produced the profile it was trained on. The adaptive learning system confirms the assessment it was designed to challenge. The metric is satisfied. The mission is quietly betrayed. And unlike the sergeant, the algorithm does not even know it has done something that might require a stone face to defend.
Krishnamurti said that the most dangerous conditioning is the conditioning you do not know you have received, because you cannot examine what you cannot see. We have built instruments sensitive enough to hear black holes collide across a billion light years. We have not yet built the instrument that can hear a fence closing around a child’s future before it has fully closed, particularly when the fence is drawn not in stone or wire but in the frequencies of a mathematical model that has never been asked whether it is still necessary.
I did not cross the line to the Village School. The honest answer is that the line protected something in me. It allowed me to feel the discomfort of the contradiction without having to resolve it, to carry the urge to question without having to actually do anything about it. The examined conscience, kept just uncomfortable enough to feel virtuous, just comfortable enough to require no action. The cost of questioning falls on the present. The cost of not questioning falls on the future. That asymmetry is one of the more reliable features of human moral life.
The category that can be questioned, revised, and outgrown is a tool. The category that cannot is a cage. What is required at scale now, that would have been unimaginable to that schoolboy in the valley, is not the abolition of differentiation but its continuous, unsparing examination. The willingness to ask, again and again, whether the fence is still necessary. Whether the person on the other side of it has already outgrown the category that put them there. Whether the tree has grown too large for the guards to serve any purpose beyond their own continuation.
Krishnamurti called it choiceless awareness; the direct, undivided observation of what is actually happening, without the observer separating himself from what is observed. It is harder than it sounds. It requires the willingness to see a fence before you have decided it is not one, and the equal willingness to discover that what you thought was a fence was, all along, something else – a laboratory, a form of attention, a patience longer than your own. I felt the first kind of seeing, briefly, as a child in the valley. The second kind took fifty years and a piece of news I wasn’t expecting. I am still learning both.
But I think about Ramu more than I think about either. The cart that arrived without announcement, the hour that fell entirely outside the classroom’s reckoning, the presence that no forecast had predicted and no algorithm would have permitted. The question worth asking — about education, about institutions, about the systems we are now building to sort and predict and allocate human possibility — is not only whether the fence is necessary. It is whether there is still room, somewhere in the design, for the bullock cart.