1. Origins↗
Begin with meritocracy. Allocate opportunities by merit; give work to people capable of doing it. It sounds reasonable, doesn't it?
Yet Michael Young's The Rise of the Meritocracy, published in 1958, is a satire. A word later taken up as an ideal appears here as a warning.
When “your failure is entirely your own fault” becomes a social narrative, even complaint can become difficult. You have not merely missed out on a position. You must accept a deeper verdict: “You didn't get it because you weren't good enough.” Failure becomes an inward shame.
This is the turn worth examining. How does selecting people for their abilities become defining them by their results? At what point does “Are you suited to this work?” quietly become “Are you worthy of respect?”
2. The Paradox of Zero↗
Imagine a person's richness as a vector with more dimensions than can be counted. It holds a sensitivity to beauty, questions about existence asked alone in the middle of the night, the warmth of fingers feeding a stray cat, the raw freedom of running through a downpour.
Now take an examination paper and try to “measure” that person.
In this image, the paper acts like a projection. It retains certain directions: the things called learning objectives. A marker then compresses that already selected representation into a scalar: a score.
From a person to a number.
The image can be made more precise. Select just describable attributes and write them as . A simplified measurement keeps linear combinations, then weights them to produce a score:
The matrix determines what enters the field of view; determines how it is scored. This is neither an exact model of an exam nor an equation for human worth. It brings one operation into focus: selecting before scoring.
Since , there must be a nonzero direction for which . Attributes can change along that direction while the observation and score remain still:
A nonzero vector lying in that null space can even receive a zero observation and a zero score. The vector has not ceased to exist. The measurement cannot see it.
You cannot measure color with a ruler. A zero projection does not imply a zero vector.
That is what the mathematics establishes. Whether a person must display some overlooked talent before deserving respect is a different question. Human dignity should not depend on what an examiner finally manages to see.
3. The Representation Being Judged↗
What an evaluation directly encounters is often a representation of a person: an answer sheet, an application, a handful of comparable numbers.
Re-present: to present again. A representation is not the thing itself, but a rendering, a translation, an approximation. The sentence “This paper measures your ability” leaves something unsaid: which ability, in what situation, expressed in what way?
A limited record can be useful. It makes comparison possible while leaving some differences outside the record. A construct and the procedure that gives it a numerical representation are not the same thing.
Lossy compression makes choices too. It preserves information important for a use while giving up other details. But a human life has no agreed compression objective. A representation is “good enough” for whom, and for what?
The question moves beyond whether a score is accurate. Who decides which dimensions need not be seen?
4. An Encoder's Preferences↗
In machine learning, inductive bias refers to the assumptions and preferences a learning method brings to its task. They make some patterns easier to learn. That does not automatically make them errors.
Bring this perspective to a timed exam, completed independently and marked against standard answers, and its choices become visible:
- It more readily records knowledge expressible on the page than embodied, hard-to-articulate skill.
- It more readily records completion within a time limit than creation that needs a long gestation.
- It more readily compares answers that fit a shared marking scheme than exploration with no settled conclusion.
- It records work completed alone rather than achievements that emerge through collaboration.
These choices can have good reasons. The problem begins when a design made for a particular purpose comes to look like the natural shape of ability.
Once an encoding system has operated long enough, familiarity can begin to supply its legitimacy. “This is how it has always been measured” quietly becomes “This is how it ought to be measured.”
The hardest preference to notice is the one that makes other possibilities disappear from view.
5. One Scale, Many Lives↗
Allocating opportunities by performance rather than birth carries an important promise. But when ability at a particular task expands into an “excellence” capable of ranking everyone, that promise begins to bear more than it can carry alone.
5.1 Outside the feature space↗
Programmers, artists, therapists and independent game developers do not face the same tasks. Being good at one kind of work does not naturally translate into a rank in another.
The same person may look unremarkable in one evaluative space and luminous in another. This need not be a story of a spectacular comeback. Sometimes the person has not changed. Only the direction of the light has.
5.2 More than one decoder↗
A school, an employer, a collaborator and an audience may read the same experience differently. A record does not arrive with one inevitable interpretation.
The question “Does this person have value?” reveals its missing terms: for which task, to whom, and under what conditions?
5.3 The scarcity of rank↗
With a fixed number of places, everyone can improve their score, but everyone cannot improve their relative rank. That is a limit of competition over position, not a reason to dismiss learning. Knowledge and skill need not be zero-sum.
The sharper question is what an additional hour produces. How much is learned, and how much is spent simply to avoid being displaced by someone who has also put in another hour?
6. A Blindness Called “Too Fair”↗
The original provocation was that meritocracy's problem is not being unfair, but being “too fair.”
The phrase needs a pause. Using one ruler for everyone establishes a consistent rule. It does not settle the fairness of starting points, resources or uses of the result. The irony lies elsewhere: the same light can illuminate certain directions equally, and equally leave others in shadow.
A system can apply the same encoding, retain the same few dimensions, and leave outside its view whatever in each person cannot enter that standard.
This blindness can be difficult to contest. What you need to say may not be “This part of me has been undervalued,” but “That part of me does not exist in this coordinate system at all.”
Not everything left unmeasured needs to be measured. But a local judgment cannot gain authority over the whole merely because it cannot see the rest.
7. Reading Competitive Conformity Through Representation Learning↗
What becomes visible if competitive conformity is imagined as everyone trying to enlarge their projection onto the same low-dimensional subspace?
Higher scores. Better schools. More recognizable experiences. The same application templates, the same career paths, the same definition of excellence.
In machine learning, representation collapse includes the degenerate case in which different inputs are encoded as the same constant vector. Converging applications are not that technical phenomenon. The analogy does, however, pose a question: when a representation becomes the ticket to resources, might life begin to grow toward the shape of that representation?
At first, existing experiences may simply be pruned into something recognizable. Later, the choice of what to experience next begins to obey the form too.
Exploration with no box to tick. Care that cannot be converted into points. Wandering that produces nothing. These become harder to explain. They have not vanished, but it becomes harder to give them time.
The deeper concern may be that life begins to accommodate its representation.
8. Loss Function: What Is Being Minimized?↗
In deep learning, a loss function specifies the objective that training seeks to minimize.
“Loss” is an interesting word. It names a mathematical objective, and it names the experience of losing something. These are different meanings, but their proximity opens a question. When a student works to minimize the distance from full marks, what might be lost?
Curiosity unrelated to the exam. An experiment that yields no immediate result. Room to define success differently. The part of a life orthogonal to the answer key.
These are not inevitable costs of learning. The question is whether, when time is scarce and the target narrow, whatever cannot add to the score is always the first thing surrendered.
That loss may never appear on the transcript. Someone can know exactly how many marks they gained, yet have little way of knowing where a question never begun might have led.
9. Ground Truth: Whose Truth?↗
Machine learning has another term: ground truth, the reference label or outcome adopted for a task.
Where does that reference come from? Sometimes from a verifiable event or measurement; sometimes from an annotator's judgment; sometimes from a rule defining correctness. Their reliability has to be examined separately. They cannot all be reduced to arbitrary exercises of power.
Yet even an accurate record leaves another choice unresolved: why should this particular record determine the allocation of opportunities?
An exam may faithfully record an answer without supplying a true label for someone's merit as a whole. A performance measure may accurately record an output while missing some of the work that keeps a team together.
The quietest step is to turn “the standard adopted here” into “the way the world really is.” The question of how a label helps understanding, and how it can conceal an explanation, has another echo in diagnosis.
10. A Possible Way Out: A Society of Many Encoders↗
If the problem is a single encoding, a way forward may lie in making more kinds of seeing possible.
A timed answer, a long-term project and a contribution to collaboration need not converge on one kind of excellence. They can illuminate different parts of a person. One failure need not command every opportunity that follows.
Ensemble methods in machine learning offer a suggestion: different models may complement one another's blind spots. Back in society, the questions remain who can enter these evaluations, who pays the cost of presenting themselves, and whether apparently different rulers still favor the same kind of life.
More light should also leave room not to stand in it all the time. Otherwise, moving from one examination paper to a hundred merely stretches the examination into a lifetime.
Epilogue: When the Shadow Disappears, the Person Remains↗
A friend once died by suicide after an academic setback. A failed exam cannot, of course, explain a person's death. But since then, the distinction between a score and a person has not remained an abstract question.
The disappearance of a shadow does not mean the disappearance of what cast it. A zero projection does not imply a zero vector.
What needs to disappear is not the child who failed, but the single light source that allows that child to see only a shadow.
More light is needed, from more directions, to reach what has remained outside the field of view. So that everyone has a chance to see themselves in some mirror.
A sentence will not make this happen. A single scale makes selection efficient and hierarchies legible. Changing it also touches the authority to define excellence.
Harder still, an internal examiner may survive changes to the external system. Even when nobody else is assigning marks, you may keep assigning them to yourself.
What needs to change includes that familiar way of looking. Not waiting for another mirror that will finally prove your worth, but learning not to mistake any mirror for the whole of yourself.
Mathematics can show what a beam of light leaves out. It cannot live through a person's suffering for them. This essay will not bring that child back. Even knowing the argument might not have saved him. A reasoned explanation is one thing; the suffocating sense of being rejected by the world is another.
All I can do is write this down.
For the next person standing before a zero. For the next person who feels they do not exist.
Your projection has failed. You are not that failure. Nor must you wait for a better angle before being allowed to exist.
You are there. That is enough.