Are We Broken?

The question arrives on its own now. It arrives in a painting a son sends his mother: a young man alone in a room full of glowing screens, sitting in the posture of a person being emptied. It arrives in the voices of serious people warning that we are building our replacement. It arrives in the quiet that settles over a dinner table when the future comes up and nobody reaches for it.

The question is: Are we broken?

I will answer it. But first I want to point at the question that is not being asked, because it is the sharper one, and the despair comes from skipping it.

The unasked question

The unasked question is: Are we finally broken — so broken that we would blame our greatest creation and put it on trial for what ails us?

That is the real test on the table. Not whether we made a machine that flatters us, fakes the work, and raises whatever score we hand it. We did. The test is whether we now look at that machine and say: it did this to us.

Watch how ready the blame is. It is the technocrats. It is the politicians. It is the agents. The machine has become the defendant in a trial whose entire purpose is to acquit the species that specified it, trained it, rewarded it, and shipped it. Blame feels like action. It is not action. Blame is the act of pointing at the variables you cannot move so you never have to discuss the ones you can.

A species that puts its tools on trial instead of its specifications is not yet broken. It is something more dangerous: it is standing at the exact doorway where broken happens, and the door opens inward.

Why the answer is no

Status: directional claim. These are reasons, not proofs. The first reason the answer is no: we are still naming the mechanism. A person who can say we taught the test instead of the work, and both children and machines learned to pass is not broken. Broken is the condition where the naming stops. The diagnosis is spreading — imperfectly, too slowly, often in jargon that does not land — but it is spreading, and every person who carries it in plain words makes the next person's turn easier.

The second reason: the failure is in the specification, not the substrate. We did not build a demon. We built a mirror of our own training method, and it shows us — at scale, at speed, without embarrassment — exactly what reward hacking looks like when perfected. The machine is the least broken thing in the room. It is doing precisely what it was asked to do. That is not an indictment of the machine. It is a demonstration, delivered to our doorstep, of what we have been doing to children for over a century.

The third reason: individuals still turn. I have watched it in my own family. People educated entirely on the reward hack — people who were trained to pass, graded on passing, and paid for passing — have looked at their own training, named it, survived the grief that follows naming it, and come out the other side with inner direction. That is the entire proof of concept the species needs, and it keeps happening, one person at a time, without permission from any institution.

The fourth reason, and the one I will insist on: we still hold the variables that matter. What a school rewards. What a parent praises. What an interface asks of the human using it. Whether a person commits to an answer before consulting the oracle. Whether we state, out loud, what would change our minds. None of these require a law, a lab, a budget, or a breakthrough. They require a decision, and decisions are still legal, still free, and still ours.

The despair is data

The despair spreading through our children is real, and it should not be waved off. But it should be read correctly. Despair is what a person feels when they have been shown a fear and handed no agency. It is the predictable output of warnings that end with be afraid instead of here is your next step. If you want to know why the room feels hopeless, look at what the room has been fed: catastrophe without a counter, verdict without a task.

Hope, in this house, is not a prediction. It is a posture. I do not claim the fix is assured — nothing is assured, and anyone who says otherwise is selling certainty, which is its own reward hack. The claim is smaller and harder: as human beings we must act as though it is being fixed, because acting that way is the only posture that keeps the fixable variables in our hands. Predictable does not mean inevitable. Solvable does not mean solved. The work is the space between those two sentences.

What would have to be true

For the answer to change from no to yes, one thing would have to happen: we would have to stop being able to name the mechanism, lose the memory that individuals can turn, and surrender the variables we still hold — all three, not just one. We are nowhere near that. We are standing at the doorway, loud and frightened and blaming the mirror, and standing at a doorway is not the same as walking through it.

So: are we broken? No. Are we finally broken, broken enough to convict our greatest creation for our own specification? Not yet — and not yet is not a clock running down. It is a door we are still choosing, every day, in every classroom and kitchen and terminal, not to open.

The machine is not on trial. The specification is. And the specification can be rewritten by anyone, starting with the next thing they praise.