Indian factory workers train AI systems that may replace them

This policy benefits the factory management and the distant shareholders of the artificial intelligence firms by a substantial, quantifiable margin in efficiency and data acquisition. It harms the garment workers in India by an immeasurable degree in dignity, security, and future livelihood. The arithmetic is uncomfortable, but the arithmetic is the argument. We must count.

Let us look at the scene: a garment worker in India, handed a head-mounted camera by a supervisor. This is not merely a tool; it is a leash. The worker’s labor is no longer just the stitching of fabric; it is the feeding of a machine that will eventually render the stitching obsolete. The management gains the immediate pleasure of increased productivity and the long-term pleasure of capital preservation through automation. The worker gains the immediate pain of surveillance - the constant, unblinking eye of the algorithm recording every movement, every pause, every breath. But the deeper pain is prospective. The worker is being paid to build the very instrument of their own obsolescence.

We must apply the Felicific Calculus with rigor. First, intensity. The intensity of the manager’s gain is high, but it is distributed among a few. The intensity of the worker’s fear is acute, but it is borne by many. The anxiety of knowing that one’s current labor is training the replacement is a unique form of suffering. It is the pain of watching one’s future dissolve in real-time. Second, duration. The manager’s gain is sustained, perhaps for years, as the AI system matures. The worker’s pain is immediate and potentially permanent, leading to unemployment or precarity. Third, fecundity. The manager’s gain has high fecundity; the data collected will spawn further efficiencies, further cost-cutting, further profits. The worker’s pain has high fecundity too; the loss of job security breeds community instability, mental health crises, and social unrest.

But here is where the calculus turns. We must consider certainty. The management is certain of the data’s value. They are uncertain of the worker’s reaction, yet they proceed. The worker is uncertain of the future, which is a source of profound distress. Uncertainty, in the moral calculus, is a multiplier of pain. To live in a state of suspended judgment regarding one’s livelihood is to live in a state of constant, low-grade torment. The supervisor handing over the camera is not just distributing hardware; they are distributing dread.

Now, let us examine the extension. How many are affected? The workers are numerous. The managers are few. The principle of utility demands that we weigh the happiness of the many against the happiness of the few. If the pleasure of the few is bought with the pain of the many, the balance is negative. But we must also look at the quality of the pleasure. The manager’s pleasure is one of convenience and profit. The worker’s pain is one of existential threat. To treat these as commensurable is to commit a grave error in measurement. The pain of losing one’s means of survival is not equivalent to the pleasure of saving a fraction of a percent in overhead.

Consider the irony. The workers are instructed to film their work processes for AI training. They are the teachers of their own executioners. This is not just exploitation; it is a perverse form of self-harm mandated by the employer. The worker is forced to contribute to the destruction of the very niche they occupy. If a man were forced to dig his own grave, we would call it murder. If a man is forced to train the machine that will dig his professional grave, we call it “innovation.” The label does not change the calculus. The pain is real. The gain is real. The net is negative.

We must also address the contested point: financial compensation. The workers are not compensated for the training data they generate, nor for the future loss of employment. They are paid only for the stitching. This is a failure of contract and a failure of justice. The value created by the worker’s data is not reflected in their wages. The surplus is extracted entirely by the management. This is not free trade; it is coercion disguised as employment.

The reform implication is clear. The legislator must intervene. The collection of biometric and behavioral data from workers for the purpose of automation must be regulated. The workers must be compensated for the data they generate. More importantly, there must be a guarantee of future support for those displaced by the systems they help build. The calculus does not allow us to ignore the long-term suffering of the many for the short-term gain of the few.

The image of the wax head atop Bentham’s skeleton: even in death, utility can be served. But the living must not be treated as mere data points. The head-mounted camera is a symbol of the reduction of human beings to inputs. We must reject this reduction. The policy is unjust because it fails the test of aggregate happiness. It concentrates gain and disperses pain. It is a recipe for social instability and moral decay. The rational legislator will ban the practice until proper safeguards and compensations are in place.

The worker looks up, not at the supervisor, but at the lens. In that reflection, they see not their own face, but the ghost of their future self, already automated, already replaced. The tragedy is not that the machine will work. The tragedy is that the worker helped it learn how.