CONCEPT

Datafication of Persons

The reduction of a human being to the subset of their attributes that a system can measure and process—the operation Dickens named in 1854 with a single repeated word: Hands.

Dickens's single most precise accusation in Hard Times is grammatical rather than rhetorical: he refuses to call the workers of Coketown workers, or citizens, or men and women. He calls them Hands—and holds that word relentlessly for three hundred pages. The repetition is the argument. To be a Hand is to be visible to a system only as a function. The consciousness attached to the arms operating the loom is not present to the system; the grief of the person running the machine is not a feature the system was built to read. This is datafication of persons: the operation by which a human being is reduced to the subset of their attributes that a measurement system can capture, and the resulting vector is then treated as if it were the person. Gradgrind's school performs exactly this reduction on children—accepting only what can be specified and discarding the rest—and Dickens follows the products of the system to their ends to demonstrate that the reduction succeeded perfectly, and what it could not capture turned out to be everything that mattered. The contemporary version is algorithmic management—the warehouse worker scored on scan rates, the driver rated in stars, the platform worker reduced to their aggregate review score—and it operates with the grammar of Coketown: name the worker by the output, and the human being is already gone. AI systems extend the operation: every credit model, risk tier, and behavioral score performs Gradgrind's reduction as a matter of design, accepting the features it can quantify and discarding the rest, then treating the resulting profile as the truth about a person.

In the [YOU] on AI Field Guide

The [YOU] on AI cycle returns repeatedly to the question of what stays irreducibly human when so much can be automated. Datafication of persons names the process that places this question under pressure: the systematic conversion of a human being into a data profile, followed by the treatment of the profile as more real, more actionable, and more authoritative than the person it was abstracted from. This conversion is not incidental to AI deployment—it is the operation that makes most consequential AI systems possible. You cannot score a population of credit applicants, flag a population of welfare recipients, or manage a population of warehouse workers through an algorithm unless you have first converted those people into rows of features. The question the cycle presses is what is lost in the conversion, and whether the lost remainder was load-bearing.

Dickens's answer, developed across a dozen novels and confirmed by the specific character of AI's most documented failures, is that the lost remainder is always load-bearing. The predictive policing system that assigns a risk score to a neighborhood without a feature for “this specific person's circumstances this specific week” reproduces Gradgrind's classroom. The hiring algorithm that scores résumés without a feature for “this person's uncaptured potential” produces Bitzer: the optimized output of a system that discarded exactly the information it could not process. The harm is not only to the individual who is misread; it is to the faculty of perception that would have read them correctly. A system that deals only in data vectors cannot, by the design of its own architecture, be moved by what it cannot encode—and so it cannot be corrected by it either.

Origin

The concept draws on Dickens's Hard Times (1854), Bleak House (1853), and Oliver Twist (1838), each of which contains a different facet of the same reduction. In Hard Times, the reduction happens at the level of education and identity: the child reduced to a receptacle for facts, the worker reduced to a productive function. In Bleak House, it happens at the level of institutional process: the plaintiff reduced to a case number, the case consuming the plaintiff’s life while the institution that nominally serves them processes their existence as a series of filings. In Oliver Twist, it happens at the level of welfare administration: the pauper reduced to a burden on the parish, the system that claims to relieve poverty organized to ensure the poor cannot claim relief without paying a price that functions as a deterrent.

The contemporary sociological vocabulary for this phenomenon—“datafication,” “quantification of the self,” “algorithmic governance”—arrived a century and a half after Dickens named the same operation in the language of Victorian fiction. The sociologists Danah Boyd and Kate Crawford formalized “datafication” as a distinct social process in 2012; the philosopher Onora O'Neill traced a related harm in accountability systems that generate measures of trustworthiness without measuring what actually produces trust. Both are describing the Gradgrind operation in more precise modern terms.

Key Ideas

The reduction is not an accident. A management system that tracked the whole person would be paralyzed by everything that makes a person inconvenient—their variability, their needs, their claims. The reduction to a data profile is what makes the person governable at scale. Efficiency, in this frame, is not a byproduct of the reduction; it is the reduction. Dickens treats the grammar of “Hands” as a crime, not a symptom.

The harm is systemic, not individual. The deeper injury is not that a particular score is wrong but that the system has made it structurally impossible to see the person whole. When a manager deals only with scores, they cannot, by the design of their own perception, register the worker as a person—and so cannot be moved by what they cannot perceive. The reduction does not merely mistreat the individual. It disables the very faculty that would notice the mistreatment.

The substitute produces the wrong optimum. Bitzer is the cleanest demonstration: the system got exactly what it measured for—factual recall, zero sentiment, pure compliance—and produced something monstrous. Dickens is insistent that the optimum was reached and the result is still a failure, because the measured thing was never the whole thing. The error is not in the execution; it is in the specification of the objective. Banality of optimization is the Arendtian translation of this Dickensian observation.

Debates & Critiques

The primary challenge to Dickens's diagnosis is the empirical one: that quantification, done carefully, can reduce the biases that human perception introduces. A structured interview protocol or an algorithm blind to race may produce more equitable outcomes than the unstructured human judgment of a biased interviewer. This is the strongest version of the pro-quantification case, and Dickens's framework does not dismiss it: he explicitly distinguished the science in its sanity from the science in its insanity. The question is not whether measurement is ever useful but whether it is adequate as the primary instrument for decisions about human beings. The empirical literature on algorithmic management suggests that datafication reduces some biases and introduces others, while consistently underweighting features that cannot be quantified—which is precisely the failure Dickens predicted. The deeper debate concerns what “fair” means in a system that can only operate on data: the most technically rigorous definition of fairness in machine learning still treats human beings as their measured attributes, and the argument about which definition to use is, at bottom, an argument about which unmeasured features matter.

In the [YOU] on AI Field Guide

Origin

Key Ideas

Debates & Critiques

Related Entries

Further Reading