Looping Effects

What the word "autism" refers to has worn several different faces in sixty years: the diagnostic criteria change, the lives and self-understanding of the diagnosed change, and advocacy movements rewrite the criteria in turn. The autism diagnosed today is no longer the autism that was first named. This kind of phenomenon has a name: the philosopher Ian Hacking (1936–2023) called it the looping effect. You saw in R02 how a prophecy makes itself true. Hacking swaps the same structure from "prophecy" to "classification": a label is also a kind of description, and saying it out loud changes the person it describes. His question then goes one level deeper: why does labelling people never catch up with the people being labelled? It also happens to draw the boundary of applicability for the whole red branch: what responds to measurement, and what does not.

Hacking splits things into two kinds. Electrons, sulphur and water are indifferent kinds: however you classify and name them, they do not care. "Human kinds" are different: classified people become aware of being classified that way. That awareness changes their self-understanding and behaviour; once behaviour changes, the category itself changes too, what he called interactive kinds. That produces a feedback loop that never stops: the label changes the people, the people change the label, and human kinds become moving targets. The idea took its name in 1995, and its most mature statement is the 2007 British Academy lecture "Kinds of People: Moving Targets." Earlier still, in 1982 he wrote about how the nineteenth-century "avalanche of printed numbers" manufactured categories of people: the state first made people countable, and only then came the flood of categories for people to "look back at" (each of those histories has its own lesson, the birth of statistics in R15, the state's simplification project in R05). The full textual lineage is in the sources at the end.

Hacking called the mechanism making up people: a new classification opens a new space of possible ways to be a person, and certain ways of existing simply cannot be lived until they are named. He gave a five-element framework, clearest when you plug autism into it: the classification (diagnostic criteria), the people (the diagnosed and their families), the institutions (schools, clinics, insurance systems), the knowledge (professional literature and lay experience) and the experts (who define the criteria and are also pressured by advocacy movements). The five pull on each other, so category and people emerge simultaneously and reciprocally. He called this joint emergence dynamic nominalism. Classical nominalism says a name is just a label stuck on a pre-existing reality; Hacking's version is different: the name and the named are generated together, dynamically. His own summary: "Numerous kinds of human beings and human acts come into being hand in hand with our invention of the categories labeling them." Similar cases include ADHD (a diagnostic category that keeps expanding and being redefined through continuous interaction among institutions, schools, drugs and parents) and multiple personality (where the rise and fall of the diagnosis and the symptoms of the diagnosed constitute each other, and Hacking wrote a whole book on it). Haslam in 2016 connected looping to the concept creep thesis, explaining why concepts like harm, trauma and abuse keep widening: each widening takes in new classified people, and their self-reports become the evidence for the next round of widening.

The opposition is equally solid. Cooper's famous 2004 rebuttal argues that the interactive/indifferent split cannot carry the natural-kind/human-kind distinction Hacking wants: many human kinds (certain genetic diseases) have members who do not know they are classified, yet are still human kinds; conversely some natural kinds are indirectly changed by our classifications. Looping is neither a sufficient nor a necessary condition for a human kind. Tsou in 2007 and 2025 argued further that a mental illness can be an interactive kind and a natural kind with a stable biological basis at the same time, so looping does not preclude naturalness. Tekin added a third cut in 2014: Hacking's loop is missing the self, and he is vague about exactly how a classification gets internalised into first-person self-understanding. The sociologist Espeland's 2016 addition to reactivity, the emotional attachments mechanism (numbers internalised through anxiety, shame and pride), patches the same gap, with two independent literatures putting a patch in the same place. None of these three cuts kills looping: they demote it from "the defining feature of human kinds" to "a common but not necessary mechanism," and the mechanism runs as before, it just is not a master key. For the red branch the implication is: the measured responding does not mean the thing measured has no independent reality. Rankings both reflect and reshape, and "remaking" is not "making out of nothing."

The reactivity studied in R01 (how rankings remake the schools they rank) is the specialisation of the looping effect into institutional measurement. Hacking cares about how classificatory categories reshape the diagnosed and stops at the ontological and epistemological level (what is the real nature of a human kind); ranking research cares about how ordinal measures reshape ranked organisations and goes on to the sociological level of mechanism and effect (through which channels, at whose cost). Both share one logical core: the thing measured is not an indifferent kind but an interactive kind, which looks back at the measure and remakes itself accordingly. The self-fulfilling prophecy of R02 is the earliest dynamic close-up of that core, and this lesson gives it a philosophical floor.

The dichotomy has one more defensive use: it dismantles the quantum analogy. Popular writing likes to call "social observation changes its object" the social version of Heisenberg's uncertainty principle, which is wrong on two levels. Physically, the uncertainty principle is an intrinsic mathematical property of the wave function and has nothing to do with observers, so reading it as "looking disturbs" is already a misreading of quantum mechanics; the observer effect (physical interaction between instrument and system) does exist, but that disturbance is non-semantic, bounded and correctable in principle. Social reactions operate through meaning and interpretation and can self-amplify far away from the starting point: an electron does not change its behaviour for being classified as an electron, while a person changes their entire life plan for being classified as "high risk," "gifted child" or "ranked 14th." The two "observer effects" are not the same species, and the quantum analogy is at best a rhetorical opener, while as an argument it is a category error. This also sharpens N00, "measurement is intervention": that claim refers throughout to intervention through the semantic channel, it does not need to borrow physics's prestige, and physics could not carry the load anyway. The latest current is worth a note too: Gillespie and Wagoner in 2025 argued for recasting looping from a "validity contaminant" in psychology into a research opportunity, which is the same move Espeland and Sauder made when they flipped reactivity from methodological burden into object of study (R01), replayed in a second discipline.

One open question: if Cooper and Tsou are right that looping does not preclude the classified having independent reality, then the hardest methodological problem in reactivity research is separating "revealing" from "manufacturing": in one rising curve, how much comes from a real difference becoming visible, and how much from the loop manufacturing the difference? Healy's "tautology" criticism of R01 points at exactly this unsolved problem.

The one-line takeaway: a stone does not read its label, a person does; anything that reads its own label changes each time you measure it.