In this post, I sketch out some preliminary ideas for introducing repetition into theories of social formation and for situating cognition at their base. The major principle for this endeavor is what I (unoriginally) propose as loops. More originally, I argue that loops take at least three different forms and that not all looping effects are created equal. Loops are equal, however, in putting the onus on repetition as a source of scaled orders and formations (e.g. “structures,” “enclosures,” “molds,” “modulations”) rather than on generality.
A loop, quite simply, is a generative process with one necessary condition: a non-identity between two parts, but two parts that repeat in their connection. A repeating loop makes a scaled order particular rather than general, because it is fundamentally a sequence. Loops might be broadly instantiated, but they cannot find equivalences everywhere. As repetitions, they need the same ingredients. As connections and cycles, they remain distinctive rather than ordinary.
Douglas (1986: 100-02) suggests that a looping effect need not directly involve a human agent at all; they are mostly natural. Knowledge of microbes leads to medications; when those medications are applied to microbes, the microbes adapt. This explains a clear feedback relation, but it is not a loop: microbes change because of knowledge about them; but the change needs no repetition. It is adaptive instead.
The thing about a loop is that it must repeat (again and again). A loop must generate its own momentum. This link between repetition and loops has been neglected to date. Such neglect means that looping effects are discussed separately from the prevalence of scaled orders and formations, for instance those that are “disciplinary” versus those that are “control,” leaving both of fuzzy meaning and provenance, used interchangeably. When we concentrate on looping effects as repetitions, it becomes clear that orders and formations of scale (e.g. not presumed to be general; that cannot find equivalences everywhere) rest largely on an edifice of cognition that, in practice, does not need to remain implicit in order to be effective.
Hacking loops (enclosures, molds)
For Hacking (1995; 2006), loops come from the process of classification and categorization that feeds a dynamic nominalism. Classifications are made by people about people, as an index of traits, of the properties displayed by the latter. In a Hacking loop, classifications made by people about people loop into their target and alter it. The classifiers “create certain kinds of people that in a certain sense did not exist before.” The onus rests on the name given to these traits that collects them together. Hacking loops therefore represent a form of nominalism: they need not become entangled with real kinds or make any difference for them. What matters more is the name, its legitimation by expertise, its elaboration by institutions, its officialization by bureaucracy, all of which reinforces its external, public and legitimate presence.
As Hacking puts it:
In 1955 “multiple personalities” was not a way to be a person, people did not experience themselves in this way, they did not interact with their friends, their families, their employers, their counsellors, in this way; but in 1985 this was a way to be a person, to experience oneself, to live in society (2006).
The intervention of 30 years meant the formulation of this name, the proliferation of knowledge and the accumulation of references under its heading, all giving it a stable external presence, by indexing various evident things (e.g. “this is that”) in a distinguishable way. While the traits of multiple personality (or manic depression, anxiety disorder, etc) might have preceded the name, this is not Hacking’s point: as a “way to be a person,” multiple personality needed a name that could index these traits and thereafter be a way of indexing of oneself (e.g. “that is me”). Such a sequence (this is that → that is me) becomes fully reversible (that is me → this is that). A name, increasingly standardized, rationalized and externalized (e.g. “discourse”) makes up and stabilizes people by feeding back into their identification (this is that ←→ that is me).
To be a certain type of person, to live in society as that person, to be interacted with as that person, and most importantly to experience oneself as that person occurs through a Hacking loop. Importantly, all of these effects require only an external process rather than a change of “inner life.” A name is “deep” in a contingent sense: the loop rests upon the identification of those who are indexed. It makes no substantive difference for the traits that are indexed that they now receive a name. In fact, Hacking loops seem to require this given the proliferation of names and entire professions and forms of expertise dedicated to classification. However, it does make a difference for those who are classified and named. Hacking loops create an enclosure or mold, from which there might be no escape so long as the name is externally maintained.
Mutually sustaining relations (structures)
A different loop is proposed by Sewell in what we might call his principle of mutually sustaining relations. The terminology here concentrates on a “mutually sustaining” loop between two different kinds of things (schemas and resources), as mentioned in the following influential formula and applied (famously) to a theory of structure:
Structures … are constituted by mutually sustaining cultural schemas and sets of resources that empower and constrain social action and tend to be reproduced by that action. Agents are empowered by structures, both by the knowledge of cultural schemas that enables them to mobilize resources and by the access to resources that enables them to enact schemas (27).
As Sewell implies, schemas and resources mutually sustain each other through a repeating connection. A schema remains an effect of resources, just resources are the effect of schemas.
When the priest transforms the host and wine into the body and blood of Christ and administers the host to communicants, the communicants are suffused by a sense of spiritual well-being. Communion therefore demonstrates to the communicants the reality and power of the rule of apostolic succession that made the priest a priest. In short, if resources are instantiations or embodiments of schemas, they therefore inculcate and justify the schemas as well (13)
Unlike a Hacking loop, Sewell’s “mutually sustaining” loop does suggest a deep effect, as the very constitution of a set of properties as “resources” are schema dependent, just as the constitution of mental categories as “schemas” are resource dependent. A resource is equivalent to the traits that a Hacking loop collects under the heading of a name, but a schema does not “name” them. Sewell characterizes the mutually sustaining link as instead “reading” or “interpreting.” A resource needs to be read as a resource in order to be a resource. A schema does the reading. A schema, presumably, is not a schema if it does not read or interpret resources. The loop can be initiated through either end: resource accumulation to a schema (resource → schema) or schema accumulation to a resource (schema → resource). A loop becomes difficult to sustain in cases that allow for too much agency (e.g. transposition of schemas), which prevents an unambiguous rendering of resources.
In cases where there are limited schemas for “reading” and “interpreting” a resource, and this is in turn “sustained” by limited resources for other possible schemas, a “structure” will result. A structure is distinguishable from a “mold” or “enclosure” in a Hacking loop. Structure, by contrast, suggests not only a potential source of resistance but also the limits of meaning. This entails the “depth” of structure as opposed to the externality of a mold. Structure refers to inner life, which it substantially depends on shaping and altering. The surface-level chaos of capitalism, for instance, only signals the depth of a schema ←→ resource loop: the schematic and repeating transformation of use- to exchange-value is a necessary condition for “resource” in this context; resources, meanwhile, accumulate to “schemas” that involve a use-to-exchange transformation.
We should expect structures to change through disruption to an established loop, via the interchangeability and replacement of both parts of structural loops (schemas and resources). This creates demands on inner life through transpositions that likely appear “impractical” in their interpretations and reading of things. The chance of resource accumulation keeps the possibility of structural change open.
Expectations-chances loops (modulations)
Hacking loops and Sewell’s mutually sustaining loops are both known well-enough by this point as to render the above discussion boring by comparison. To finish this post, I want to make two proposals: first, that Hacking’s “molds” and “castings” and Sewell’s “structures” are both loops found within a disciplinary order. This suggests a relative limit on their generality, though equally they remain contingent on repetition (as loops). Second, I want to understand a disciplinary order as distinct from a control order based on a different loop that engages cognition differently than naming, reading, or interpreting (Deleuze 1992). This is a expectations-chances loop that works according to (objective) prediction and (subjective) guessing (see Bourdieu 1973: 64).
In one version of this loop, the tale is told indicatively as follows:
Acrimonious debates about the calculative abilities of individuals and the limits of human rationality have given way to an empirical matter-of-factness about measuring action in real life, and indeed in real time. The computers won, but not because we were able to build abstract models and complex simulations of human reasoning. They bypassed the problem of the agent’s inner life altogether. The new machines do not need to be able to think; they just need to be able to learn. Correspondingly, ideas about action have changed (Fourcade and Healy 2017: 24).
Hence, a proposal for non-intentional action becomes applicable to data-gathering mechanisms, but the “index” is different in this scenario, as it includes “inner life” no longer. “Culture” is an association rather than internalized pattern generator. It does not have effects, but rather stands for a history of traces:
When people are presumptively rational, behavioral failure comes primarily from the lack of sufficient information, from noise, poor signaling or limited information-processing abilities. But when information is plentiful, and the focus is on behavior, all that is left are concrete, practical actions, often recast as good or bad ‘choices’ by the agentic perspective dominant in common sense and economic discourse. The vast amounts of concrete data about actual ‘decisions’ people make offer many possibilities of judgment, especially when the end product is an individual score or rating. Outcomes are thus likely to be experienced as morally deserved positions, based on one’s prior good actions and good taste.
A theory of action remains, then, even despite the absence of inner life; because data is simply action. Data can modulate action through a “herding” or directing effect, creating futures based on past performance and subsequent encoding. Since there is no inner life, classification is based on information collected at junctures that create possible futures. The causes of action are not of interest (only that action happens), though there are consequences to action. This can exercise a disciplinary effect through anticipation, as facilitated by the rationalization of trials. Since there is no ideal model (or name gathering of characteristics a priori), however, this is not integral to control. There is only the fact that one must have been through certain trials and then out of them.
Predictions made through data protocols interface with predictions made in action. Trials introduce uncertainties that meet with anticipations; a certain future is achievable when possibilities are presented algorithmically and displace an otherwise “wild’ cognition. Control becomes an algorithmic modulation of future possibilities rather than a generative modulation of guesses.
The systematic production of “good matches” is based on controls exercised on the means of prediction from both ends: the expropriation of the means of prediction and the controlled distribution of what they predict. This keeps the loop closed between the (objective) provision of possibilities and (subjective) anticipations or guesses, making “this matching feel all the more natural because it comes from within—from cues about ourselves that we volunteered, or erratically left behind, or that were extracted from us in various parts of the digital infrastructure” (Fourcade and Healy 17).
Modulation takes place through cognitive loops, constructing a “self-deforming cast that will continually change from one moment to the other, or a sieve whose mesh will transmute from point to point” (Deleuze 4). Conventionally, the connection between “schema” and power is content-laden and substantive: it provides a way to “read” resources (Sewell 13). An expectations-chances loop finds no equivalent to “reading” (or interpreting or naming); the key process is guessing instead. A non-individual recorder or record-keeper (qua technology) can guess even if it cannot read, and it can adapt its guesses, improve them. Here looping is incompatible with “molding” or “casting”; “structure” is static by comparison. After all, you can know when you leave the “cast” and its standard no longer applies.
The theory of power embedded in a schema-resources loop puts the onus on schemas that “read” resources; this is where we find agency. In a disciplinary context, an ideal or standard (a telos) is enforced and sought after. In control contexts, such a standard goes missing. Trials are not examinations. A model is volunteered rather than enforced. An individual is a record, though there is no record-keeping individual (“examiner” or “recorder”). Rather than being incorporated into a structure (through schemas), agents are made precise as a code or classification. They do not exercise effects (structural or otherwise) but are given possible futures. They are not shoehorned into the fixed parameters of a schema. They bootstrap themselves into sequences that look increasingly like their own good matches.
We should therefore expect the genesis and transposition of expectations just as we do those of schemas or names, in looping connection with chances, as a way of inviting chance in or taming it. But there is a catch. The consequence of a “controlled” expectations-chances loop can be similar to the amnesiac returning to memory after several long years: “My God! What did I do in all those years?” (Bourdieu  quoting Deleuze 1993). Consider, along exactly similar lines, a “coming to” after diving down an algorithmically modulated rabbit-hole. The explanation must be cognitive because this occurs through repeating loops. Disciplinary formations can achieve (reflexive) “consciousness” and nothing will change; the same is not true for control formations.
Bourdieu, Pierre. (1973). “Three forms of theoretical knowledge.” Social Science Information 12: 53-80.
Bourdieu, Pierre. (1995). The State Nobility. Stanford University Press.
Deleuze, Gilles. (1992). “Postscript on societies of control.” October 59: 3-7.
Deleuze, Gilles. (1993). The Fold: Leibniz and the Baroque. University of Minnesota Press.
Douglas, Mary. (1986). How Institutions Think. Syracuse University Press.
Fourcade, Marion and Kieran Healy. (2017). “Seeing like a market.” Socio-Economic Review 15: 9-29.
Hacking, Ian. (1995). “The looping effects of human kinds.” Pp. 351-394 in Causal Cognition: A Multidisciplinary Debate.
Hacking, Ian. (2006). “Making up people.” LRB 28.
Sewell, William. (1992). “A theory of structure: duality, agency and transformation.” American Journal of Sociology 98: 1-29.
Do you see the Hacking loops as described in this piece as being a sort of extension of Labeling Theory, or distinct from that? It feels like maybe a modern interpretation with a more clear repetitive and/or programmed loop than Labeling, but curious what you think as far as it’s antecedents in past theory and future implication. Seems more applicable with modern technology – twitter, Instagram, hashtag use – which can create fast short repetitive loops of label and identification via external terms.
Out of practice sociology major looking to refresh on current topics and trends so forgive the perhaps unsophisticated approach to this question!