socio-algorithmic construction of identity categories based on anethnographic study

 One of the triggers for my interest in AI KR  has been a certain
divergence I observed between
reality and the labels, These observations then spun the interest into
bias, system integrity, cognitive dissonance, schizophrenia as briefly
mentioned in various threads. KR is key to solve this divergences, although
it comes with challenges.

I received this article via a tweet (credit to FP) and managed to access
the article below, which I think is relevant albeit behind paywall I paste
the concluding remarks

The return of the social: Algorithmic identity in an age of symbolic demise
Dan M. Kotliar
<https://journals.sagepub.com/doi/abs/10.1177/1461444820912535?casa_token=SPnD5M0aiAYAAAAA%3ARant3XSJzRQm-9enJOgz0P-HCjaYPE4TMpqz7J8IdFZhDFMP5gKZujH_3NkMlgt5ou9VgQjcGU8UPg&journalCode=nmsa>
<https://orcid.org/0000-0001-7028-1678>
First Published July 22, 2020 Research Article
<https://crossmark.crossref.org/dialog?doi=10.1177%2F1461444820912535&domain=journals.sagepub.com&uri_scheme=https%3A&cm_version=v2.0>
https://doi.org/10.1177/1461444820912535

The return of the social: Algorithmic identity in an age of symbolic demise
Dan M. Kotliar Stanford University, USA and The Hebrew University of
Jerusalem, Israel Abstract This article explores the socio-algorithmic
construction of identity categories based on an ethnographic study of the
Israeli data analytics industry. While algorithmic categorization has been
described as a post-textual phenomenon that leaves language, social theory,
and social expertise behind, this article focuses on the return of the
social—the process through which the symbolic means resurface to turn
algorithmically produced clusters into identity categories. I show that
such categories stem not only from algorithms’ structure or their data, but
from the social contexts from which they arise, and from the values
assigned to them by various individuals. I accordingly argue that
algorithmic identities stem from epistemic amalgams—complex blends of
algorithmic outputs and human expertise, messy data flows, and diverse
inter-personal factors. Finally, I show that this process of amalgamation
arbitrarily conjoins quantitative clusters with qualitative labels, and I
discuss the implausibility of seeing named algorithmic categories as
explainable

..........article behind paywall

  Conclusion Data-mining algorithms can indeed create unnamed,
incomprehensible categories, and in that, they may disregard language,
social theory, and social expertise. But such obscure algorithmic outputs
often get re-socialized. The old epistemic devices, the traditional
symbolic means, get re-introduced in response to people’s values and needs.
That is, while these categories are automatically created by algorithms
(according to the internal algorithmic structure, and the types of data
they run on), the categories’ names, and in effect, their perceived
meanings, are often added after the fact, in light of various social
factors. Hence, the creation of identity categories entails at least two
stages: an algorithmic, calculative one—which includes the creation of
opaque, quantified, clusters; and a qualitative, communicative one—which
re-introduces language, social theory, and social expertise into the
equation. Identity categories may stem from an algorithmic analysis of some
datafied part of society, but their names often originate from, and are
similarly designed as, something different—as an attempt to answer to the
economic, occupational, or even 1162 new media & society 22(7) emotional
needs of the people who buy and use such systems. In other words,
algorithmic identities are more than “measurable types” (Cheney-Lippold,
2017: 47), and algorithmic categorization involves more than the
calculative version of reality. The identities that these algorithms create
are “qualculated” (Cochoy, 2008), as much as they are calculated. They stem
from automated mathematical analyses, but they also echo the
inter-organizational relationships, intra-organizational tensions, and the
idiosyncratic values of the people who develop, buy, and use such
algorithms. Thus, the creation of algorithmic identity categories does not
signal a complete epistemic turn, nor an absolute symbolic demise. Instead,
these categories are created from what I term epistemic amalgams—complex
blends of algorithmic outputs and human values, messy data flows, and
traditional social labels. The “social” returns, but it is not a
re-emergence of the traditional disciplinary powers, but their reinvention,
their complex amalgamation with other epistemes. Hence, a user may be
described as an “introvert,” but this description is not based on his or
her psychological evaluation. Instead, these labels are primarily aimed at
evoking certain responses in certain cultural actors. And thus, even if
users can see the categories they have been put into, and even if these
categories seem familiar, users cannot know how companies name their
categories, or how many likes, to which kinds of cars, turn them into
“introverts.” Language returns, but the gap between the lingual signifier
and its algorithmically formed signified is wider than ever before.
Therefore, while the return of the social may seem like a ray of light into
algorithmic black boxes, and while named categories may be presented as
enhancing the explainability of algorithmic outputs, it is, in fact, an
epistemic mirage, a simulacrum of sorts (Baudrillard, 1994). The
qualitative stages behind these categories are just as opaque, and just as
black boxed, as the calculative ones. Nevertheless, the amalgamated
structure of these categories, and the fact that they remain opaque, does
not make them meaningless, nor powerless. Thornton (2018), following Kaplan
(2014), recently described the current era as an era of “linguistic
capitalism”—one in which words get commodified and auctioned by online
advertising services, such as Google AdWords. In this case, however, it is
not a “commodification of words” (Kaplan, 2014: 62), but a commodification
through words—a monetization made through re-symbolization. That is, while
the commodification of language strips language from its social context,
and deprives it of much of its symbolic value, the process I describe here
does the exact opposite—it turns the algorithmic clusters into social
categories, thus bringing them back into the social domain. As Dourish and
Gómez Cruz (2018) recently argued, data need to be narrated in order to
work. In this case, identity categories get commodified only after they get
re-socialized, only after a subjective, symbolic value gets re-attached to
them. Thus, while data can indeed form new, digital, types of orality
(Papacharissi, 2015), it seems like human language is still an essential
component in contemporary capitalism, in turning symbolic value into
economic value— an epistemic bridge between quantification and
monetization. Similarly, while language, theory, and expertise may be
superfluous to some algorithmic systems, they are still essential
components for the functioning of algorithmic powers, and for their gradual
instilment into society. This work contributes to our understanding of
algorithms and their power, as well as to our conceptualization of identity
in an algorithmic age. As algorithmic profiling enters almost every social
realm, it is crucial that we better our understanding of the Kotliar 1163
socio-cultural construction of such profiles and the complex epistemic
structures behind such categories. Moreover, the account presented here
contributes to the growing literature on algorithms and culture (Christin,
2018; Dourish, 2016; Ribak, 2019; Seaver, 2017), by illustrating how
people’s values and needs, and their specific social position, affect the
ways people are algorithmically described. Finally, while the research on
algorithmic identity has mostly dealt with the datafication of reality,
this article calls for a scholarly focus on the re-socialization of
datafied life. Seeing algorithmization as a multi-stage process, and
particularly focusing on its qualitative stages, could help identify the
meeting points between algorithms and society, and the exact circumstances
in which algorithmic outputs are culturally shaped