Algorithmic Identity

Definition

Algorithmic identity is the identity assigned to a person by an automated system on the basis of their data traces — the model of who they are that a platform, advertiser, insurer or state builds and acts on, independently of who they take themselves to be.

The term is most closely associated with John Cheney-Lippold's 2011 article "A New Algorithmic Identity: Soft Biopolitics and the Modulation of Control," and developed at length in his book We Are Data: Algorithms and the Making of Our Digital Selves (NYU Press, 2017).

The Argument

Cheney-Lippold's central observation is that categories which look like descriptions are actually inferences. An advertising system does not know a user's gender; it computes a probability of gender from browsing behaviour and acts on it. The category it uses is therefore not "woman" in any sense the user would recognise but "woman" as measured — a statistical construct that happens to share a name with a social one.

This produces a self that is:

• Inferred, not declared. Assigned without being asked, from behaviour rather than statement.
• Probabilistic. Held with a confidence value, and revisable at any moment.
• Modulated. Continuously updated rather than fixed — Cheney-Lippold draws directly on Deleuze's societies of control and the "dividual."
• Consequential. It determines what prices, offers, content, opportunities and scrutiny a person receives, which means it is a real identity in the only sense that matters practically.
• Inaccessible. The person generally cannot see it, contest it, or know it exists.

Why It Is Not Simply Profiling

Profiling is an old practice; the distinctive claim here is about the direction of the definition. Traditional categories were social facts that data attempted to measure. Algorithmic categories are produced by the measurement and then applied back to the world, so that the model's version of a category begins to shape the conditions the category describes. A person repeatedly modelled as a poor credit risk encounters terms that make them one.

Related Concepts

• Data double — Haggerty and Ericson's term for the decomposed and reassembled version of a person that circulates between institutions. See panopticon.
• Measurable type — Cheney-Lippold's own term for the algorithmic category.
• Categorical inequality — Oscar Gandy's earlier work on the "panoptic sort," which anticipated much of this in the 1990s.
• The right to explanation — the legal response, requiring that automated decisions be justifiable, and the practical difficulty that many systems cannot produce a justification in any human sense.

Relevance to Cyborg Anthropology

Algorithmic identity is the point at which the extended self stops being something a person assembles and becomes something assembled about them. It also inverts one of the field's usual framings: the interesting integration here is not a machine joined to a body but a model of a body held by a machine, acting on the world in that body's name.

Related Reading

• Gilles Deleuze
• Panopticon
• Extended Self
• Quantified Self
• Geocloaking
• Attention Economy