At the core of the ontological and political shifts brought about by the emergence of artificial intelligence, the body becomes a crucial site of conflict and tension over which meanings, identity, and agency are negotiated in the age of machine intelligence.
The Dream of Disembodiment
The story begins at MIT Media Lab when Joy Buolamwini sat down to test a piece of facial-detection software of her own making. She leaned towards the camera. Nothing happened. She adjusted the light, tried again, gave the machine every chance, and still, it could not find her face. Then she noticed a white plastic mask on her desk, the sort sold cheaply for Halloween. Half out of curiosity, she held it up to the camera instead. The screen responded at once. Face detected. The software had recognised a blank piece of plastic before it recognised her. That small, unsettling moment sent her digging further, and led to her landmark research, which exposed how commercial facial-recognition systems performed reliably on lighter-skinned faces, and far less reliably on darker-skinned ones, particularly women’s.
The software in question was produced by some of the world’s largest technology corporations and was already being marketed to police forces, airports, and employers as a neutral and objective means of identifying human beings. The incident would culminate, in 2018, in one of the most consequential research papers of the past decade: Gender Shades, co-authored with Timnit Gebru, which demonstrated that commercial facial-recognition systems misidentified darker-skinned women at rates up to 24.4 per cent higher than those for lighter-skinned men.[1]
Oprah Winfrey, Michelle Obama, and Serena Williams among the most photographed women on earth were variously classified as male, or failed altogether to be recognised as female, by systems then widely available on the commercial market. The story is often recounted as little more than a debugging anecdote: flawed training data, remediable bias, a technical problem awaiting further optimisation. Yet it signifies something far more profound. It is a parable of what it means to be rendered visible to a machine built upon a fantasy far older than computing itself: the fantasy that intelligence can be severed from the body. That vision may exist independently of any situated observer, and that knowledge can somehow be purified of the historical and embodied beings who produce it.
The deepest problem with machine vision is not simply that it is biased, though it is; nor that its datasets are incomplete or skewed, though they are. It is that the underlying operation, extracting a centre, discarding deviation, and projecting recurrence as prediction, has been mistaken for objectivity, when in fact it encodes a determinate political and philosophical inheritance. A particular epistemology, several centuries in the making, has been sedimented into infrastructural form and then re-presented as the absence of epistemology itself.
Buolamwini’s mask was, in this sense, an inadvertent philosophical intervention. It disclosed that the system was calibrated to recognise a very particular kind of face, before universalising that face as the human as such. All others appeared only as deviations, errors awaiting correction within the machine’s operational logic. The mask discloses, in miniature, a far older and more consequential sorting: the machinery by which a society decides who is recognisable as one of its own, and who arrives, and remains, only as an anomaly to be corrected, deported, or denied.
This fantasy is at least four centuries old. When René Descartes divided the world into res cogitans and res extensa — thinking substance and extended substance — he instituted a hierarchy that Western philosophy would spend centuries attempting to undo, and Western technology would spend centuries attempting to perfect. Mind, within this schema, became the privileged site of knowledge. The body was reduced to a recalcitrant vessel: vulnerable to appetite, error, decay, and death. Immanuel Kant refined the formulation further through the figure of the transcendental subject of reason, abstracted from the material and historical conditions of any particular knower.
It is precisely this lineage that contemporary artificial intelligence inherits, typically without acknowledgement. The transhumanist rhetoric animating much of Silicon Valley, from Marc Andreessen’s Techno-Optimist Manifesto[2] to the assorted evangelists of artificial general intelligence (AGI), amounts, in many respects, to Cartesianism financed by venture capital. The body appears here as friction: inefficiency, limitation, an obstacle to computational transcendence. Intelligence, properly conceived, is imagined as reducible to computation; everything else is merely flesh.
The promise presents itself as emancipatory. To transcend the body is, ostensibly, to transcend the constraints the body imposes. Yet the body silently presupposed within this fantasy has never been universal. It has always belonged to someone in particular — young, white, able-bodied, masculine, a body historically afforded the privilege of forgetting its own situatedness. As Amanda du Preez observes, men were granted the luxury of “embodying transcendent minds”, whilst women were consigned to “minding immanent bodies”.[3] Artificial intelligence, then, does not abolish the embodied subject. Rather, it universalises a singular and historically privileged form of embodiment while disavowing its particularity.
The Body Returns
Phenomenology emerged in the twentieth century to insist that the Cartesian division between mind and body was less a description of lived existence than an analytic fiction. Maurice Merleau-Ponty gave this critique its most influential formulation through the concept of the body-subject:[4] the human being who encounters the world is not a disembodied consciousness peering outward through a sheath of flesh, but an embodied creature whose knowing takes place through reaching, touching, walking, stumbling, and seeing. The body is not an impediment to perception; it is the very condition of perception itself.
This matters for the present argument because it suggests that the supposedly disembodied intelligence produced by machine learning is itself constituted through a body-subject of a peculiar and highly distributed kind, dispersed across server-racks, mineral extraction sites, and industrial cooling infrastructures; sustained by the eyes and judgements of underpaid data annotators in Jharkhand, Manila, Nairobi, and Caracas; animated by the labour of women and people of colour whose names seldom appear in technical papers or corporate press releases.
To describe the resulting systems as “disembodied” is therefore merely to reproduce René Descartes’s error in computational form. There is always a body. The political question is whose body is permitted to disappear behind the abstraction, and under what conditions.
Michel Foucault politicised this phenomenological stance. Power, in his account, does not simply repress pre-existing bodies; it actively produces them. Modern biopolitics names, classifies, and segregates bodies; it disciplines and hierarchises them; it distributes visibility and invisibility through institutions that come to appear natural, even inevitable. Kimberlé Crenshaw’s[5] formulation of intersectionality sharpens this analysis further: bodies are never singular objects of governance but are always differentiated along multiple, intersecting axes such as race, gender, sexuality, class, disability, such that any lived experience emerges at their intersection.
There is no abstract subject of harm, only situated and differentiated embodiments of it. Feminist theory has, for decades, been mapping precisely this politics of visibility. Laura Mulvey’s analysis of the male gaze shows how visual regimes are structured by asymmetries of looking; Luce Irigaray exposes how woman is positioned as the invisible ground against which the masculine subject constitutes itself; Judith Butler demonstrates how gender is not an essence but a performative repetition, produced through acts of citation and normativity; and Caroline Criado Perez has shown how these abstractions are materially sedimented, for instance in the “gender data gap” that renders women 47 per cent more likely to suffer serious injuries in car crashes because crash-test dummies were modelled on male bodies.[6]
None of these are incidental distortions of representation. They are the infrastructure of recognition itself: the architecture through which some bodies are assumed in advance, while others appear only as deviation, exception, or error. When this analysis is taken seriously, Gender Shades ceases to resemble a technical malfunction. It begins to appear instead as the system performing precisely as it was configured to do.
The lineage is not incidental. Many of the statistical techniques underpinning contemporary machine learning were consolidated in the late nineteenth century by the British eugenicist Francis Galton and his student Karl Pearson. Galton’s composite photography, the superimposition of multiple portraits to generate an “average” face, was not a neutral precursor to modern image classification. It was an instrument within an explicitly eugenic project concerned with identifying ideal types and deviant forms, criminal physiognomies and racial hierarchies.
Galton named this programme eugenics. The statistical apparatus he helped develop is now embedded, in transformed but traceable form, within many of the systems used for facial recognition today[7] that essentially foregrounds a colonial vision. I do not argue that contemporary engineers consciously reproduce Galton’s politics; most do not, and many would be disturbed to encounter his intellectual inheritance. Rather, the point is that the technical gesture of extracting (a centre from a distribution and treating deviation from that centre as error) is itself already a political operation. It is an epistemic decision about normativity: about what counts as typical, legible, and correct, and what is relegated to the status of noise.
Once seen in this way, the architecture of contemporary machine learning appears less as a neutral apparatus of prediction than as a repetition of a long-standing logic of classification. Training datasets that overrepresent white faces; predictive policing systems that direct attention toward already over-surveilled neighbourhoods and then treat the resulting arrests as confirmation; emotion-recognition tools calibrated to Western expressive norms; medical imaging systems trained predominantly on male anatomy — these are not isolated biases but iterations of a shared operation, reproduced across domains. What is often described as the “neutrality” of these systems is, on closer inspection, the conversion of historically sedimented forms of inequality into apparently objective statistical outputs.
The machine does not simply reflect the world; it metabolises its histories of power, translating them into the language of prediction, and nowhere is this operation more consequential than at the border, where the question of who is legible to a system is also, quite literally, the question of who is permitted to exist within a territory at all. Amoore’s account of the ‘biometric border’ describes how systems such as IDENT, ADIS, and APIS do not encounter asylum seekers and migrants as persons but as data doubles: composite profiles assembled from fingerprints, travel records, and risk scores, then sorted according to which cluster of prior data their attributes most closely resemble.[8]
To belong to a cluster, as Amoore observes, is not a matter of shared characteristics or met criteria, but a ‘spatialised proximity or distance’ from a statistical centre[9], which is the same operation, in different clothing, that produced Buolamwini’s misrecognition. What has shifted, in Amoore’s more recent account of the deep border, is that this sorting no longer occurs at a line on a map. It is inscribed into the body itself and redistributed across software systems and institutional actors, such that the border travels wherever the body travels, and discretion once exercised by a frontline officer is now exercised by an algorithm operating several jurisdictions removed from the person it assesses.
Glitches and Refusal
The essay upholds the encouraging possibility that none of this is inevitable, and a substantial body of feminist and posthumanist thought has already been working to articulate what an alternative might look like.
Donna Haraway’s Cyborg Manifesto[10] refused, as early as 1985, any simple choice between technophobia and technological surrender, insisting instead on hybrid figurations that unsettle the binaries of human/machine, nature/culture, male/female. Karen Barad’s concept of intra-action[11] displaces the assumption of pre-existing entities that subsequently interact, proposing instead that subjects and objects emerge together through specific material entanglements, meaning that what an algorithm “sees” is never a neutral discovery but a co-production.
Rosi Braidotti’s account of the nomadic subject, alongside the vocabulary of becoming and assemblage developed by Gilles Deleuze and Félix Guattari, similarly describes identities as processual, relational, and in flux forms of becoming that statistical classification can only ever approximate by flattening.[12] In practice, these conceptual interventions have begun to find experimental traction. Legacy Russell’s notion of Glitch Feminism[13] reframes moments of machine failure, the misrecognised face, the unreadable body, the image that cannot be stabilised, not as errors to be corrected but as openings in the system’s epistemic authority.
The artist Caroline Sinders has developed a deliberately small and transparent Feminist Data Set[14] through participatory workshops that foregrounds situated knowledge rather than scale or extraction. The Paraguayan artist Kira Xonorika uses generative AI to produce indigenous and trans imaginaries that exceed the predictive horizon of the training data itself.[15] These are not solutions in an engineering sense. They are refusals of a framing in which the only available politics is the incremental repair of systems whose fundamental grammar remains untouched.
The deepest problem with machine vision is not simply that it is biased, though it is; nor that its datasets are incomplete or skewed, though they are. It is that the underlying operation, extracting a centre, discarding deviation, and projecting recurrence as prediction, has been mistaken for objectivity, when in fact it encodes a determinate political and philosophical inheritance. A particular epistemology, several centuries in the making, has been sedimented into infrastructural form and then re-presented as the absence of epistemology itself.
A body is not a data point. A body is a history: of who has been rendered visible and under what conditions; of who has been required to labour invisibly for systems that claim abstraction as their virtue; of which bodies have been used to calibrate the norms by which all others are judged. Crash-test dummies, medical imaging baselines, annotated datasets, policing archives — these are not neutral repositories but stratified records of whose existence has been treated as normal and whose has been treated as deviation. The machine sees patterns. Bodies live histories. The struggle over artificial intelligence, therefore, is not only a technical dispute over accuracy or fairness. It is a struggle over what will count as knowledge itself, and over whether the futures being assembled now will be capable of recognising the histories that their patterns were never designed to see.
Image Credit: Wikimedia Commons (Public Domain)
Notes
[1] Joy Buolamwini and Timnit Gebru, ‘Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification,’ Proceedings of Machine Learning Research 81 (2018): 77-91.
[2] Marc Andreessen, ‘The Techno-Optimist Manifesto,’ Andreessen Horowitz, 16 October 2023, https://a16z.com/the-techno-optimist-manifesto/
[3] Amanda du Preez, Gendered Bodies and New Technologies: Rethinking Embodiment in a Cyber-era (Newcastle: Cambridge Scholars Publishing, 2009).
[4] Maurice Merleau-Ponty, Phenomenology of Perception, trans. Donald A. Landes (London: Routledge, 2012 [1945]).
[5] Kimberlé Crenshaw, ‘Mapping the Margins: Intersectionality, Identity Politics, and Violence Against Women of Color,’ Stanford Law Review 43, no. 6 (1991): 1241-1299.
[6] Caroline Criado Perez, Invisible Women: Data Bias in a World Designed for Men (London: Chatto & Windus, 2019), 107.
[7] On the eugenic genealogy of statistical pattern recognition and its translation into machine vision, see Lila Lee-Morrison, Portraits of Automated Facial Recognition: On Machinic Ways of Seeing the Face (Bielefeld: transcript, 2019); Julia Ponzio, ‘Embodying Genre: From Galton’s Generic Faces to Peirce’s Embodied Ideas,’ Chinese Semiotic Studies 19, no. 3 (2023); and Wendy Hui Kyong Chun, Discriminating Data: Correlation, Neighborhoods, and the New Politics of Recognition (Cambridge, MA: MIT Press, 2021).
[8] Amoore, L. (2006). Biometric borders: Governing mobilities in the war on terror. Political Geography, 25(3), 336-351. The foundational paper describing the biometric border, including the role of systems like IDENT, a biometric database that stores and identifies electronic fingerprints on all foreign visitors, immigrants and asylum seekers, and ADIS, which stores travellers’ entry and exit data, along with APIS.
[9] Amoore, L. (2021). The deep border. Political Geography, advance online publication. https://doi.org/10.1016/j.polgeo.2021.102547 – the concept of ‘deep border’ is spatially reimagined as a set of always possible functions, features, and clusters as a ‘line of best fit’, and the neural-network sense of “depth” that resonates with state ambitions to reach into population-level attributes.
[10] Donna Haraway, ‘A Manifesto for Cyborgs: Science, Technology, and Socialist Feminism in the 1980s,’ Socialist Review 80 (1985): 65-108.
[11] Karen Barad, Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning (Durham: Duke University Press, 2006).
[12] Rosi Braidotti, ‘A Theoretical Framework for the Critical Posthumanities,’ Theory, Culture & Society 36, no. 6 (2019): 31-61; Gilles Deleuze and Félix Guattari, A Thousand Plateaus: Capitalism and Schizophrenia, trans. Brian Massumi (Minneapolis: University of Minnesota Press, 1987).
[13] Legacy Russell, Glitch Feminism: A Manifesto (London: Verso, 2020).
[14] Caroline Sinders, ‘Feminist Dataset,’ 2020, https://carolinesinders.com/wp-content/uploads/2020/05/Feminist-Data-Set-Final-Draft-2020-0526.pdf
[15] Kira Xonorika, ‘Do You Believe in Aliens?: Re-Indigenizing the Algorithmic Tropes of Intelligence,’ Momus, 2024, https://momus.ca/do-you-believe-in-aliens-re-indigenizing-the-algorithmic-tropes-of-intelligence/
References
Amoore, L. (2006). Biometric borders: Governing mobilities in the war on terror. Political Geography, 25(3), 336-351. https://doi.org/10.1016/j.polgeo.2006.02.001
Amoore, L. (2021). The deep border. Political Geography, 92, Article 102547. https://doi.org/10.1016/j.polgeo.2021.102547
Andreessen, Marc. ‘The Techno-Optimist Manifesto.’ Andreessen Horowitz, 16 October 2023. Available at: https://a16z.com/the-techno-optimist-manifesto/
Barad, Karen. Meeting the Universe Halfway: Quantum Physics and the Entanglement of Matter and Meaning. Durham: Duke University Press, 2006.
Braidotti, Rosi. ‘A Theoretical Framework for the Critical Posthumanities.’ Theory, Culture & Society 36, no. 6 (2019): 31-61.
Buolamwini, Joy, and Timnit Gebru. ‘Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification.’ Proceedings of Machine Learning Research 81 (2018): 77-91.
Butler, Judith. Gender Trouble: Feminism and the Subversion of Identity. New York: Routledge, 1990.
Chun, Wendy Hui Kyong. Discriminating Data: Correlation, Neighborhoods, and the New Politics of Recognition. Cambridge, MA: MIT Press, 2021.
Crenshaw, Kimberlé. ‘Mapping the Margins: Intersectionality, Identity Politics, and Violence Against Women of Color.’ Stanford Law Review 43, no. 6 (1991): 1241-1299.
Criado Perez, Caroline. Invisible Women: Data Bias in a World Designed for Men. London: Chatto & Windus, 2019.
Deleuze, Gilles, and Félix Guattari. A Thousand Plateaus: Capitalism and Schizophrenia. Translated by Brian Massumi. Minneapolis: University of Minnesota Press, 1987.
Descartes, René. Meditations on First Philosophy. Translated by John Cottingham. Cambridge: Cambridge University Press, 1996 [1641].
du Preez, Amanda. Gendered Bodies and New Technologies: Rethinking Embodiment in a Cyber-era. Newcastle: Cambridge Scholars Publishing, 2009.
Egbert, Simon, and Monique Mann. ‘Discrimination in Predictive Policing: The (Dangerous) Myth of Impartiality and the Need for STS Analysis.’ In Automating Crime Prevention, Surveillance, and Military Operations, edited by Aleš Završnik and Vasja Badalič, 25-46. Cham: Springer, 2021.
Foucault, Michel. The History of Sexuality, Volume 1: An Introduction. Translated by Robert Hurley. New York: Pantheon Books, 1978.
Haraway, Donna. ‘A Manifesto for Cyborgs: Science, Technology, and Socialist Feminism in the 1980s.’ Socialist Review80 (1985): 65-108.
Irigaray, Luce. Speculum of the Other Woman. Translated by Gillian C. Gill. Ithaca: Cornell University Press, 1985.
Lee-Morrison, Lila. Portraits of Automated Facial Recognition: On Machinic Ways of Seeing the Face. Bielefeld: transcript, 2019.
Merleau-Ponty, Maurice. Phenomenology of Perception. Translated by Donald A. Landes. London: Routledge, 2012 [1945].
Mulvey, Laura. ‘Visual Pleasure and Narrative Cinema.’ In Film Theory and Criticism: Introductory Readings, edited by Leo Braudy and Marshall Cohen, 833-844. New York: Oxford University Press, 1999.
Ponzio, Julia. ‘Embodying Genre: From Galton’s Generic Faces to Peirce’s Embodied Ideas.’ Chinese Semiotic Studies19, no. 3 (2023).
Russell, Legacy. Glitch Feminism: A Manifesto. London: Verso, 2020.
Sinders, Caroline. ‘Feminist Dataset.’ 2020. Available at: https://carolinesinders.com/wp-content/uploads/2020/05/Feminist-Data-Set-Final-Draft-2020-0526.pdf
Xonorika, Kira. ‘Do You Believe in Aliens?: Re-Indigenizing the Algorithmic Tropes of Intelligence.- Momus, 2024. Available at: https://momus.ca/do-you-believe-in-aliens-re-indigenizing-the-algorithmic-tropes-of-intelligence/
