When AI Decides Who Counts as Human: The New Crisis of Recognition

Artificial intelligence systems no longer simply identify people; in many institutional settings they now help decide who is seen, trusted, and included. A new theoretical paper published in Humanities and Social Sciences Communications argues that this shift has created something more troubling than the familiar problem of algorithmic bias. According to authors Min Tang and Jinqi Yan of Southeast University in Nanjing, China, algorithmic systems increasingly participate in practices of social recognition, and when those systems fail, the failures can harden into forms of structural disrespect. The paper, published open access on 15 October 2026, reframes debates about AI fairness through the lens of a philosophical tradition that stretches back to the nineteenth-century German philosopher G. W. F. Hegel.

The central claim is deceptively simple: recognition, the social process through which individuals are acknowledged as full persons, is no longer mediated only by other humans. Classification systems with institutional authority now shape how people are treated at borders, in hiring pipelines, in credit markets, and by content platforms. When such classifications misfire, the authors contend, the harm is not merely an error to be corrected in a model. It becomes a form of disrespect embedded in the structure of institutions themselves, because the misclassified person is not just mishandled by a machine but diminished in the eyes of the organizations and publics that rely on the machine’s judgments.

To give this phenomenon analytical precision, Tang and Yan introduce the concept of algorithmic disrespect, which they argue manifests across three distinct dimensions. The first is embodied disrespect, which occurs when certain bodies are excluded from visibility altogether. Facial recognition systems that perform poorly on darker skin tones, gender-diverse faces, or older faces are the canonical example: a person whose face a camera cannot reliably read is, in a practical sense, rendered less legible to the institutions that depend on those cameras. The failure is not abstract. It can mean being flagged for additional screening, being denied entry, or simply not being counted by systems that allocate attention and resources.

The second dimension is agential disrespect, which arises when predictive models preempt human action. Here the concern is not that a system fails to see someone but that it acts on a forecast of that person’s behavior before the person has done anything at all. Risk scores, recommendation filters, and automated eligibility decisions all operate on this logic. A model that predicts an applicant will default on a loan, or that a claimant is likely to commit fraud, treats a possible future as a present fact. The person’s capacity to act, to explain, and to contest the prediction is bypassed. In the vocabulary of recognition theory, the individual is no longer addressed as an agent capable of giving and asking for reasons, but as an object of statistical management.

The third dimension is symbolic disrespect, which concerns how human value is indexed through computational relevance. Ranking and recommendation systems determine whose speech circulates, whose work is discovered, and whose contributions register at all. Because visibility on digital platforms is allocated by relevance metrics, the authors argue that social esteem has become entangled with optimization objectives that were never designed to honor human worth. A creator buried by an algorithm change, a scholar whose work is under-ranked by search systems, or a community whose language is poorly supported by a large language model all experience a version of the same injury: their symbolic standing is mediated, and can be silently devalued, by infrastructures they do not control.

Underneath these three mechanisms, the paper diagnoses what it calls a crisis in the social grammar of recognition. The phrase borrows from the recognition theory developed by the contemporary German philosopher Axel Honneth, who built on Hegel’s insight that persons become selves through patterns of mutual acknowledgment: legal recognition that grants rights, social recognition that grants esteem, and intimate recognition that grounds self-confidence. Tang and Yan argue that data is replacing dialogue in each of these registers. Where recognition once required a relationship between subjects who could address and respond to one another, it is increasingly issued by systems that neither see nor hear in any meaningful sense. Visibility, meanwhile, has become a double-edged instrument, functioning simultaneously as a condition of inclusion and as a tool of domination, since those who control the metrics of visibility also control who can appear.

This reframing has consequences for how the technical community thinks about fairness. The dominant approach in machine learning research treats bias as a property of models to be measured and minimized, typically through statistical parity constraints, calibration targets, or error-rate balancing across groups. The paper does not dismiss these interventions, but it insists that they are insufficient, because the harms it identifies are not reducible to unequal error rates. A system can be statistically well-calibrated and still preempt agency, still render certain bodies invisible, and still allocate symbolic worth in ways that people cannot contest. Fairness metrics, in other words, operate at the level of model outputs, while the crisis of recognition operates at the level of the institutional conditions under which those outputs acquire authority.

The authors therefore propose a relational ethics of recognition grounded in three commitments: justification, accountability, and institutional responsiveness. Justification requires that the criteria by which algorithmic systems classify and rank people be publicly defensible, not merely technically optimized. Accountability requires that there be identifiable parties who answer for the judgments these systems issue, rather than a diffuse attribution of responsibility to the algorithm itself. Institutional responsiveness requires that organizations build genuine channels through which those affected can contest classifications and have those contests change outcomes. The point of this framework is to shift attention from model-level corrections to the infrastructural conditions under which recognition is distributed, asking not only whether a model is accurate but whether the arrangement of institutions around it treats people as persons.

On this account, recognition is repositioned as both a site of harm and a potential locus of democratic struggle in algorithmic society. The paper’s implication is that contests over AI systems, from protests against facial surveillance to campaigns by gig workers against automated management, are not merely consumer complaints or technical disputes. They are struggles over the terms of personhood in digital institutions, analogous to earlier social struggles over legal standing and social esteem. Framing them this way, the authors suggest, clarifies what is actually at stake when people demand to be seen correctly by machines: not convenience, but confirmation that they exist as full members of a shared social world.

The paper is a work of philosophy and social theory rather than an empirical study, and it contains no experiments or human-subjects research; the authors declare no competing interests, and they note that a large language model was used only for translation and language editing, with all theoretical work attributed to the human authors. Its contribution lies in the conceptual architecture it offers: a vocabulary for harms that fairness engineering struggles to name, and a reminder that the deepest questions raised by AI are not about what machines can do, but about who gets to count as human in the systems that now organize social life. As algorithmic classification continues to expand into new institutional domains, the authors argue, the task ahead is to rebuild the conditions of recognition, dialogue, and justification on which that counting depends.

Subject of Research: Algorithmic disrespect and the crisis of social recognition in AI-mediated institutions

Article Title: Being recognized as human by AI: algorithmic disrespect and the crisis of recognition in the digital age

Article References: Tang, M., & Yan, J. (2026). Being recognized as human by AI: algorithmic disrespect and the crisis of recognition in the digital age. Humanities and Social Sciences Communications. https://doi.org/10.1057/s41599-026-09423-x

Image Credits: AI Generated

DOI: 10.1057/s41599-026-09423-x

Keywords: algorithmic disrespect, recognition theory, Hegel, Axel Honneth, AI fairness, facial recognition, predictive modeling, symbolic visibility, social recognition, digital ethics, algorithmic bias, democratic accountability