AI aesthetics are not neutral: a new framework maps how power shapes what we call beautiful

A new conceptual analysis argues that aesthetic values in AI-mediated visual culture are not universal or psychologically neutral, but socially constructed and shaped by power relations. Published in Frontiers in Communication by Mengyao Xia and Melanie Sarantou of Kyushu University’s Graduate School of Design, the paper takes an existing psychological model of aesthetic values and rebuilds it through the lens of critical social theory. The result is a power-sensitive framework that identifies six domains in which aesthetic life is structured by hierarchy, from the distribution of aesthetic needs to the politics of judgment, and applies each domain to the algorithmic systems that now produce, circulate and evaluate visual content.

The starting point is Ye Zezhou’s 2023 aesthetic value framework, a three-level model that organizes aesthetic experience as a psychological pathway: aesthetic needs drive engagement with objects, mediated by taste, cognition and judgment, culminating in an aesthetic response such as pleasure. The model draws on empirical research on Chinese aesthetic psychology, including studies measuring aesthetic taste and social perceptions of physical beauty across regions, professions and age groups, and surveys of public aesthetic values spanning individual, interpersonal, lifestyle and material orientations. Ye’s framework renders previously diffuse concepts measurable and comparable, distinguishing between definitions that require all specified conditions to hold and those based on family resemblance, where partial overlap suffices.

The authors identify three interconnected limitations in this architecture. The first is psychological reductionism: the model presupposes a unified aesthetic subject whose inner processes can be mapped as general cognitive mechanisms, without asking who is recognized as a legitimate bearer of aesthetic needs or whose judgments are institutionally affirmed. The second is implicit universalism: although grounded in Chinese empirical research, the framework presents aesthetic needs, cognition and judgment as general psychological structures rather than historically and culturally situated practices, risking a false precision that stabilizes culturally specific norms as universal parameters. The third, and most fundamental, is silence on power: the model treats each element as a relatively neutral cognitive mechanism, describing aesthetic variation without explaining it.

To fill this gap, the researchers drew on Arnold Berleant’s work on aesthetic rights and negative aesthetics, and on Pierre Bourdieu’s theory of cultural capital. Berleant’s account of negative aesthetics catalogues aesthetic deprivation, offense and humiliation in contemporary urban life, arguing that acknowledging deprivation implies recognizing aesthetic rights as innate human rights. The authors distinguish three separate claims here: a general capacity for aesthetic experience, broadly shared; the specific contents and standards through which aesthetic value is defined, which vary across groups; and the resources needed to realize that capacity, which are unequally distributed. Bourdieu’s contribution is the demonstration that judgments about beauty are never purely subjective but reflect social position, education and accumulated cultural capital, transmitted through habitus, the internalized system of perceptions and dispositions formed during socialization.

Methodologically, the paper is a critical conceptual synthesis rather than an empirical study. It proceeds in three phases: analytical deconstruction of each component of Ye’s framework to expose implicit assumptions about power and agency; critical reconstruction, which introduces power mechanisms into each component; and internal theoretical validation, assessing logical consistency, theoretical coverage and positioning against competing perspectives such as universalist aesthetic theory. The authors are explicit that this validation establishes argumentative adequacy prior to empirical testing, not external verification. Illustrative examples drawn from historical, digital and subcultural contexts are used to demonstrate the framework’s scope, with particular attention to aesthetic resistance.

The reconstructed framework yields six domains. Aesthetic needs become sites of aesthetic stratification, the unequal distribution of aesthetic opportunity, resources and recognition, informed by a multi-level model spanning neurocognitive, embodied-emotional, sociopolitical and ecological levels. Aesthetic relationships are revealed as mediated by hegemonic structures, including educational systems, media industries and policy frameworks, that determine which objects become available for contemplation. Aesthetic responses are socially differentiated through unequal access to cultural codes and interpretive frameworks. Taste is reframed as a political economy, transmitted intergenerationally through families and legitimized by institutions. Aesthetic cognition is understood as ideologically mediated rather than neutrally perceptual. And aesthetic judgment is exposed as a political practice in which credentialed critics, prestigious venues and dominant groups systematically naturalize their standards as objective measures of quality.

Crucially, the framework also documents resistance. The resurgence of yamikawaii in global youth culture exemplifies emotional resistance through aesthetic ambiguity, integrating breakdown makeup, trauma imagery and declarations of psychological distress to subvert the normative positivity of mainstream kawaii aesthetics. Similarly, the rise of ugly-cute aesthetics on platforms such as TikTok and Instagram since 2023 offers satirical resistance to algorithmically enforced beauty standards, with creators deliberately deploying awkward angles, exaggerated features and chaotic color schemes to disrupt the polished, hyper-filtered norms that recommendation systems reward. These practices do not merely reject dominant codes; they appropriate and distort them, using the tools of the dominant aesthetic regime against its own logic of optimization.

Applied to AI-generated media, the six domains map onto distinct stages of the sociotechnical pipeline. Training data overwhelmingly represent the visual cultures of economically dominant societies; research cited in the paper found that text-to-image models systematically rendered South Asian subjects through a Western, stereotypical gaze. Dominant norms are then encoded directly into model weights and architectures, replacing human cultural gatekeepers with opaque proprietary systems. Digital visual literacy stratifies audiences’ capacity to detect synthetic imagery, with lower-literacy groups relying more on surface-level aesthetic cues as credibility heuristics, a finding with consequences for political epistemology when disinformation images are judged by affective plausibility rather than critical scrutiny. Platform recommendation algorithms produce filter bubbles of aesthetic experience, forming what the authors call an algorithmic habitus, while the sheer velocity of generative content institutionalizes fast aesthetics, modes of perception most susceptible to ideological influence. Finally, content moderation and governance are concentrated among a small number of technology corporations whose cultural composition is far from representative of global visual cultures.

The authors are careful about the limits of their claims. The framework is a theorized foundation, not an empirically verified model; some propositions are supported by independent research while others are theoretically grounded inferences awaiting verification. Its theoretical architecture remains substantially rooted in European and North American traditions, and its applicability to Chinese, Japanese, South Asian and African aesthetic systems has not been systematically examined. The researchers also flag the need for reflexivity about their own social positions. They call for operationalizing constructs such as aesthetic stratification and algorithmic habitus formation using established measurement scales, conducting training data audits, platform behavioral tracking and audience experiments, and pursuing participatory research with communities whose aesthetic practices have been marginalized by algorithmic systems.

The paper’s normative horizon is aesthetic democracy, understood not as an achieved state but as an ongoing redistributive practice. Drawing on Rancière’s distribution of the sensible and on participatory workshop research showing that technical access to generative tools does not automatically translate into democratized aesthetic participation, the authors specify three interlocking conditions: redistribution of material access to computational resources and training data representation; recognition of marginalized aesthetic vocabularies as legitimate rather than filtered out by moderation systems; and representation through participatory governance mechanisms giving affected communities meaningful voice. As a benchmark for evaluating AI governance frameworks, the framework offers conceptual vocabulary for identifying, within everyday aesthetic life, the sites where redistribution, recognition and representation remain most urgently required.

Subject of Research: A power-sensitive reconceptualization of aesthetic values in AI-mediated visual communication

Article Title: Reconfiguring aesthetic values: a power-sensitive framework for AI-mediated visual communication

Article References: Xia, M., & Sarantou, M. (2026). Reconfiguring aesthetic values: a power-sensitive framework for AI-mediated visual communication. Frontiers in Communication, 11, 1893436. https://doi.org/10.3389/fcomm.2026.1893436

Image Credits: AI Generated

DOI: 10.3389/fcomm.2026.1893436

Keywords: aesthetic values, AI-generated media, visual communication, cultural capital, algorithmic bias, critical theory, aesthetic democracy, power relations, content moderation, digital visual literacy, Bourdieu, negative aesthetics