AESTHETIC NEGATIVITY VS. ECONOMIC NATURALISM: A CLOSE READING OF AN INDIAN AI-ART PRACTICE

INTRODUCTION

Harshit Agrawal’s AI-based art invites a question that contemporary digital culture often resists: whether a medium engineered for novelty and platform-friendly consumption can sustain genuine aesthetic critique. His 2021 exhibition, EXO-stential: AI Musings on the Posthuman, was promoted and circulated (notably via the gallery’s own framing) through a language of institutional ‘firsts’, a framing that can obscure the deeper stakes of his practice Emami Art. (2021), Baas (2024). This essay reads Agrawal’s work as a case study in a fundamental structural tension: how artworks claiming critical intent can resist being absorbed into the smooth, affirmative aesthetic logic they theoretically oppose Han (2017), Gawronski (2021). The question is urgent because it touches on a paradox where critique is present in declared concept, yet the aesthetic encounter remains consumable and fully compatible with commodity logic Gawronski (2021).

Agrawal’s practice operates at the intersection of three overlapping pressures:

First, there is the mediation of AI art by platforms and speculative markets, which tend to reward novelty and visibility Gawronski (2021). Second, his work is shaped by a global AI infrastructure that predominantly encodes Western visual canons and training datasets Crawford and Paglen (2021). Finally, Agrawal must navigate the historical marginalization of digital and new-media work within Indian art institutions, where traditional, tangible objects have long been favored Barua et al. (2024), Bishop (2012).

His decision to foreground Indian visual archives such as Kathakali masks and Theyyam ritual forms represents a deliberate intervention into this triple constraint Agrawal (2019); Barua et al. (2024). Yet, this local specificity coexists uneasily with global platform logic, creating a productive contradiction between critical intent and market visibility Gawronski (2021).

To make these stakes legible, this analysis triangulates four theoretical frameworks: Byung-Chul Han’s critique of digital “smoothness” Han (2017), Adorno’s negative dialectics Anadol (2021), Alexander Gawronski’s account of “economic naturalism” Gawronski (2021), and Baudrillard’s theory of the hyperreal Baudrillard (1994). Through these lenses, I examine three anchor works, The Anatomy Lesson of Dr. Algorithm (2018), Masked Reality (2019), and Strange Genders (2020), to ask where Agrawal’s practice sustains friction and where it risks succumbing to the very logic it theoretically resists Agrawal (2018), Agrawal (2019) NVIDIA. (n.d.).

The mediation problem

A recurring difficulty in AI-based practice is that the work can change meaning when it shifts from embodied encounter to documentation. In installation, scale, duration, and spatial negotiation can cultivate opacity and delay; online, the same work is compressed into an “Instagramable artifact,” where the caption often carries the argument while the image circulates as novelty. In what follows, each case study therefore includes a short mediation paragraph that distinguishes (a) what the work demands as installation and (b) what it becomes when flattened into shareable media Han (2017), Gawronski (2021).

Genealogy: Computational Art, Coldness, and the Question of Agency

Understanding Agrawal’s contemporary intervention requires a brief, pointed return to computational art’s longer history, not to narrate progress, but to isolate three recurring problems that will structure the readings that follow: (a) the distribution of agency between artist and system, (b) the question of authorship as a social fiction sustained by institutions and markets, and (c) the historically persistent charge of “coldness” as an affective diagnostic that attaches itself to algorithmic procedure Mohr (n.d.).

Manfred Mohr’s computer-generated works in the late 1960s and 1970s were received through precisely this affective tension. A response panel from his 1971 Paris exhibition records reactions from viewers that oscillate between fascination with method and suspicion toward the image’s sensuous capacity; the repeated word “FROID” condenses an early anxiety that algorithmic images may look resolved while feeling emotionally distant (Mohr, n.d.). The point here is not to reaffirm the cliché that computation lacks feeling, but to note how “coldness” historically functions as a shorthand for a perceived gap between procedural production and embodied encounter, a gap that will return, differently, when we consider Agrawal’s painterly medical abstractions.

Harold Cohen’s AARON program offers a clarifying precedent for thinking about authorship under computation. Cohen did not design AARON to replace his artistic thinking; he built it to externalize and test how an artist thinks through image-making, insisting on the system as tool rather than autonomous author Cohen (1995). This insistence is not nostalgia for the “hand.” It is a methodological correction to techno-mythologies of machine autonomy that become especially persuasive in market and media contexts.

Within India, the genealogy of digital and new-media art since the 1990s provides a second hinge. Practitioners have worked with video, networks, interactive installations, and hybrid technocultural forms that interrogate how technology reshapes subjectivity and power relations, while also negotiating infrastructural divides Barua et al. (2024). Yet digital art has often remained unevenly institutionalized within mainstream infrastructures of exhibition, collection, and criticism Barua et al. (2024), Bishop (2012). This unevenness is constitutive for Agrawal: his practice becomes legible through novelty claims (circulated as ‘firsts’), even as the deeper question is how critique survives when technological newness is itself a form of value Emami Art. (2021), Gawronski (2021).

Taken together, these genealogical points authorize a specific reading strategy for Agrawal: rather than treating AI as an autonomous creator or as a neutral tool, the case studies below track how agency is distributed across datasets, systems, curatorial framings, and viewer expectations; how authorship is narrated (and sold) as novelty; and how the recurring charge of computational “coldness” can be redeployed as an affective diagnostic for moments where the work’s surface coheres too smoothly to sustain negativity Mohr (n.d.), Cohen (1995), Han (2017).

Theoretical Framework: Negativity, Smoothness, and the Stakes of Form

Han’s diagnosis in Saving Beauty establishes the contemporary problem with precision. He argues that modern digital media and interfaces are systematically optimized to be agreeable and shareable: they “do not injure,” and they tend to solicit rapid affirmation rather than sustained reflection Han (2017). Smoothness, in Han’s terms, is not merely an aesthetic quality; it is a cultural logic that disciplines perception through frictionless pleasure Han (2017). When applied to AI art specifically, the problem becomes acute: machine-learning systems are valued precisely for producing outputs that look resolved, polished, and immediately coherent. If the primary effect of AI-generated images is smoothness and visual pleasure, where does negativity reside? Where is the work’s capacity to refuse, to delay, to create discomfort?

Adorno sharpens these stakes through the lens of negative dialectics. He argues that art’s critical force does not reside in political content or explicit messaging; it inheres in form’s refusal of the world’s coerced equivalences Adorno (1973). A work that resolves its contradictions too quickly, that offers tidy moral closure or can be consumed as affirmative “experience,” risks reinforcing the very order it claims to critique Adorno (1973). For AI art, this becomes a stringent demand: machine-learning images are often evaluated through the vocabulary of novelty (“what the model can do”) or process (“how it was trained”), rather than through the difficult question of what sensuous, formal, embodied encounter the work demands from the viewer. In short, critique must be legible in aesthetic behavior, through delay, discomfort, opacity, refusal, and formal contradiction, not merely stated in accompanying text Adorno (1973).

Gawronski’s intervention into neoliberal aesthetics explains why maintaining such behavior proves structurally difficult. He argues that neoliberalism naturalizes itself as ambient condition: economic logic becomes “everywhere and nowhere,” invisible precisely because it is pervasive Gawronski (2021). Contemporary art circulates within systems that reward visibility, novelty, and brandable difference, and even gestures of critical negation are readily recoded as market value Gawronski (2021). This is not a moral accusation against individual artists; it is a structural diagnosis. The problem is what Gawronski calls self-annulling critique: works that question their own commodity status while functioning perfectly as commodities, whose critical visibility becomes the very thing that makes them marketable Gawronski (2021). AI art, entangled with platforms, hype cycles, and speculative economies, is particularly exposed to this dynamic.

Baudrillard provides a final theoretical layer by clarifying the ontological condition of AI-generated imagery. When a generative model produces an image by sampling and synthesizing statistical patterns learned from massive datasets, the output has no singular referent in lived reality; it is a modeled “real without origin” Baudrillard (1994). In Baudrillard’s terms, such images occupy a hyperreal condition: they are often more consistent, more aesthetically resolved, more consumable than the world itself; the image can replace the world as the primary site of experience Baudrillard (1994). In this context, the critical question shifts: it is not whether AI images are authentic or fraudulent, but whether the artwork can construct an aesthetic structure capable of resisting the hyperreal’s seductions, or whether it becomes complicit in intensifying simulation.

Case Studies: Negativity, Spectacle, and Formal Ambivalence

The Anatomy Lesson of Dr. Algorithm (2018): Hyperreality and the Problem of Beauty

Agrawal’s The Anatomy Lesson of Dr. Algorithm explicitly invokes Baudrillard’s theoretical terrain through its reference to Rembrandt’s iconic anatomy lesson. Where Rembrandt depicted an empirical cadaver, a singular, individuated body subject to empirical observation, Agrawal substitutes the AI’s vision: a synthetic reconstruction trained on medical videos of surgeries and dissections, producing images that have no grounding in any actual body Agrawal (2018). The resulting works are formally ambiguous in ways that matter for aesthetic analysis: they present as painterly abstraction, with floating color fields and organic fragments that hover between recognition and entropy, between figuration and abstraction Agrawal (2018).

This is also the place where the older accusation of computational “coldness” becomes newly legible. The images can appear sensuously lush while remaining affectively distant, an effect that echoes the historical “FROID” response not as a verdict against the work, but as a clue: the more the algorithm resolves the body into coherent visual pleasure, the more the encounter risks losing the resistant, bodily unease that an “anatomy lesson” historically stages Mohr (n.d.), Han (2017).

This ambiguity can operate in two contradictory registers. First, it can estrange the clinical gaze that medical imagery typically stabilizes. Anatomical clarity is withheld. The body becomes uncanny again, no longer available as instrumentally legible information, and the viewer confronts an unfamiliar sensuousness at the site where “knowledge” is usually demanded Baudrillard (1994). Second, the same ambiguity can be aestheticized into pleasure: the work’s painterly density can translate what should disturb into what merely delights, and in that conversion the negative force risks being smoothed into surface Han (2017). The formal question is whether this ambivalence is enacted through the work’s structure or merely stated as a conceptual premise.

Mediation and circulation

As installation, the work relies on duration and proximity: the viewer needs time for the image to hover between medical legibility and painterly seduction, so that unease can accumulate rather than resolve. As documentation, however, it tends to circulate as a series of seductive stills, “AI dreams of anatomy”, where the hyperreal surface reads as visual pleasure and the critical claim migrates into explanatory text, producing exactly the smoothness that threatens negativity Han (2017), Gawronski et al. (2021).

Masked Reality (2019): Cultural Archives, Spectacle, and Temporal Friction

Masked Reality represents Agrawal’s most analytically instructive intervention because it makes cultural material do more than ornamental work. The piece uses facial-recognition and transformation algorithms to overlay viewers’ faces with digital masks drawn from Indian theatrical and ritual traditions, explicitly including Kathakali and Theyyam, so that the self appears as composite rather than essence Agrawal (2019). The interactive effect is deliberately uncanny and playful: the viewer sees themselves in live video, then suddenly their features are replaced by a shifting theatrical mask Agrawal (2019). Identity is staged as assemblage, an archive of inherited forms.

Yet the same strategic move carries risk of spectacularization. Interactive works can produce instantaneous delight, the quick “Like,” and that interface pleasure can cancel duration, the condition under which negativity might develop Han (2017). Here temporal pacing becomes crucial. If the mask-overlay persists long enough for the viewer to experience discomfort at the instability of the self, the work can exceed filter logic; if it collapses into instantaneous novelty, critique is neutralized into circulation Gawronski (2021).

Mediation and circulation

In embodied installation, the work’s critical force depends on pacing and social situation: the viewer is held in front of a system that mirrors them back as archive, and the time spent negotiating the shifting mask can produce genuine discomfort. Online, the same operation collapses into a short clip or selfie-like before/after transformation, the mask becomes a filter, the politics becomes caption, and the work’s negativity is structurally tempted to vanish into shareability Han (2017), Gawronski (2021).

Strange Genders (2020): Classification as Visible Violence

Strange Genders, developed in collaboration with the artist collective 64/1, offers the clearest model of aesthetic negativity within Agrawal’s practice because it turns AI’s classificatory logic into formal material rather than leaving it as invisible infrastructure. The project trains a generative model on naïve drawings labeled “male” and “female,” then applies a classifier that assigns probability scores to outputs that are deliberately ambiguous, images that refuse easy gender categorization while the system still insists on decision NVIDIA. (n.d.). A form may be visually indeterminate, yet it is pinned to a number, “female 71% / male 29%”, and the work holds that mismatch in view rather than resolving it.

That methodological preoccupation with rendering the machine’s internal decision-making visible also dictates the work’s spatial logic. The generated drawings are organized into a loosely concentric, circular arrangement that reads less like a neutral display and more like a diagram: a visual map of the model’s latent space, positioning ambiguous figures within a taxonomy that makes the violence of classification perceptible. Here, the “content” is not gender as theme; it is classification as pressure.

The formal strategy is rigorously Adornoian in that it stages contradiction instead of reconciling it. The work does not offer a consoling message about fluidity or a triumphant transcendence of binaries. It keeps the conflict between visual ambiguity and categorical demand in perpetual play: the drawings resist masculine/feminine legibility, while the numerical scores keep forcing a binary verdict Adorno (1973), NVIDIA. (n.d.). Negativity emerges not from what the work says, but from what it refuses to let the viewer do, namely, settle the opposition into harmony.

Formally, Strange Genders operates in a post-conceptual register. It privileges procedure, indexing, and systemic display over pictorial seduction. The aesthetic effect is therefore primarily cognitive rather than sensuous, but cognition here is not smooth or reassuring. It produces intellectual friction: the viewer has to sit with the awkwardness of percentages attached to images that do not fit the boxes those percentages claim to describe. Difficulty, rather than pleasure, becomes the site of aesthetic inquiry.

Mediation and circulation

The mediation problem is relevant here, but for a different reason than in the other case studies. In installation, the work’s meaning is spatial and comparative: you read it by moving, scanning across many near-variations, and sensing how an apparently fluid field is organized into a systemic map. Online, the piece often collapses into a single “provocative” example, one image with its score, or a cropped fragment of the circular arrangement, which converts a sustained experience of taxonomy into an easily shareable illustration of bias. That compression reduces the work’s formal pressure (repetition, placement, slow comparison) and makes the numbers feel like captions rather than an aesthetic ordeal Gawronski (2021).

Global Context: The International AI Art Field and Agrawal’s Position

Situating Agrawal within international AI art reveals both the specificity of his intervention and the field’s dominant temptations. Refik Anadol’s data-driven installations pursue computational spectacle and immersive scale through machine-learning processes trained on vast archives Anadol (2021). Mario Klingemann’s work has been framed through market and critical discourse as a case where neural-network portraiture becomes uncanny, unstable, and continuously generative, including within the auction context Rea (2019), Whiddington (2019). Anna Ridler’s practice insists on making visible dataset labor and the politics of training data, often linking machine learning to histories of speculation and value Ridler (n.d.). The collective Obvious became emblematic of how quickly AI images can be absorbed into major-auction narratives and speculative economics through the 2018 Christie’s sale of Edmond de Belamy Christie (2018). These diverse positions share a common structural tension: they either surrender to spectacle and novelty or attempt to expose the internal mechanics of AI systems.

Against this landscape, Agrawal’s contribution is distinctive precisely because it does not primarily pursue spectacle. Instead, his strongest works foreground mediation as a cultural and political problem: how identities and archives are recomposed through algorithmic systems; how classification produces legibility through force; how the hyperreal becomes an environment rather than a theme Agrawal (2018), Agrawal (2019) Baudrillard (1994). Yet global art-market logic remains inescapable, and AI art is often packaged as cutting-edge commodity. Visibility can neutralize critique Gawronski (2021).

The Central Paradox: Aesthetic Negativity in an Age of Smooth Critique

The deepest insight into Agrawal’s practice emerges from holding together what appear to be contradictions. The work succeeds most forcefully when it makes AI’s conditions of production aesthetic, distributed agency becomes visible, classification becomes pressure, and archival identity becomes form Agrawal (2019), NVIDIA. (n.d.). Yet dilution occurs when conceptual vehicles, posthuman discourse, cultural specificity, and technological novelty, carry the work so completely that the encounter itself becomes smooth, resolved, and quickly consumable Han (2017). The critical question facing Agrawal’s practice is therefore structural, not moral: can the work produce aesthetic conditions that remain incompatible with frictionless circulation, or does it become optimized for the very regime it critiques? Gawronski (2021).

Conclusion: Negativity as Ongoing and Unresolved

Harshit Agrawal is best understood as a practitioner who embodies a productive contradiction. His work expands the formal vocabulary of Indian contemporary art by integrating machine learning and compelling global AI infrastructures to engage with local visual archives Agrawal (2019) Barua et al. (2024). At the same time, his practice operates within a global techno-cultural atmosphere that values ‘firsts’ as promotional and curatorial currency, polished interfaces, and platform circulation Emami Art. (2021), Han (2017). The work is most effective when it maintains contradiction, opacity, and friction in opposition to the smooth digital surface Adorno (1973), Han (2017). Ultimately, this lack of resolution is not a shortcoming of Agrawal’s practice but a structural necessity for making critical art under conditions where critique is easily converted into market value Gawronski (2021).

Implications beyond Agrawal

The method developed here, reading AI works through negativity, smoothness, and hyperreality while tracking the mediation problem, can be extended beyond Agrawal without turning every digital practice into a morality tale about technology. It offers a way to evaluate whether a work’s critique is enacted in form (through delay, awkwardness, opacity, repetition) or outsourced to explanatory discourse that travels with the image as caption Adorno (1973), Han (2017). Applied to other subcontinental digital practices, the approach also keeps institutional “firsts” in their place: as conditions of visibility rather than proofs of aesthetic or critical force Emami Art. (2021); Barua et al. (2024). It encourages a sharper attention to how local archives enter global infrastructures, and whether “cultural specificity” functions as friction or as marketable content Crawford and Paglen (2021); Gawronski (2021). It also insists that documentation is not neutral: online circulation can compress duration into novelty and convert complex works into illustrative artifacts, so scholars must read installation demands and platform afterlives as part of the same aesthetic object. Finally, the framework makes room for comparative work across India’s uneven digital art infrastructures, where critical practice may emerge less from technological sophistication than from the careful construction of viewing conditions that refuse smooth consumption Barua et al. (2024), Gawronski (2021).