Work of art in the Age of Its AI Reproduction

Abstract

From a Benjaminian point of view, AI-generated art is distinct from both ‘traditional’ art and technologically enabled reproduction, for example, photography and film. Instead of mere mechanical representation of the world as it is presented to a device, AI-generated art involves identification and inventive representation of data patterns. This specific mode of data-based generation exceeds mere surface-level mimicry and enables deeper meaning, namely, an insight into the collective unconscious of the society. In this way, AI-generated art is never detached from society and the predominant social conditions while also reflecting the technology-induced transformations that today’s societies are undergoing. Thus, AI-generated art can be seen as capable of partly reversing the loss of auratic capacities that hand ensued with mechanical reproduction. Still, as a matter of continuity, AI-generated works enable the maximisation of exhibition value and capacity for audience enjoyment, rendering AI-generated art perfect for the age of increasing distraction.

To paraphrase the opening sentence of The Work of Art in the Age of Its Mechanical Reproducibility, when Benjamin undertook his critique of mechanical reproduction, this mode was still in its infancy. ¹ Of course, he was able to trace a history of reproducibility all the way from founding and stamping to photography and film. Nevertheless, the experience on which Benjamin could rely pales in comparison with the mushrooming of photo and video content first with digitalisation and, in particular, with the internet and social media. Moreover, we currently stand at the infancy of yet another revolution – that of AI-generated art, with the latter being bound to grow in both quality and diversity. The rapid emergence and mainstreamisation of this new kind of content and the new mode of its production (generative AI models) already necessitate a consideration of its art-ness, effect and function.

It is often asserted that due to the growing power of artificial intelligence, the development of a proper distinct form of machine creativity that goes beyond mere mimicry of human creative output is imminent. ² Of course, such claims are to be seen within the broader context of AI promotionalism, whereby machine capacities are presented as universal solutions to the world’s most pressing problems and capable of substituting humans in effectively all domains. Nevertheless, one should be wary of such grand and hubristic claims and not fall for what Morozov has already years ago dubbed technological solutionism. ³ Still, with the popularisation of text, image and video synthesis, debates around creativity and the ‘art-ness’ of AI-generated content have suddenly been rendered extremely acute. Simultaneously, the emergence of such technologies opens up a further question that had already manifested its significance in Benjamin’s time: that of technology’s impact on the idea, status and function of art and the relative standing of such new ways of (re)producing the world. In particular, this pertains to the changing value and societal significance of the work of art as such, with the move towards exhibition value seen here as a key factor. Hence, while not intended as a direct update or an explicit continuation of Benjamin’s work, this article presents a Benjamin-informed analysis of the status of AI-generated works (of art). Of course, there is a rich and ongoing debate over the adequacy of the term ‘art’ as applied to AI-generated content. While full engagement with this debate is beyond the scope of this article, it is assumed that its actual status notwithstanding, a subset of AI-generated content functions as art in a way comparable to that of, for example, photography in Benjamin’s time.

This article opens with a discussion of the novelty and data-based embeddedness of AI-generated art as well as its relationship to human creativity. The status of AI-generated art is located in-between reproduction and inventiveness. Next, attention shifts towards the position of AI-generated art vis-à-vis what Benjamin calls the ‘aura’ of the work of art, identifying the ways in which AI-generated art is situated in and reflective of historical and cultural circumstances. The discussion subsequently shifts towards Benjamin’s idea of the growing centrality of exhibition value of art: as AI is generally at its best when optimising tasks, it is shown how the optimisation of exhibition value is among the key premises of AI-generated art. The final section, then, deals with matters of authenticity and human relationship with AI-generated works. Notably, far from mere reception and perception, such relationship is also seen to entail a democratisation of the creative process. Ultimately, AI-generated works are understood to have the capacity of making visible the collective unconscious of today’s societies – the rich layers of data, artificial agents and humans themselves – in ways that are both adjacent to and dissimilar from the revelatory capacity that Benjamin had ascribed to film.

The complex relationship between human and AI creativity

It is next to impossible to discuss completely new ways of art creation without simultaneously rethinking the underlying conditions for the origination and enjoyment of artistic works. Indeed, Benjamin raises a crucial point by observing that discussions over the art-ness of, first, photography and, subsequently, film had missed something from the very beginning: an account of how the technologies of mechanical reproduction have ‘transformed the entire nature of art’ altogether. ⁴ Hence, instead of stretching old concepts and ideas originally developed for human-created works of art to mechanical capture and reproduction of the world, there was a need to develop new concepts – which Benjamin himself did with great insight. Crucially, the same need for a leap of conceptual thought also applies to works produced by AI. Like photography and film, AI art is materialised by fixing a snapshot onto a medium, either analogue or, more recently, digital. However, unlike these two, AI-generated content does not affix that which is presented in front of a device in a material form but, instead, a snapshot of data patterns that it has identified during the training stage, that is, when ingesting and analysing sets of data on which the future performance of the AI model will be based. Like human-created art, AI-generated works represent a synthesis of world-knowledge and invention; however, unlike human artists, AI-generators acquire world-knowledge as sets of data and, therefore, their inventiveness is a matter of identifying novel ways of combining and extrapolating from this data.

Indeed, extremely large sets of data scraped from across the internet are at the very heart of generative – and, indeed, any – AI; the specificity of generative AI, though, lies in its ability to go beyond mere recognition of patterns in such data and prediction of future occurrences based on such patterns and move towards more complex tasks, such as generating new content. ⁵ This is not to say that pattern recognition and prediction are superseded – instead, they remain central because generative AI produces content by way of putting data-derived predictions to a medium. Such predictions could pertain to the distribution of pixels, sequences of words or sounds etc. However, there is more to such prediction than mere imitation because the content generated tends to be qualitatively new and unlike the examples in the training data because the contents of the training data set are restructured and not merely rearticulated. ⁶ Therein, it could be argues, lies the modicum of creativity in generative AI, without prejudice to the ongoing debates as to the actual agency (human or artificial) behind the processes of content generation.

Notably, as Zeilinger asserts, AI ‘has the potential to reshape the aesthetic, cultural, and socio-economic valences of the concept of creativity’, thus simultaneously destabilising the figure of the author. ⁷ Indeed, authorship in the traditional sense has always been tied with the idea of a human person behind the work. In case of AI generation, meanwhile, even if there is a human person involved (tinkering with training data and parameters or sometimes doing as little as providing a prompt), the end result is ultimately determined by AI. After all, as Turner notes, the core premise of AI is an autonomous ‘ability of a non-natural entity to make choices in an evaluative process’. ⁸ It is often assumed that once AI has been trained on the sample material and has learned the core principles of artwork generation, it can be held to be sufficiently creative to produce new art. ⁹ In a way, this is not dissimilar from the way humans learn from their mentors, peers, great masters and the broad history of art. ¹⁰ Nevertheless, AI-generated content still is a work of art of a completely novel kind. The task is, therefore, to determine the characteristics of this new art form.

AI-generated content is always embedded in the sense that such tools learn from data samples, which might be a specific set of previous works or, much more broadly, scraped contents of the Web. Arriagada, for example, underscores the matter even further by stressing that AI-generated art ‘is fundamentally based on Big Data, which is the most social thing we have’. ¹¹ The sociality of Big Data must, however, be understood in a very specific sense. Instead of offering universal transparency and unbounded predictive capacities, as stipulated not only in progressivist accounts of digital technology but also in critical accounts of the ever-growing power of technology companies, ¹² big data should be seen as performing an intertwining function. In other words, Big Data are constantly multiplied as endless digital decorporealisations of the world while simultaneously challenging the traditional Western understanding of human primacy ¹³ and, instead, moving towards an understanding of life as ‘an ongoing composition in which humans and non-humans participate’. ¹⁴ In this way, everyday realities acquire a ‘more-than-human’ character, one marked by assemblages and interembodiments of human, digital (data and algorithms/AI) and physical summands. ¹⁵ Under such conditions, as Braidotti stresses, presence is best defined in terms of affective capacity – one that knows no boundaries but, instead, traverses across different modes of existence. ¹⁶ Following this approach – which can be loosely defined and posthumanist – AI-generated art is based on ever-morphing recursive relationships whereby flows and modifications of data act as socio-technical glue, never stable but always in the process of affecting and being affected in return.

In this way, AI-generated art is never completely machinic and detached from human experience – instead, it is permanently interwoven with human experience and expressions of the world. After all, AI creativity being based on data, its core function is to rearrange all of these bits in a way that is simultaneously novel but still recognisable to a human audience. Likewise, creativity that is thereby learned is machinic in the way it is performed but not as a matter of specific machinic aesthetics or sensitivities – because there simply is no such thing as independent machine aesthetics or sensitivities (at least for now), and even if there were, machines are not the target audience – at least as long they do not have purchasing power of their own. Consequently, machine aesthetics are bound to, at least for the time being, remain human-centric, in terms of both learning and the intended audience. However, it is precisely this capacity to emulate human creativity that also acts as a source of perceived threat to humans, particularly to human artists who would see their livelihoods threatened. ¹⁷ Still, the threat would only become imminent if there was also a simultaneous change in societal perceptions of art – similar to that which had taken place alongside the emergence of mechanical reproduction so as to render photography and film artistically and commercially viable.

Of course, not everyone is sold on the idea of machine creativity: after all, it is a legitimate observation that AI lacks consciousness and intent and, therefore, cannot be seen as creative in the human sense. ¹⁸ It is indeed the case that the generative tools of today are not making independent macro-level decisions (whether to create or not, what the subject matter should be, what style to adopt etc.). On the mezzo- and micro-levels, nevertheless, the heavy lifting is done by the AI. Overall, this removal of the need for humans to carry out their ideas in full (from the initial thought all the way to a finished item) can be expected to lead to a significant growth in the amount of content. This growth would also instigate an explosion in the volume of AI-generated content. Still, this does not mean that such changes would spell the end of human creativity or that art would become completely ubiquitous. Just like photography is today an established art form but not every photo one takes with their smartphone is considered to be art, so only a fraction of all AI-generated creative content will be considered to be suitable for inclusion in the ‘art’ category. Nevertheless, controversies to that end are already present, with AI-generated artwork being capable of winning art competitions against human artists. ¹⁹

As Benjamin notes, ‘[i]n principle, a work of art has always been reproducible’ but with a crucial caveat: notably, ‘[m]an-made artifacts could always be imitated by men’; meanwhile, mechanical reproduction, according to him, ‘represents something new’. ²⁰ AI-generated content, in turn, represents something in-between: on the one hand, there is an element of machinic seriality, whereby data patterns in the training sets are identified and restructured into ne, yet recognisable, forms but, on the other hand, AI does not replicate the world from some detached vantage point but, instead, generates output on the basis of and thus renders visible a very specific type of reality – objects, styles and likenesses as they appear in data. For this reason, it might even be possible to say that AI-generated art is truly the art of the times: just like art is generally reflective of the society and of its relationship with technology and the natural world, AI-generated art is reflective of today’s dominant mode of engagement with the world – data with which humans are intimately enmeshed. By making the layers of data and trends and patterns therein visible, AI-generated art reveals the collective unconscious of today’s society, including its deep human-machine entanglement. That might in itself be a source of value pertaining to AI-generated content – a revelatory capacity to render visible the actual digitally enmeshed and entangled conditions of life in contemporary societies. ²¹

The above does not necessarily have to amount to technological progressivism: instead of data as an external objective essence, one should, instead, focus on ‘relational connections, affective forces and agential capacities’ arising from the assemblages of humans and data. ²² Crucially, the introduction of AI underscores a specifically posthuman element by introducing a new – artificial – kind of agency into the mutually affective human-data interactions, turning them into a triangular interaction of humans, data and AI. ²³ As Lupton argues, such assemblages ‘can be viewed as ever-changing forms of lively materialities’, ²⁴ which is a far cry from the progressivist understanding of data analytics as a quasi-magical tool that makes societies visible to an external gaze (perhaps best epitomised by Peter Thiel’s borrowing of his data analytics company name – Palantir – from The Lord of the Rings). Instead, focus shifts onto ‘the indeterminacy of the distinctions between human and nonhuman that human-data assemblages enact’. ²⁵ In this way, AI tools are revealed to dwell within humans as major elements within cognitive processes and causal chains while humans do also dwell in both digital agents and data as datafied subjects but also as sources of machine learning. ²⁶ Moreover, such relationships are never static but, instead, operating in constant recursive loops as no action, datafication process and automated decision is ever pure, standalone and uncontaminated but is, instead, based on multiple similar instances that had occurred prior. ²⁷ Therefore, the collective unconscious uncovered and made visible by generative AI is never a revelation of some real and objective essence but, instead, a glimpse of the interconnected assemblages that underlie societies today.

Notably, as Zeilinger observes in his defence of AI creativity, ‘imitation, mimicry, and copying form the core of how human agents acquire language, learn a craft, and, indeed, create art’. ²⁸ Benjamin also appears to have been largely in favour of such a view, citing the making of replicas ‘by pupils in practice of their craft’ as one of the instances of human reproduction of art. ²⁹ Of course, photography then is seen as perhaps the most significant addition to reproduction capacities. Nevertheless, the emergence of AI-generated works raises a further issue with regards to reproducibility: it is possible to assert that generation by AI is mere ‘making’, that is, reproduction of data patterns, which is significantly inferior to creativity proper. ³⁰ Following the above argument, while humans do engage in copying when acquiring a new skill, once the latter is achieved, they move towards independent self-sufficient creation. While engagement with the criticism of the ideas pertaining to such allegedly monadic existence of human artists is beyond the scope of this article, it must be stressed that such strictly individual focus deflects attention from the way in which human personality as well as social and cultural imprints that are all necessary to artistic output are grounded in the rich interrelations and networks that embed every individual within their lived environment. In particular, the striving for human autonomy and self-sufficiency and, therefore, denial of interrelatedness are criticised by posthumanist scholars on the basis of such an anthropocentric view being discriminatory and only revealing a biased worldview of a privileged minority in the West. ³¹ According to Benjamin, photography had ‘freed the hand of the most important artistic functions which henceforth devolved only upon the eye looking into a lens’. ³² While, to some extent, the collaboration between the human and the machine is preserved even in AI-generated works, ³³ the end-goal, at least for those developing and marketing the technology, is machine autonomy in creativity. With AI-generated art, in turn, both the hand and the eye are to be freed, with the end result solely subjected to machinic processes of generation as ‘humans have no immediate bearing on the process of production’. ³⁴ Here, focus shifts towards ‘developing machine process and machine creativity’ as opposed to simply mimicking human creativity. ³⁵ With the inclusion of deep learning in particular, AI generators increasingly acquire the plasticity and independent form-seeking that had thus far been seen as characteristic of the human brain. ³⁶ In this way, it is claimed, novel and creative works can be generated with no humans being directly responsible for them. ³⁷ Once again, the real currency behind such statements needs to be carefully considered and evaluated, particularly with regards to instigation: as stressed above, regardless of how independently the creative process itself would run, it still needs to be started and directed by humans and it still serves human aesthetic (and consumption) needs. Hence, AI-generated art remains human-centric at both ends: decision to create and appreciation of the end result.

Locating AI-generated works between auratic and exhibition value

For Benjamin, a crucial feature that is lacking in reproductions, however perfect, is the original work’s ‘presence in time and space, its unique existence at the place where it happens to be’. ³⁸ Of course, reproductions have the benefit of bringing artistic works closer to the person enjoying them – such as a photograph of a building or a musical recording that come out of their original contexts into one’s living room. ³⁹ Any capture and reproduction, however, can only come at a cost – the taking away of authenticity and the aforementioned situatedness in time and place; as Benjamin famously puts it, as a result of this displacement, ‘that which withers in the age of mechanical reproduction is the aura of the work of art’. ⁴⁰ Indeed, it is key to the Benjaminian understanding that ‘[t]he uniqueness of a work of art is inseparable from its being imbedded in the fabric of tradition’ which, while changeable and prone to reconstituting how any given work is perceived and interpreted, nevertheless, accumulates as the aura of that work. ⁴¹

It must be acknowledged that the aura is a complex and multifaceted concept that lacks stability within Benjamin’s oeuvre, in particular with regards to its relation with mechanical reproducibility, such as in photography. In important ways, one can witness a clear nostalgic aspect to it, as in describing ‘a state which had already become obsolete’. ⁴² In conventional terms, aura is lost when authenticity and uniqueness are replaced with seriality and replication, not just of the work itself (e.g. one can produce numerous photographs from the same negative) but also of the ideological and material setting (with Benjamin’s particular focus being on the staged bourgeois studio portraits). ⁴³ In fact, it transpires that aura can emerge against and in contrast to the particular setting, as in an encounter with the subject of an otherwise mundane and serial photograph that refuses to be reducible to convention. Generally, the aura refers to that aspect of the object that goes beyond even if the artefact as such is within reach – the situatedness and history of the object that make them unique and more than mere representation or mere following of artistic or ideological convention. ⁴⁴ It is the experience of the object; a quasi-human relationship of gaze laid and returned ⁴⁵ – what might, in today’s terms, be called a parasocial relationship. A crucial case in point here is the ‘imaginary encounter between viewer and image’. ⁴⁶ There is a reciprocity of the encounter between the object and the individual interacting with it, as manifested in the laying and return of the gaze: rather than being located in the very act of reproduction, mechanical or otherwise, the aura of the work is in the specificity of the audience encounter. ⁴⁷ Such encounter, then, ‘breaks down rigid boundaries between self and other, creating a play of identification between viewer and image’. ⁴⁸ It is, of course, an open question as to whether such a break with convention and ability to return the gaze can be attributed to AI-generated content because that would necessitate a significant amount of agency. As for Benjamin, he had already denied such capacity to photographic equipment. ⁴⁹ Nevertheless, AI-generated content does fulfil the aspect of situatedness in time and place; and as the aura is a relational trait, it can still be read into AI-generated objects by the audiences.

Meanwhile, AI art enthusiasts would downplay any doubt regarding the artistic adequacy of AI-generated content, not least by proclaiming such doubt to be merely ‘a temptation to cling to romantic notions of creativity and innovation, praising human genius that created real art’ – a sort of Luddite position. ⁵⁰ Overall, however, the latter position itself appears to be rather defensive, effectively implying that although fully fledged autonomous creativity cannot (at least as yet) be achieved by machines, perhaps it is because it had never existed in the first place. Nevertheless, such an entrenchment does not have to happen in the first place. In fact, there is an argument to be made that AI-generated works possess a degree of situatedness and embeddedness – at least to a greater extent than, for example, photography. While not necessarily manifesting auratic qualities, precisely as a result of their data-based nature, AI-generated works possess the capacity of reflecting their time and place and, in a certain way, tradition. Admittedly, for example, Aubry uses their own version of the Benjaminian argument – namely, that ‘[a]rt is a fluctuating domain of human life that is utterly intertwined with culture, technology, and history’ – with the aim of discarding the very idea of ‘art-ness’ of AI-generated art. Nevertheless, the opposite can, in fact, be deduced from such claims: namely, that AI-generated art is fundamentally entwined with and part of today’s culture and society, situated at a precise moment in history through both its focus on the data record and its reflection of the dominant technology of the day. ⁵¹ As per the above, such works make visible the underlying essence of the contemporary society – recursive interrelationships between data from the multiple uploads and interactions that we engage in online, digital artefacts (AI agents in particular) and humans themselves. Hence, contrary to photography, which merely captures and reproduces whatever happens to be in front of the lens, AI creativity has a revelatory aspect to it, bringing forth the deeper, invisible strata of life at this particular moment in time – including the multiple societal biases inherent in the data from which the generator learns about the (human and non-human) world. ⁵² In this way, while AI-generated content as such cannot have a critical purpose, it is entirely possible for the audience to ascribe a critical reading, thus turning such output into a tool for social criticism through the content-audience relationship.

As already evident from the above, questions are raised – and with good merit – with regards to AI-generated output vis-à-vis common understandings of creativity and, therefore, art. While the development of a fully fledged account of AI creativity is beyond the scope of this article, it must be stressed that the position embraced here is critical towards mainstream anthropocentric (or, rather, Western-centric) accounts of creativity. Indeed, creativity, as Stephensen observes, is permanently embroiled in what could be called a ‘definition game’, that is, a constant project of rearticulation in light of changing economic and political concerns as well as research agendas, and not something objective or pregiven. ⁵³ To this effect, there can be no such thing as a natural or neutral definition of creativity (or assertion of the absence thereof). For Henriksen, Creely and Mehta as well, creativity is interpreted as ‘a fundamentally politicized concept’, one that is less about processes and artefacts and more about the exertion and normalisation of power over who and what counts as creative and, therefore, valuable (and, likewise, who and what can be discarded as worthless). ⁵⁴ It has thus served as a tool for erasure of non-Western cultures and indigenous views on creativity, while even within the West, women, for example, have for centuries been largely denied the status of creative agents; the same strategy could then be seen as manifesting itself with regards to AI-generated content. ⁵⁵ Moreover, as such Western-centric politics of creativity has been further entrenched by becoming entangled with popular understandings of fundamental human attributes, resistance to AI creativity becomes an (almost) automatic defensive reaction. ⁵⁶ Still, the preceding is not intended to suggest that AI-generated content is by definition creative and, therefore, potentially classifiable as art, just like today most human-created photo or video content, intended for uploading to social media or just mindlessly produced, contains at best a modicum of creativity and can hardly be ascribed the status of artistic works. Nevertheless, the implication is that AI-generated content can in principle be creative and, therefore, amount to a work of art, albeit usually with a different societal function than the more ‘traditional’ artworks.

Crucially to elucidating the above function, Benjamin notes that the work of art has its origin in ritual – a heritage that, he claims, remains at the very least implicit: for example, this ‘original use value’ remains ‘recognizable as secularized ritual even in the most profane forms of the cult of beauty’. ⁵⁷ Mechanical reproduction, meanwhile, separates the work of art from ritual, shifting emphasis from ‘cult value’ to ‘exhibition value’. ⁵⁸ Such a shift is certainly fundamental and signifies a complete change in function: what first was a move away from an object of magic to an object of art, now has turned into ascription of exhibition value to which the artistic function could become accidental. ⁵⁹ Something similar is echoed by Manovich with regards to AI-generated content in terms of a potential dissociation of AI-generated art and creativity. ⁶⁰ The latter point, nevertheless, needs to be partly objected in the same way Benjamin criticised the stretching of the old concepts to completely new types of expression. Instead of dissociation, a more productive alternative would be to reconsider creativity as an ancillary feature intended to maximise exhibition value if not in exchange for monetary reward, then at least as a generator of social value, which in today’s societies are perhaps most prominently reflected through views and likes. To this effect, it might seem, AI-generated art would continue the trend of disenchantment and exhibition focus as identified by Benjamin but without completely losing creativity out of sight.

In the same vein, there is a change in the appreciation of a work of art, with it shifting towards easy and unchallenging enjoyment as ‘[t]he conventional is uncritically enjoyed, and the truly new is criticized with aversion’. ⁶¹ Unsurprisingly in this context, AI creative output is at its most appealing to the audiences when it is just slightly different from the broader context but not too much so that audience expectations are not violated. ⁶² After all, it is becoming clear that AI is capable of generating results that are simultaneously novel and pleasant, ⁶³ and serialising such capacity transpires to be the goal of AI-powered generative systems. In this way, it is not difficult to assume that art-ness is in the eyes of the audience and not in the objective qualities of the work itself ⁶⁴ – at least, once again, if one accepts the Benjaminian argument of a shift towards exhibition value as a key consideration. Here, again, creativity is found to be in the form of developing and implementing novel attention – and reception – maximising ways of rendering data patterns visible.

The shift towards the exhibition value of the work of art is also clearly visible in the corresponding shift towards art as (intellectual) property. In this way, art leaves the domain of creativity and enters the economic domain, becoming a form of capital. ⁶⁵ Symptomatically, then, the core arguments of those proposing protection of AI-generated art in a way at least similar (if not analogous) to human-produced works focus on the need for investors in AI technologies to reap a profit. ⁶⁶ Likewise, one can increasingly witness art following a path trodden by other forms of digital content – one that leads towards platformisation. ⁶⁷ Here, again in line with Benjamin’s foresight, artistic function has become secondary to return on investment and its precondition – exhibition value. However, there are still ways in which the situation here is less than straightforward because, at least to some extent, AI also works as re-enchantment. Notably, there is a rather prevalent tendency to understand AI in (quasi-)religious terms, from awaiting some sort of divine intervention (such as the arrival of superintelligent AI) to ascribing machine creativity the power to lead towards salvation. ⁶⁸ Hence, contemporary society seems to be prone to ‘deification of AI’ through ascription of intent and (almost) supernatural powers. ⁶⁹ In creativity, for example, AI can thus be seen as a source of superhuman fantasy and inspiration while the revelatory nature of art-from-data generation can be understood, at least from an industry perspective, as an almost supernatural insight. ⁷⁰

One should, however, be sceptical about unbridled promotionalism either: such focus on supposedly higher insight from a hierarchically superior vantage point would only mean substitution of one set of biased assumptions (anthropocentrism) with its opposite. Instead, one should see the input of AI precisely in the more-than-human sense discussed above: as a demonstration of how social practices ‘are always situated in the lively web of interdependencies’. ⁷¹ Only in this way is it possible to move beyond the narrative of competition between humans and machines and opt for a more egalitarian, interdependent relationship. ⁷² Hence, once again, generative AI capacities should best be seen in terms of expansion of the breadth of interconnectivity and interactivity rather than temporal or qualitative substitution. Quantitative change (the growing ubiquity of AI-generated output) is, of course a very different matter and a driving force behind the platformisation of AI-generated content in the broad sense, but, as in other kinds of platforms, it has little to do with artistic endeavours properly conceived.

Nevertheless, while in the past the above acts of insight would have been singular occurrences, AI-generated content unavoidably has an element of seriality to it. Photography also has rendered authenticity and the idea of one work as ‘the original’ nonsensical as any number of prints, all of the same quality (and manifesting the same lineage from the act of reproduction, that is, taking of the photo as such) could now be produced. ⁷³ This is also the main reason why AI-generated works cannot have full auratic quality: despite being embedded in the fabric of culture and society, they nevertheless retain the impossibility of distinguishing an original from a copy as they can be multiplied with only a few clicks. Admittedly, there are also attempts to change that equality of digital artefacts, for example, through the use of Non-Fungible Tokens (NFTs). By associating a particular item with a blockchain entry and providing an indelible trace of ownership, NFTs purport to reproduce the idea of the original, alongside the materialistic aspirations of an art collector, thus simultaneously capitalising on supposed authenticity and exhibition value (through creating artificial scarcity). Nevertheless, such attempts are better seen as artificial means of monetising investor desire for exclusivity rather than changes to the nature of the digital work itself.

AI-generated works as a set of optimisations

Benjamin stresses how the aura of the work of art is lost particularly in film (contrary to, say, a theatre performance), since it is shot in a non-linear manner, cut and then reassembled from multiple isolated pieces, visually devoid of technology and yet completely permeated by it; moreover, film, according to Benjamin, lacks the direct, always one-off, interaction with the audience within which the situatedness (authenticity) of the performance lies. ⁷⁴ This is not entirely different from Du Sautoy’s criticism of algorithmic creativity: for him, proper creation only happens as a ‘meeting of minds’, that is, when there is room for audience’s co-creation with the artist while there is no such space vis-à-vis an artificial entity. ⁷⁵ On the other hand, it can be argued that the disappearance of aura in film pertains to, not least, the mismatch between the mode of the world’s presence and the mode of its representation. In a world that is physical-first, technological representation, cutting and pasting constituted an end result that was devoid of authenticity and a capacity for the gaze to be rested upon and returned while also, by being malleable, easily conforming with the conventional – all detrimental to an auratic encounter. In a world that is digital-first, meanwhile, digital rendering of creative endeavours constitutes perhaps the sole authentic way of representation (and, given the nature and scope of data, likely the only feasible way of representation). As Benjamin writes with regards to film, ‘[t]the equipment-free aspect of reality here has become the height of artifice’. ⁷⁶ Meanwhile, the use of AI marks a change in the sense that there is even no contraption of reality to be hidden – but, instead, enmeshments of humans, data and digital technologies.

A further transformation, according to Benjamin, pertains to the proliferation of works themselves: from few creators facing large audiences to an increasing capacity of every individual to express themselves. ⁷⁷ Such an assertion, of course, sounds even more familiar in times of social media. Indeed, technological transformations and an associated change in social habits during the recent decades have made the creation and (particularly given Benjamin’s comments on the emergence of ‘exhibition value’) display of mechanical reproductions of the world seemingly banal. Generative AI capacities may well venture into the same domain, democratising the production of art and making it an everyday occurrence. ⁷⁸ This is not dissimilar from the effect that smartphone-mounted cameras have had on photography, taking the latter way beyond the level of proliferation that Benjamin would have imagined. While cameras as such, and subsequently their increasing ubiquity and integration into phones, drones and elsewhere have democratised reproduction of the world, AI-generated art does something similar with what goes beyond mere reproduction and representation. Paintings, sketches, collages etc. have always necessitated a human hand, along with specific skill and ability. AI has democratised the capacities of content creation so that anyone with easy-to-use tools could generate works that had previously been associated with a human touch as well as a (high) level of skill and ability.

What is notable here, though, is the irruption of technology into the work of art in a way which is simultaneously plain and hidden: on the one hand, it is plain because technology is the essence and substance of the work; on the other hand, it is hidden because that which passes for exhibitionally valuable (popularity becoming the ultimate criterion) is that which enables the viewer to forget, or at least to conveniently ignore, the absence of human involvement. At the very least, even when technological involvement is plain and well-known, the techno-centricity of today’s societies has likely resulted in a desensitisation of audiences to AI being embedded ever more deeply into the fabric of the everyday. Again, exhibition value becomes the core consideration, as opposed origin. Indeed, popularity and public enjoyment were seen as new and emergent value criteria already by Benjamin ⁷⁹ and have only increased in prominence since.

One more change in terms of relationship to an underlying reality can be inferred from comparing AI with Benjamin’s contrasting of the painter and the cameraman. For him, while ‘[t]he painter maintains in his work a natural distance from reality, the cameraman penetrates deeply into its web’. ⁸⁰ AI, meanwhile, goes both ways: simultaneously penetration (by digging through data to produce a representation inferred therefrom) and distance (building an artifice rather than direct representation). Likewise, Benjamin asserts, while ‘[the picture] of the painter is a total one, that of the cameraman consists of multiple fragments which are assembled under a new law’. ⁸¹ Here, again, both aspects are manifested in AI generation: assembly from bits (data) but from a total inventive perspective rather than minuscule ordering of a plot or a tightly framed shot.

In addition, Benjamin does read psychoanalytic capacities into mechanical reproduction of reality, particularly in film, through technological capacities such as close-up and slow motion which are seen as potentially revealing glimpses into the unconscious by making visible what would normally escape the naked eye. ⁸² AI, meanwhile, elevates this excavation work to the collective level by allowing a glimpse into the collective unconscious digging across the complex interrelationships of humans, data and technology. Once again, elements and aspects that would otherwise escape the naked eye are rendered visible, revealing the data-based collective unconscious of contemporary societies. Thus, AI-generated art acquires a revelatory capacity of sorts. Zeilinger claims to offer an explanation by asserting that AI-generated works are perhaps best understood as ‘copies without originals’ in a Baudrillardian sense, creating a simulacrum of the real (from the data crunched and analysed and patterns thus identified) but without actually referring to anything beyond itself. ⁸³ That, however, is a misleading interpretation. Instead, the object of reference (data), being digital and only available to be parsed by AI tools, tends to escape human attention, thereby creating the illusion of AI-generated works as simulacra.

Hence, a more complex account of the revelatory capacity of AI-generated content is necessary – one that accounts for the complex interrelationships between humans, AI and data. It must be stressed, at the risk of repetition, that ‘the world is not composed of preexisting and already-formed entities awaiting discovery’ either by the humans themselves or by tools and other means created by them (such as such as data-crunching algorithms); instead, the very practices of knowing and revealing such entities ‘play a constitutive part in bringing their objects of study into existence’. ⁸⁴ In addition, one must inquire to whom things are revealed and in relation to whom any knowledge and understanding is formed as the post-anthropocentric framework espoused in this article ‘displaces the notion of species hierarchy and of a single, common standard for “Man” as the measure of all things’. ⁸⁵ Closely related with the preceding is the widespread push for transparency and against algorithms and AI agents acting as ‘black boxes’ that one cannot see through. As stressed by Dewandre, transparency in this context becomes ‘a ground for experiencing autonomy and control’. ⁸⁶ Against that, adopting a much more vulnerable position of a ‘relational self’ ⁸⁷ would entail relinquishing the desire for autonomy and power through supposedly neutral knowledge. Hence, the revelatory capacity of AI-generated content and its ability to make visible the collective unconscious of contemporary society lies in the complexity and impurity of representation. Just like the unconscious does not manifest itself verbatim, so AI-generated content brings forth the messy interrelationships of agents that are, at least partially, opaque to each other: humans are never fully datafied and, inasmuch as they are, data capture differs between the diverse sets of tools, techniques and platforms. The pathway towards particular decisions or content items churned out by AI is often inexplicable, and the potential for ever new uses of data is hard to exhaust. In a way, it is the ever-present likelihood of errors, glitches, deformations and hallucinations in AI-generated content that signifies its revelatory potential vis-à-vis the complexity of today’s societies.

The above, however, does not automatically mean that such revelatory capacity is generally appreciated. What AI shares with mechanical reproduction is not only the importance of exhibition value but also that their reception takes place ‘in a state of distraction’. ⁸⁸ If anything, it transpires that the distraction has only increased as a result of the ever-growing abundance of digital content available online. For this reason, the attention available to any work of art has shrunk, meaning that there is a need to maximise impact. Crucially, being always informed by data, AI-based generators can also learn and adapt themselves to emerging trends and changes in public taste and perception, thereby optimising their output for maximum exhibition value. ⁸⁹ Moreover, AI-generated works challenge the paradigm that art is something stable and identical to all. Of course, art has never been identical to all in the sense that it has typically been understood and interpreted differently by different people and at different times. Nevertheless, they would all have encountered the same work (withering and deterioration notwithstanding). Meanwhile, AI-generated content can, at least in principle, be adapted to the data of the individual accessing it, thereby increasing appeal. In a slightly different domain, meanwhile, AI is also enabling personalisation of mass culture in ways ranging from content moderation (selection and display of particular items) to personalised generation, that is, generators adapting to the preferences and style of their user. ⁹⁰ Hence, it transpires, AI is increasingly poised to take over the entirety of individuals’ cultural and aesthetic experience.

It is, therefore, no surprise that AI begins to be seen as a potential competitor in the artistic field as well. Indeed, the automation and jobs substitution discourse was initially focused on the more manual and tedious jobs. Nevertheless, it is becoming evident that AI is capable of pushing ‘traditional’ creators out of the market. ⁹¹ This might not be the case with the high-end art market (at least not in the short term). Nevertheless, it is easy to imagine those working in the creative industries as well as less established artists being at risk of automation. While it is correct that photography, contrary to the original fears, never ended up replacing painting, AI-generated art is more like painting than photography is. After all, photography is about capture and reproducibility in the strict sense, whereas AI generation is a matter of data pattern rearrangement and reassortment that uncovers things beyond visibility.

Conclusion

The debate pertaining to the ‘art-ness’ of AI-generated content notwithstanding, it transpires that at least an affinity between human and AI-generated art can be established. Of course, these are works of an entirely different kind, but so was photography and film in Benjamin’s time. It is also worth noting that the question of ‘art-ness’ is only relevant to a small fraction of the content generated by AI (even should one narrow the scope of discussion down to the two types that at the time of writing are attracting the most attention – images and text); after all, to again draw an analogy with photography, while photographs abound, particularly on social media, only a very small fraction of them would meaningfully invite discussion in artistic terms. Nevertheless, the very introduction of AI-generated artistic artefacts constitutes a shift that needs further conceptualisation.

Overall, it transpires that while AI-generated art represents a continuation and even intensification of some of the trends identified by Benjamin, it by no means constitutes a linear progression. In fact, in important ways, AI-generated art stands in-between ‘traditional’ art and Benjamin’s take on photography and film. Notably, AI art is societally embedded in the sense that it is based on data generated by human societies. In this way, at least some aspects of authenticity and auratic quality are retained. Likewise, AI-generated art has a revelatory quality, making visible the layers of the collective unconscious of today’s societies – that is, data patterns – in a way that is in line with the psychoanalytic capacities that Benjamin saw in photography and film. More generally, it is possible to claim that while Benjamin saw photography and, even more so, film as a matter of artifice, AI-generated art constitutes perhaps the most genuine representation of today’s societies characterised by the enmeshment of humans, data and digital tools.

Nevertheless, continuity-wise, AI-generated art’s optimisation towards maximising its exhibition value is noteworthy. Moreover, it must also be kept in mind that such art is still anthropocentric in terms of its aesthetics: not only it learns from humans but also the target market are humans. Hence, the difference is less in terms of the technical aspect of creation, since it retains a human orientation (just like both a painting and a photograph are human-oriented) but, instead, the source and manner of representation: while the earlier technological tools (such as the photo or video camera) merely reproduce what is set before them, AI-generated art comprises of novel representations of (the world seen as) data, human and AI assemblages.