Expansive Images: Understanding Photography and Generative Imagery
Image: Liliana Farber, Untitled No. 3 from the series Isolarii, 2021. © Liliana Farber
As public discourse rages – debating the ways in which artificial intelligence (AI) systems will automate, streamline and disrupt human life – the artists, scientists, engineers and scholars who have long engaged with this field of technology let out a collective sigh, writes Catherine Troiano.
The past two years have seen rapid development in the AI ecosystem. In 2021, DALL-E, a text-to-image model, was released by OpenAI. It was quickly followed by other similar systems, such as Stable Diffusion or Midjourney and, in 2022, OpenAI released their language model called ChatGPT – now a household name synonymous with the surge in AI capabilities.
The creative potential of computational technologies has evolved in tandem with their rise. This issue of Edition One argues that artists engaged with AI prior to the recent boom can help to organise, reflect on, and think through the myriad ethical, social and cultural issues that AI systems raise. But looking towards artists in this way also requires looking towards the institutions and frameworks that platform, promote and engage with them, considering how and with what impact cultural output diffuses into its wider environment.
An Age-Old Question: ‘What is Photography?’
AI is already impacting one area more than others: imagery. Generative imagery has upended the creative and financial economies of image cultures, broadly understandable as two groups: images perceived as photographic and images not. The 200-year-old question of ‘what is photography?’ has been energetically revived by generative imagery: separating photographic imagery, or images that mimic photographic qualities (also known as photo-realism), from others.
For much of the medium’s history, definitions of photography relied upon the involvement of light and light-sensitive materials, and, to a lesser extent, camera equipment. These tenets served to retain a somewhat singular definition of photography, even during the ‘digital turns’ of the 1980s and 1990s.[1] Cameras – having previously been relatively incidental, as demonstrated by a rich history of camera-less photographs – ironically held together a photographic iconography specific enough to unite both sides of a bitter dispute about the so-called ‘authenticity’ of photography made, manipulated or otherwise informed by digital technologies. Such notions of authenticity, originality and authorship stretch back to the advent of photography itself (when it was accused as the assassin of painting), through many technical and conceptual revolutions that have caused photography’s repeated death (spoiler: it is always reliably resuscitated), and into the present paradox.
AI Upps the Ante
That said, AI appears to have raised the stakes, and not only for photography. Concerns about models that scrape copyrighted images from the internet to train algorithmic software has led to numerous calls for regulation; though responding to this problem with capitalist solutions such as IP and copyright law has also been challenged.[2] Whilst the advent of digital technologies in the 1990s framed a debate between photographers, albeit of different creed, the advent of AI has changed the issue to be between photographers as a whole, and machine-empowered corporate greed that threatens to displace an entire group of creatives.
To compound this, images generated by text prompts have moved photography away from an exclusively visual medium: a particularly difficult conceptual shift for a notoriously unwieldy, but fundamentally image-led, sector. So much so, that Boris Eldagsen called for the term ‘promptography’ to be used for pictures generated by text-to-image models, after he won (and declined) the creative category at the 2023 Sony Photography Awards for an image made using DALL-E2, allegedly unbeknownst to the judges.[3]
The Sony debacle doubled down the perceived need to define – or re-define – photography, in a way that excludes generative imagery.
Pushing Against a Limiting Debate
Attempting to singularly define photography when the premise of photography itself has been exponentially expanded, altered and reconfigured would be arbitrarily limiting. It would flatten the nuance of generative imagery in an ‘AI binary’, much like the binary discussions around ‘digital’ and ‘analogue’ practice flattened over a century of layered photographic developments.
It would also discount that generative imagery can be inherently and indisputably photographic, even when ‘only’ photorealistic, and crucially reliant upon human input. Caroline Sinders and Anna Ridler embody this in their contribution <INSERT HYPERLINK TO THEIR ART ARTICLE IN THE EDITION>, and even the case of Eldagsen emphasises an aspect of human activation – not only in making images, but in their reception.
Regulation is clearly important for any advance in technological capabilities with the social and cultural potential of AI, but the recent shift in discourse positions generative images as ‘deceptive’ and ‘verifiable photography’ as ‘truth’. ‘Truth’ has never been an appropriate descriptor for the photographic condition, and if we think instead of the demonstrable potential of photography to be speculative, then such impossibilities are rendered irrelevant anyway.
How Institutions Can De-Limit Debates
In 2020, the V&A, supported by the Manitou Fund, commissioned Jake Elwes to create a new iteration of The Zizi Show for display in the V&A Photography Centre. As this issue explores in detail, The Zizi Show features deepfake[4] versions of performers from London’s drag scene, and it is an example of generative imagery with both deeply photographic origins and a clearly human framework.
Elwes’s project shows deepfakes deployed for a constructive purpose, rather than for more sinister ends, raising questions about the biases, ethical failings and real-life discrimination reproduced by AI systems. It is the inaugural display in the V&A’s Digital Gallery at the heart of the museum’s Photography Centre, a suite of seven galleries dedicated to showing the V&A’s extraordinary collection of photography.
This means that visitors can traverse nineteenth-century photographs, a deepfake drag cabaret and varied other practices spanning the whole history of photography in a single visit.
Surely it is more helpful, here, to present an expansive understanding of photographic image-making, without expectation that everyone will agree, rather than focus on whether something is sufficiently photographic?
It is entirely plausible that the verbiage around photography will grow to accommodate the notion of generative imagery. But how do we conceive of the ‘new real’ in real time? How do museums engage with rapidly changing technologies against a backdrop of necessarily lengthy schedules? And how can we curate incisive digital or contemporary practices whilst remaining mindful of the past, in institutional environments deeply connected to photographic histories and trajectories that have led us to this moment?
One strategy – which underpins the V&A’s digital programme in photography – is to engage with AI systems through creative practice, as this issue makes a compelling case for. The V&A was one of the first museums in the world to collect and exhibit photography, from 1852 and 1858 respectively. To engage, then, with contemporary artists shaping the future of photographic practice is simply to continue a long and effective V&A tradition of looking to contemporary cultural production to work through new ways of making. New developments were not always met with widespread public acceptance, and curatorial provocations are arguably more useful than curatorial directives.
Art and Curation that ‘Debates Forward’
Following its commission, The Zizi Show was acquired for the V&A’s permanent photography collection, one of various recent acquisitions that engage with generative imagery and AI. Another is work by Liliana Farber, who uses AI to explore ideas of mapping and computer vision. Her project Terram in Aspectu shows google-earth lookalike images of islands originally referenced in historic cartography, but which were later proven to not exist. Farber fed information taken from historical sources to an open-source machine learning algorithm, trained using satellite photographs taken from Google Earth, which reproduced the erroneous islands as their own ‘satellite photographs’. Her later project, Isolarii, continues this line of enquiry with a different methodology: the works are data collages, ‘woven’ together by custom machine learning software processing a range of imagery collected by Farber to create ghostly evocations of early world maps. Farber’s work visualises the colonial endeavour to map the world, referencing the techno-industrial reproduction of colonial patterns and the ecological contradictions of photographic or image-based practice. It enquires into the consequences of systems or technologies that determine ‘reality’ and the creep of corporate reach, critiquing the providers and makers of the technologies she implements. Farber’s acquisition also challenges traditional assumptions of what digital practice looks like: these works are paper-based inkjet prints, framed and wall-mounted like many other pre-digital photographs.
It is worth noting that artist-driven practices represent an approach that broadly subscribes to the cultural institutional frameworks of the ‘art world’, and that there are other institutions and even other colleagues within the V&A, such as curators of digital design, who frame notions of cultural production in a less individualised or maker-centric way.
But if we do not desire a single understanding of what photography is or can be, then we do not require such an understanding to be universally reflected in the endless images, output, interactions and mediations that could be experienced as photographic.
We thus leave our institutions and our audiences open to the fullest spectrum of contemporary culture, recognising the paradoxical realities of photography, imaging and technologies.
—
[1] Mario Carpo, The Second Digital Turn: Design Beyond Intelligence (MIT Press), 2017.
[2] Rob Horning, ‘The Theft of Alien Labor Time’, in Internal Exile, April 2023.
[3] The term ‘promptography’ was originally used by Christian Vinces, a Peruvian photographer, before being popularised by Eldagsen. See: https://www.instagram.com/christianvincesfoto/ (accessed 2/11/23).
[4] Daniel Palmer and Katrina Sluis position deepfake technology as the final precursor to text-to-image models, in a lineage stretching from image aggregation and processing platforms such as Google Images (2001), Flickr (2004) and Amazon Mechanical Turk (2005); to the development of neural networks – adversarial or not – and ‘deep learning’. See: Daniel Palmer and Katrina Sluis, ‘Photography after AI’, in After AI, Artlink 43 (2), spring 2023.
