In 2021, OpenAI released DALL-E, a tool that produces realistic images based on a text prompt. It was aptly named after the surrealist Salvador Dalí and the robot WALL-E. This was the start of a new revolution in AI art. Since then, OpenAI have released updated versions with ever more impressive results and companies such as Midjourney and Stable Diffusion developed their own tools, providing anyone with the opportunity to generate high-quality images from a few words at the click of a button. This caused an explosion of AI-generated art accompanied by a mixture of shock and excitement from the creative community, eager to explore the new technology, yet at the same time concerned about its future.
Artists have historically been working with the latest computer technologies of their time for decades. From the 1960s onwards, as access to computers became possible at universities and research labs, artists such as Frieder Nake and Vera Molnar began to develop algorithms to generate visual patterns, followed by Harold Cohen’s AARON, a series of computer programmes that used tools such as pens and robotic arms to produce original artworks. In the 2010s, deep learning technologies began to take hold as the combination of large datasets, neural network algorithms and increased computational power meant that AI could quickly learn patterns to produce images, texts, and sounds. This filtered out into the artistic space beginning with Google’s DeepDream garnering public attention in 2015 with its psychedelic puppy-slug creatures, followed by the generative adversarial networks (GANs) peaking around 2016-2020 with their marked leap in image generation capabilities and then the emergence of generative pre-trained transformers (GPTs) for text generation from 2019 onwards. Artists such as Mario Klingemann, Anna Ridler and Memo Akten flocked to the new tools, developing their own models, painstakingly making their own datasets, and evaluating the technology critically. Major exhibitions at institutions including The Barbican and HeK Basel followed. The sale of Obvious’s Portrait of Edmund Bellamy, a blurry AI-generated portrait signed with a GAN model equation, then printed out for display in a traditional golden frame, fetched a record $432,500 at Christie’s in 2018. AI art had finally become mainstream, or so we thought, as the rising interest encouraged accessible tools, commercial opportunities, and the application of GANs and GPTs across all niches.
In some ways, DALL-E and Midjourney brought a breath of fresh air into the AI art field, encouraging a new generation of creators including the exhibition artists Boris Eldagsen and Ben Millar Cole. With his decades-long experience in photography, Eldagsen saw an opportunity to make images without shooting on location. His prompt-writing skill and multi-stage image editing process have led to the creation of works such as The Electrician, a black and white image of two women, that was due to be awarded the Sony Photography Prize until Eldagsen rejected it on the grounds that it was not photography, but ‘promptography’, a term coined by the artist to describe the new art form based on images generated from prompts. Eldagsen took a stand, inviting the prize organisers and the broader audiences to consider the difference in process between light-based photography and text-based promptography as well as the implications of the new technology. Meanwhile, the work of Ben Millar Cole combines unexpected phrases generated by older recurrent neural network-based (RNN) systems from the mid-2010s, which generate text one letter at a time, with the latest image generation possibilities of Midjourney, bridging two different eras of AI art: one that glorifies the creativity of the limitations and errors of the machine and the other that is much closer to perfection, but still with the existence of the uncanny valley. This is underscored by the close relation- ship between titles – put together they could form a poem – and the image, guiding the viewer’s interpretation of the human-like figures, themselves almost a remnant of the early GAN era, in which images with mislocated limbs were much more common.
A creative technologist by training, Nouf Aljowaysir works with multiple AI technologies in her series Salaf including image recognition for insights into machine perception and image segmentation, which splits an image into different parts, before using StyleGAN2, a GAN-based image generation model that consists of two neural networks, one that generates images and the other that determines whether the data is real or fake, an interplay that yields high quality results. The absent segments in the generated images, where the detailed background unexpectedly meets a void in the foreground, lend the pieces a unique aesthetic in AI art, like a digital iteration of cut-outs. Whilst AI offers a realm of possibilities of creative expression, it is just as known for its intrinsic issues with bias. The AI systems of today are typically trained on large datasets containing millions of images, which despite the size are not always representative of the world or categorised in a manner suitable for their applications. This theme of bias is explored through stereotypes by the exhibition artists. Ben Millar Cole’s work questions the associations we might have upon looking at the hooded figures at the edge of the woods, the initial military impressions dissolved through thoughtful titles presenting each group anew. Boris Eldagsen’s works show men and women from the 1940s with unusual twists in the subject matter – cables, intestines, and strings form part of the image, sometimes breaking into the third dimension to challenge our perceptions of black and white photography. Finally, Nouf Aljowaysir’s work highlights the biases within the AI systems that are unable to correctly identify the widespread traditional Saudi clothing and focus on stereotypical images of war-torn Baghdad and Western visions of Saudi Arabia, without paying attention to the local portrayals. These are more frequently passed on through the oral tradition as opposed to image documentation and are therefore missing from the datasets used to train AI systems, something the artist underscores by removing parts of images, where we would expect to see a person or an element of clothing. Aljowaysir assumes the role of a storyteller, visually narrating her exploration of her own heritage through her interaction with AI systems and their stereotypes, captured in her film Ana Min Wein: Where am I from?.
The artists in the exhibition engage with the current possibilities of creative collaboration with AI tools, harnessing the unique affordances brought on by the various technologies, whilst thinking about their implications. Image recognition tools highlight the imperfection of the machine gaze, whereas photorealistic text-to-image models focus on portraying our collective imagination down to the smallest detail, with the prompt engineer at the steering wheel – taking the viewer to the next stage of art history.
