What Are We Really Doing With Generative AI?
Uncovering the historical implications within creative technology of iteration and blurring boundaries between tool, medium, collaborator and ourselves
By placing generative AI (genAI), in the historical context of technology and creativity’s interplay we see, similar to traditional art forms, it as part of an iterative process of self and technological discovery. Previous technological artistic advances (photography, graphic design, acrylic paints) show how new expanded forms of expression come with challenges to the status quo. GenAI doesn’t need to be (and I argue cannot be) built with the ‘automation of creativity’ as a goal. Instead, by building and using it in terms of its potential for inclusive and non-productivity-based features we can augment and enrich our aesthetic experiences. Perhaps the greatest novelty is that we now have a technology that blurs the lines between tool, medium, collaborator and ourselves - the task for us is to acknowledge, identify and develop answers to the questions this poses.
AI’s intersection with art is turning us all into Detective Del Spooner (Will Smith) from the 2004 film adaptation (or in my opinion, the shameless ‘action film-ification’) of Isaac Asimov’s I, Robot. Towards the end of the film, Smith’s character confronts the clinical apple-iMac-inspired-robotic-AI-being to ask questions that now consume many of our conversations relating to AI today: “Can a robot write a symphony? Can a robot turn a… canvas into a beautiful masterpiece?" to which the robot responds, "Can you?”.
This theme of question is not new with one of the most famous versions written by Walter Benjamin in his iconic 1935 essay. However, questions that in the past were able to remain somewhat hypothetical are being confronted with incredibly capable and real algorithmic processes. This requires practical creative, legal and technical work to create systems that can withstand and embrace change in ways that seem fair, inclusive, and fulfilling.
So, to the question of this article’s title - what are we doing with generative AI?
iterati, iteratio, iteration
When using genAI, I feel a contradiction. Whilst part of me wants the output to be 'complete', I also feel I’m simply taking part in an ongoing process. Perhaps after one or many generations, I may feel compelled enough to share the latest output on social media or point my phone at a friend, hoping it will evoke a moment of intrigue or perhaps a small laugh (usually the digital equivalent of a sharp exhale of breath through the nostrils). Every time I do this I feel I learn a little more about what and how I am trying to communicate whilst also learning what the capabilities of varying genAI applications are. It is this iterative process of self and technological discovery that I believe to be what we are currently doing with genAI.
This process of iteration is consistent with approaches taken in the making of many varied media of art. Each medium comes with its own history and means of iteration. Reductively, nearly all art comes from humans using a technology (defined broadly in this case as an invented tool or process) to make visible the feelings and thoughts evoked when living in their context of systems and histories (which have also ultimately led to the creation of the technologies they are using).
For each medium, the underlying feelings, expressions or definitive destination seem not to be known at the beginning of the process but, like genAI, come through a process of iteration that informs creators’ ideas and their artistic manifestation. For some media, this iteration can happen within the same piece. X-ray scans of paintings show this iteration has a long history both as a means of refinement and re-imagination.
materiality of ai as a medium
Every technology we employ to manifest our innermost thoughts and feelings inherently shapes our expression. The nature of the technology, often rooted in its material characteristics, predisposes certain aesthetic experiences.
For example, graphic design emerged from technologies of printing being implemented for artistic means at the beginning of the 20th century and still often evokes its characteristic high-contrast, blocky modern feel. Though it's predominantly digital now, those physical origins still echo.
At about the same time, photography was becoming cheaper and portable through Kodak and Polaroid’s focus on compact cameras and faster-developing methods. Photography had already disrupted the art world’s obsession with hyper-realism paving the way for movements like Impressionism in the late 19th century (now that anyone with a camera could create realistic representations). The 20th century saw photography become an amateur activity focused on commemorating life events and social change. Subjects became fleeting candid moments rather than the previously favoured staged formal group shot forced by material properties and costs of previous photographic techniques.
Shortly after, in the mid-20th century, acrylic paints allowed for paintings to maintain their vibrancy and detail for a fraction of the cost and effort. This gave access to many more artists and subjects previously considered too mundane or abstract for most painters, bringing new modes and motifs of visual expression images into the mainstream.
These examples underscore a common thread: each technological evolution in the arts opened new avenues of expression while challenging the status quo. I can only imagine the discomfort of traditional artisans—wood etchers, calligraphers, and oil painters—trying to understand the new and unfamiliar worlds of graphic design, mainstream photography, and acrylic paints. This historical pattern resonates with today's concerns for sectors considered at high risk from automation like manufacturing and transportation but now this extends to creative industries such as graphic design, animation and script writing to name a few.
The examples given above point towards a future where genAI disrupts the creative world but, rather than automating, becomes a new medium expanding expression for new types of artist. This is of course riddled with assumptions and begs the question “Will genAI be different?” which will be the focus of the final sections.
inspirational stimuli
Is there a difference with genAI or is it just a technological continuation in the history of art-making? There are certainly similarities but a key difference is that in the past nearly all media required the creator to define the creative direction taken. Questions of 'why', 'how' of 'what else' were left to the creator to answer in moments of 'inspiration'. These moments relied on countless and largely intangible stimuli found ‘by chance’ during walks, dreams/hallucinations, books, conversations etc. With AI I believe we have a new additional method. It doesn’t seem there is anything inevitable about this new approach having to ‘streamline the creative process’ or ‘automate human creativity’. Rather, genAI can (and already has in some areas) extend existing creative experiences by offering new realms for expression in ways not previously available.
i taste ai tastes
In the same way as developing a piece of art in the past may have required many iterations and repeated inspirational visits to the artist's favourite hilltop (or bar) before the right moment of creativity hit, so too with genAI. The repeated iterative approach I mentioned earlier allows for the identification of aesthetic features (used in its broadest definition as the subjective interpretation of beauty) that the creators may want to amplify or reduce whilst also uncovering more about the creators’ underlying and evolving tastes and judgements.
In addition to learning our own tastes from this process, we learn about the ‘tastes’ of the genAI model being used. There still remains a question of whether learning a genAI application’s characteristics is just an advanced version of the previously mentioned 'getting to know the technology's material properties' as one might when learning painting techniques. Whether this is the case or there is something else going on, anyone who has extensively used genAI can tell you that different models display idiosyncratic characteristics, some more obvious than others.
Without meaning to, I often find myself attempting to find reasons for why these characteristics have developed. Is it the data used in training? Is it a feature explicitly added by the developers? Is it a reaction to my specific method? Is it an emergent behaviour from the underlying code? These are partly answerable questions but often there is no definitive verdict and perhaps never will be.
Often, characteristics of AI are only called into question in the critiques of models, such as their "not being able to do hands" when generating the wrong number of fingers. Or comments on the particularly distinctive style of Midjourney images which go for an epic (but now cliché) photorealistic look. These all include fragments of questions I think are essential to ask: How could the aesthetics of these images be different? What feelings emerge towards outputs from particular models? What moments were disappointing or frustrating during the process? When was a different programme needed to get the desired results?
It is in experimenting and asking these questions I believe we can appreciate how deeply 'in progress' the AI project is and define the location and shape of the areas in which we want it to improve. Through this investigation and participation, hopefully, a much wider group of people can contribute and develop future models which align with and inform their (hopefully beneficial) desires. In asking ‘why’ we care so much about six-finger hand generations perhaps we get at an underlying set of beliefs. Is the gold standard of image generation really photorealism? If not, then how can models and applications be built to suit the needs of different groups and individuals needs?
extending expression
What we are doing with AI can feel like automatically generating photos and so critiques on the photorealism of an image may make sense. I can’t help but make the comparison to the initial scepticism of photography, from figures such as Baudelaire in the 19th century, that photos lack the expressivity of paintings because of their mechanical nature. Noticing the brushstrokes and composition of a painting still evokes wonder today but so too does the preparation, timing and skill of photography. Attempting to judge a new media production by old paradigms seems like an approach that is unlikely to allow us to see the opportunity and beauty in an ever-changing world.
The aspiration and beauty of genAI I see is the broadening of access to processes of self and material discovery and expression previously reserved only for those with the necessary confidence, resource, and artistic competency. There will continue to be countless moments of enchantment from art, plants, animals, other people or daydreaming.
We do not need to try and replace or automate inspirational experiences but add to them. In what may be a unique moment in human history, we have the opportunity to relate to and learn from something familiar yet unpredictable and with the capability of responding partly in our own means of expression but taking us beyond typical patterns of communication. GenAI may not be able to give us ‘truth’ as is currently desired but, to me, it is able to enhance our subjective worlds through the expression and investigation of our experiences.