Amidst Words and Wires – Discussing the past, present, and future of AI image generators in Poetics of Prompting at MU
The group exhibition Poetics of Prompting at MU Eindhoven delves into the world of prompting generative technologies: the process of asking machines something in a way that gets you the information or image you’re looking for. Chatbots like ChatGPT and image generators like Midjourney have popularised this mechanism by making AI available to the public. But what gets lost in these artificial translations?
Poetry can produce effective allusions to complex concepts, by navigating through the constraints of language. Nowadays, prompting AI generators similarly reflect grammatical and literal restrictions. It is urgent however that we consider what is lost in translation when we aim to share a language with the disembodied. What can we learn from the poetic urge for allusion when machines leave no space for metaphor? How does a manifestation of human experience effectively translate into statistical code? And what should we ask AI in order to find out? From 4th October to 24th November, The Hmm & MU Hybrid Art House investigates the fringes of these literary obstacles in their ‘crash course in speaking machine.’ In the former factory building, the exhibition Poetics of Prompting brings together 21 artists and designers eager to inform, innovate, and iterate at the intersection of art, language, and technology.
The first work in the reception area is a 20-minute documentary by Alan Warburton called The Wizard of AI (2023). The frames brightly shift as the narrator discusses the past, present, and future of AI image generators and the nature of what they produce. An especially memorable ‘scene’ is cut together from animations of strange digital fish on scientific slides and other fish swimming in 2D fish bowls. They are of a variety of colours and shapes and some even feature familiar logos among their scales. They look like an unnatural amalgamation of different fishy features that have come together to produce a ‘fish-like’ image. These digital animals introduce the idea of the ‘dredge’ which Warburton uses to explain what AI companies like Midjourney did to create image generators. Like a net in the ocean that is unbiased in what it picks up, these programs made a ‘pixel soup’ from user-generated images found across the internet. The introduction of the ‘super fish’ imagery, an image that is cut up from and reformed from other images available on the internet, cuts right to the point: any type of fish imaginable can now be generated. It is bright, bizarre, and available in every style, ‘if not a little dead behind the eyes.’ These fish are likened to an invasive species in a sea of images, which has turned into a swamp of synthetic life. This entry claims to be the first of its kind—almost entirely made using AI—but unlike some of its contemporaries, such as the Dead End Gallery’s The Most Successful Organism (2023) the depth of references and the humour of the narrative make this mesh of messy images stand out in this exhibition, ultimately emphasising the humanity of the themes it presents.
Thema's
Moving to one of the larger rooms, another projected video is surrounded by various types of chairs. This piece, called Sorting Song (2021) by Simone C. Niquille, features children’s voices narrating over visuals of household objects fluctuating on screen. The physical arrangement of chairs paired with the video illustrates the challenge of categorising even the most common objects. A child’s voice echoes through the space: ‘Where does the cup end and the vase begin?’ It highlights the challenges that come with the literal language interpretations machines are programmed to process. How can data distinguish a wheelchair from an office chair? And how does this affect the robots programmed to interact with humans?
One of the highlight pieces that centres on the direct ‘prompting’ of the exhibition is Prompt Battle Training Station (2023) by Sebastian Schmieg. In this installation, guests can sit down and see if the results of their prompts replicate the images displayed on the screen. The challenge escalates as the timer counts down, with a futuristic figure on the screen offering instructions and feedback. Participants can add their names to a scoreboard, showing how they performed compared to other guests. This piece mimics prompt competitions, offering a taste of the peculiar yet trainable skill of ‘speaking’ the language of machines—i.e. coaxing this data system filled with randomness to produce the desired results.
How can data distinguish a wheelchair from an office chair? And how does this affect the robots programmed to interact with humans?
As visitors navigate the industrial exhibition space, they find the last piece on display, which is particularly amusing. A small square painting, approximately 25×25 centimetres, depicts what looks like a vacuum on its side at the bottom of some stairs. It is titled Robot Assistant (2024) and is placed next to a rectangular screen connected by wires to an exposed motherboard, positioned just above a looping video. This video, What Do We Dream About When We Dream About Machines (2024) by Mariana Fernández Mora, features a dramatised script asking questions like ‘Do they dream of electric sheep?’ and ‘Maybe he wanted to escape the same things we do?’ Meanwhile, the screen flashes with clips of Roombas (small vacuums that roam your house) attempting to suck up rice and zoom around in circles. The pairing on this wall is wonderfully absurd. Even though machines do not dream, nor have desires or seek freedom from labour, their wants and needs seem to speak to us. This work evokes two notions which become increasingly unavoidable as one watches the video. Firstly, it is worth reflecting on the vulnerability we attribute to machines, considering the resources we invest in them; it prompts us to consider what is revealed about ourselves when we ask machines to do tasks we do not wish to do. Secondly, it suggests that we should care for these machines because they, in turn, assist us—not because they are sentient, but because they make us better. They are helpful and sometimes even amusing. If we take care of them, they will last longer, and we can enjoy not only their service but the delight we feel when watching a Roomba contemplate jumping a ‘cliff’ (the stairs).
All in all, this exhibition offers many wonderful insights. It balances both information and allegory well, touching on intriguing trends and proposing answers to common questions such as ‘What happens after we establish a statistical core for a style? Is an anti-generalisable, anti-core possible in aesthetics? What happens to human art in this proliferation of AI-generated images? Do human-made images become more valuable because they are unlike the statistical ‘Frankenstein’s’ churned out by image generators?’ These are warranted and relevant explorations. Nevertheless, some of the works are admittedly more convincing than others, especially when considered alongside the Poetics of Prompting title of the exhibition. The works where dramatised metaphors and lyricism are used to create “deep” expressions of pseudo-philosophical questions can become tiring, especially in an exhibition that has set the stage for an effective exploration of the importance of a common language between AI and human realities. Trying to find ‘more’ meaning does not garner ‘more’ understanding; the work needs to stand on its own, regardless of moody stylistic choices. Regardless, through the large selection of works, most pieces are original, and some are organically interactive. Even literature about AI, creativity, and language is available for all to read. As one leaves, more knowledgeable about the language of AI and the nature of human-machine relationships. Poetics of Prompting proves itself to really be ‘a crash course in speaking machine.’
The exhibition Poetics of Prompting can be visited until the 24th of November at MU Hybrid Art House
Julija Zagurskytė
works as a Junior UX writer for a tech company in Amsterdam and recently completed both a Bachelors and Masters thesis on the topics of AI generated images and AI art; in other words she writes and she writes and she writes.
