Engaging with new technology as an artist often feels like a double-edged sword. On one hand, these tools offer new modes of expression, one that could only be expressed through the media specifies. On the other, the very act of creating with them can lead to the practices being reabsorbed into the consumer ecosystem. As Simon Penny noted in 1995: "Artists who engage these technologies also simultaneously engage consumer commodity economics."[1]
The recently opened gym had adapted video screens embedded into treadmills, it displays real-time simulations of running through Google Maps-style environments. It reminds me of Jeffrey Shaw's The Legible City (1989)[2] where as participants use a stationary bicycle to navigate through a virtual cityscapes composed of three-dimensional letters. While The Legible City doesn't involve a treadmill per se, its concept of physical movement through virtual environments parallels the installations in a gym 30 years later. Or even take TerraVision by ART+COM[3], a 1990s project that let users explore the globe virtually. Years later, the artists sued Google, alleging it became the foundation for Google Earth. Art and technology have long been intertwined, but for those working with emergent tools, especially in the digital and computational realms, there's a recurring pattern: certain innovations loop back into commodification, and commodity changes the way to work.
Back when OpenAI published its first version of DALL·E, it outputted strange textures and glitched compositions. Those images felt uncanny, raw, and alien in a way that today's clear outputs have polished away. I felt the excitement to work more on it, and yet those models are not accessible anymore. Working with these tools means entering a feedback loop. As the software changes due to the need for better quality for further commercializing, the expectations also shift gradually. It limits how we perceive tools and what kind of output we are able to generate.
Furthermore, as production progresses, there is also demand for new hardware. I started out working with electronic art using Arduino, a lightweight, open-source microcontroller platform great for prototyping and physical interaction. You write some code, wire up a few sensors or LEDs, and suddenly, the idea takes on a physical form. It was affordable, hackable, and felt empowering. Eventually, I shifted to using Raspberry Pi, which is a single-board computer with full operating systems capable of handling more complex tasks, including interactive videos with sensors. It gave me more flexibility and stability, yet cost not only more price but also more time in order to understand how to program on a Linux interface. Now, with the demands of working with AI, especially image generation or training models, I'm being pushed into yet another tier of hardware: buying new graphics cards (GPUs), a power-hungry and expensive product that is necessary to work with generative tools.
The pace of updating GPUs is even more rapid; the cycle of upgrades can feel like a hamster wheel. The earlier AI images, like generative adversarial networks, have completely different requirements than those used with transformer-based tools. Any new updates on software or framework could cause the old hardware to become obsolete due to its lack of computational power. Niklaus Wirth, the creator of the Pascal programming language, articulates this observation known as Wirth's Law, that "Software is getting slower more rapidly than hardware becomes faster."[4] He highlights the phenomenon where software's increasing complexity and resource demands often outpace the advancements in hardware capabilities.
This imperative to upgrade isn't just about budgeting; it also reflects a deeper contradiction. The technology I use to critique or explore is embedded in the systems I might be questioning. As an artist, I must confront the irony that my means of expression can undercut my message. These tensions led me to a surprisingly basic but essential question: What even is a graphics card?