For years, AI has been represented and branded as an illuminating entity from product interfaces to marketing and culture. Glowing orbs, sparkling icons, radiating brains, glistening robots, looming horizons; a visual shorthand for the presence and promise of machine intelligence. An aesthetic that is ambiguous by design and heavy on metaphor; creating the sense that AI is elusive, magical, powerful and beyond-human. 

AI is visualised according to the agendas of those making – and paying for – the images: brands selling generative models with open scopes, media outlets chasing engagement, artists questioning and critiquing. We can't disentangle the visual elements from the veins that connect them to their creators and context.

I’m interested in exploring the aesthetics of AI and why we currently need more pluralist visual narratives and cultural representations.

One way that I’ve been thinking about this is the paradox of illumination. When we illuminate, we either shine a light on something to make it brighter, or we explain something difficult so it can be understood more clearly [1]. There’s an opportunity to shift power away from a few big companies showing AI as an abstract, luminous entity, towards diverse communities creating images that actually explain it: the maths, the infrastructures and the effects on people and the planet.

The problem with the current narrow visual landscape isn’t just the omnipresence of luminous forms, it’s the singularity of the approach. A handful of dominant voices set the majority of narratives, so a lot of AI product and marketing images look fairly similar. The aesthetic comes across as neutral and broadly appealing but it’s actually hyper-local. Much of the industry is funded by or coming from the privileged few in a handful of cities in the U.S., and the visual culture reflects their shared reference points, sets of values and world views.

AI is way too multifaceted to be imagined as inaccurate or noncommittal forms that all feel like they’re derived from a similar source. It’s a sprawling field of tools and methods applied across countless contexts [2], at a scale so vast we can only ever experience individual pieces of it. And a good deal of what AI is doing to society takes place inside our minds. What does cognitive atrophy or the fondness towards an LLM even look like? As film-maker Adam Curtis put it: “if you want to make a film about computers, there's absolutely nothing to film.” [3]

At its core, AI is intricate, high-dimensional maths, which is not straightforward to make sense of, let alone make beautiful or compelling. Of course, it’s possible to show the physical containers of that maths: phones, laptops, drones, robots, self-driving cars, security cameras. But these only get you so far.

AI isn’t completely invisible though, it has physical forms; they’re just not enchanting and many are frankly depressing. Chips etched from silicon. Data centres made of concrete and steel. The land flattened to make space for them. The water that cools the systems. The millions of people who rate the vast datasets scraped from the internet and moderate content. None of it is impossible to document, it's just not commercially desirable to do so because the harder AI is to understand, the more magical it appears. 

One 2025 study found people with lower AI literacy were more likely to experience awe at machines doing things that seem uniquely human [4]. Awe is a politically-charged emotion. It can foster wonder, creativity and collaboration, but it can also be an instrument of domination. Psychology professor Dacher Keltner defines awe as “the feeling of being in the presence of something vast that transcends your current understanding of the world” [5] and while talking to an LLM shouldn’t be compared to seeing a waterfall in real life, the vastness and magic-ness of AI is exactly what branding and headlines are selling. 

Visual culture compounds the aesthetic narrowness of AI branding with a similarly limited set of go-to metaphors, such as the prolific use of glowing blue brains, white hairless robots and Creation-of-Adam hands, reinforcing that AI is otherworldly, powerful, white and masculine. Dr Kanta Dihal and Tania Duarte analysed this landscape via Better Images of AI, which is the home to a library of artwork that moves beyond the problematic tropes. 

Commercial symbolism is also shaped by science fiction, in a cycle Dr Dihal and Stephen Cave call the Californian Feedback Loop – “the co-creation of narrative and technology between Hollywood, Silicon Valley, and academia.” [6] A handful of Anglophone films, like Terminator, Her and Ex Machina, have significantly shaped not just the public perception of AI but how it’s been researched, deployed and regulated. Visual choices matter because they have an effect on the mental models we all have about AI.

Whatever metaphor a brand reaches for, it needs to be appropriate. A Color Bright's analysis of 23 AI brands earlier this year observed companies leaning on the natural world to lend "organic pleasantness to a potentially terrifying technology". Yet scenes of nature feel a little on-the-nose when huge amounts of water, minerals and energy are used throughout the AI lifecycle [7]. Likewise, relying on warm neutrals, serifs, matte paper textures and hand-drawn sketches that borrow the visual grammar of publishers and universities comes across as insincere when we hear about rare books being destroyed en masse after being scanned for AI training data. [8]

However, a barrier to pluralism is access. Creatives don't need to make work with AI to make work about AI, but they do need access to people, tools and information that are typically hidden inside the “black box”. Of course there are artists like Trevor Paglen and Anna Ridler who have been working in this space for years and have extensive technical knowledge but many illustrators, animators, photographers, film-makers and artists don’t have that expertise. When they can get closer to the technical world beyond a prompt input box, and can talk to engineers and research scientists who are able to translate what’s going on, we may see more imagery sparked by the machine learning process itself. 

Subfields such as computer vision, for instance, are inherently picture-based and lend themselves well to creative interpretation. In 2020, my colleague Emily, a software engineer, showed me how a machine learning model perceived concepts inside a collection of photographs. The model's perception appeared as heatmaps – hot patches of colour marking where "sharpness-ness" or "orange-ness" lived in each image. As a visual thinker, interpretability techniques like these invited me into an intimidating world that I didn't feel like I belonged to because they translate the statistical patterns of a high-dimensional space into a language of colours, edges and light. I also found the aesthetics beautiful in their own way – lurid, sometimes glitchy and entirely functional.

Unlike the un-designed interpretability techniques that Emily showed me, another colleague, Shan, added a very conscious layer of visual communication to the technical diagrams and interactive descriptions he created on distill.pub, a scientific journal that ran from 2016-2021. Shan translated maths in the form of vectors into fascinating visual stories and one of my favourite articles was Activation Atlas, where he and his co-authors visualised millions of activations from an image classification network to reveal how a neural network represents concepts. Even today, when I think of the internet soup that powers generative AI models, the mental model I have in my head is based on his trippy maps.

It’s too easy to say that image-makers should only work with big tech or startups in order to have access to the “black box” and create work about what they experience, though there are programmes such as Artists + Machine Intelligence at Google that do this well. Independent organisations also provide the space, time, budget and access to tech that fosters cross-disciplinary projects free from commercial agendas. These include AIxDesign, NewInc, LACMA Art + Technology Lab, LG Guggenheim Art and Technology Initiative, BRAID, Serpentine Arts Technologies, Somerset House Studios and Mozilla Foundation’s Creative Futures to name a few. The more structures like residencies, embedded artists and paid fellowships, the better, because it’s not only films that shape our mental models of AI, but culture at large. 

This is where AI can feel awe-inspiring (in a good way) again: the cultural interpretation space that is close to the technical realities rather than fictionalised. For example, the partnership between Kate Crawford and Vladan Joler sits between tech research and visual communication, translating what is usually obscured into something people can query and critique. Their Anatomy of an AI System (2018) plots the entire lifecycle of an Amazon Alexa including the mined materials, the supply chains, the human labour, the technical infrastructure. They make the invisible veins behind AI products visible and have continued to build on this work in Calculating Empires (2023). 

If frontier AI labs continue to develop huge models with no specified purpose in mind, it’s likely that the aesthetics of mainstream AI will remain on the more abstract, metaphorical and vague side. My hope is that as small, carefully scoped, purpose-built, community-owned and open-source models become the norm, we’ll see the emergence of more diverse and transparent visual narratives because there is less to conceal and more to be proud of.

Genuine illumination and the “good kind of awe” isn't just a property of images, it's about relationships between people who are in the “black box” and artists, creatives and designers who can visually translate what’s happening from multiple perspectives. It’s time we introduce new feedback loops.