Toward better images of AI
Our network shadows. Artist: J Knowles
We all know the problem. Images representing AI in media and marketing are pretty terrible. I’m not the first to say it, but I have been grumbling about this for years as a tech writer and all through my AI degree. Just do an image search for AI and you will see what I mean. Mostly blue, lots of flying binary and maths, some shiny white robots that are their own category of problematic and maybe the odd depiction of a brain.
What’s the issue? AI doesn’t look like that. Fair enough, a panel of code might not float your boat or generate clicks to your wonderful article, but misrepresentation of AI is a problem. We all know there’s more AI in our lives, apps, home assistants etc. AI comes in many forms, machine learning, NLP, visual identification, networks that can generate art and writing. Mostly it is invisible though and making it an embodied character changes our expectations.
A note before we kick off here. Though I do find AI-generated images interesting, it’s not what I am going to bang on about. For clarity I mean the images we use to illustrate news stories, marketing and papers or factual websites with. That done, let’s get on shall we?
Good and Bad
AI, like any technology and duct tape — has a dark side and a light side and is holding a lot of our digital universe together. It is bad when it is fed crappy data that is biased, it is good when we can use it to screen for health issues faster than before. It’s probably not great if it is writing your academic papers for you, but it’s excellent at wading through acres of data faster than we can and leveraging our own creative abilities. It’s what you make it and how you use it. Which also means it’s not the Terminator.
Early sketches working on Better Images of AI. J Knowles
You’ll maybe notice I have not chosen to illustrate this piece with assorted damning examples of ‘bad’ AI art. I don’t think the art in itself is bad, the rendering and images are sometimes very beautiful and imaginative. At the same time, this is the work of highly creative people who have likely been commissioned to create those images.
I’m not entirely damning the outlets and publications that use hackers in hoodies, white robots and raining binary. Why? Because often there is no other option.
Those who are familiar, know I worked in newsrooms for decades, often writing about tech and needing to get news out and published. When the topic you are covering is somewhat abstract, you need a metaphor to illustrate it and this is what they are. A stock of images you can access quickly, use repeatedly and readers will know what you mean. You also want something stimulating. As so many of us receive our information and news from endless reams of headlines with images, editors naturally want images to stand out, so people will click and hopefully read. (Economics of news publishing for another time.)
Early sketches of people and their data networks. Descartes joke thrown in for free. J Knowles.
There is a sort of vernacular to AI images already that is embedded in many cultural norms. Blue and technology goes back a long way to early technology, it represents the new, novel, exclusive and hard to create for a number of reasons you can read up on for yourself. AI being a topic where one zenith is the achievement of human intelligence — or the appearance of it (see Turing Test) and we associate this with the robot, the golem and the humanoid. It’s not surprising, but it is a bit misleading. There are few common robots using human-level AI at time of writing.
Black Box technology. Embroidery — J Knowles.
We use this visual language to share an idea. I’m sort of ok with that, but I want it to be more accurate. I’ve wanted this for a very long time and I have found it somewhere between frustrating and infuriating to see that writers don’t get to choose the images, editors don’t have many options and image makers — especially in photography — are commissioned to create pictures that are not helpful. So, let’s change that.
I’ve started to work on creating alternative images of AI. As an illustrator, I can pretty much go anywhere my imagination takes me. In photography, I am somewhat stymied by the things I can take a photo of. Algorithms are not so photogenic. But I am working on that.
I work with the language we have for AI — black box technology, artificial intelligence, neural networks etc, and I make images that cross over between the actual language we use, the real imagery of working with these technologies and enough imagination to create images that are useful in publishing, marketing and elsewhere.
The good thing is, I’m not alone. I’m currently working with BBC R&D on solving the stock photography problem. I keep in touch with organisations like We And AI and those that are looking to create a better image repo to tilt the balance away from robots and binary. I’m having amazing conversations with practitioners, computer scientists, other artists and tech leaders, but I am always up for talking to anyone who has a vision in mind around this technology — one that is based on fact.
Embroidered neural network map. By J Knowles.
I have a question for you readers. If you work in AI, what do you ‘see’? Literally and figuratively. Are there motifs or images that come to mind? I can’t be the only mind working on AI visuals. There are a number of brilliant illustrators already working on tech pictures and we all embrace the abstract because we have to, but not so abstract that the technologies are not easily communicated.
If you are an artist that makes images of AI, even better a photographer going for the harder problem of AI visuals, or if you are an AI practitioner with a visual sense of what you do, I’d love to hear about it. Drop me a line here.
Bad images of AI don’t help people learn about what this technology is, how it works or what it means for them — good or bad. AI has been here long enough, so let’s work out better ways to visually describe it.
