Let me ask you a question: would you like to see your work featured in major publications like The Guardian, VICE, or even on my favourite comedy show, The Late Show with Stephen Colbert? Well, there’s a good way to achieve that kind of attention, and then there’s the very wrong way—just ask the researcher whose paper included a rather questionable AI-generated image.
When AI Goes Wrong: The Tale of the Endowed Rat
Imagine this: you're flipping through the latest peer-reviewed paper, and you stumble upon Figure 1—a rat.
But not just any rat. This rat has a GIANT penis hanging from the ceiling. Yes, you read that right. This bizarre image was an AI-generated monstrosity that somehow eluded peer review and got published in a Frontiers journal. The rat’s face was the only part that looked somewhat right; everything else was a visual disaster. Even the text labels were gibberish (Zoom in on the image, it’s worth it 🤣).
This paper didn’t stay up long—just three days, to be precise. But in that short time, the rat went viral immediately, the media caught on, and it was downloaded 200,000 times, making it perhaps the most popular research paper of the year—but for all the wrong reasons.
I’m sharing this cautionary tale to warn you about the potential pitfalls of AI, especially in its current state, and to highlight where its limitations lie—this wasn’t even an isolated episode; I've noticed similar gibberish AI images in multiple peer-reviewed papers.
So, is this AI thing useful for anything in visual science communication? Let’s see…
AI and Visual Science Communication: A Double-Edged Sword
A few words about myself: I’m Dr. Tullio Rossi, founder and director of the science communication agency Animate Your Science. We’ve been making science animations and infographics, for seven years now. Like many, I’m fascinated and a bit scared by recent developments in AI, particularly in image generation. But as with anything new and shiny, it’s essential to proceed with caution.
AI-generated images are often made for fun—think of those viral images of the Pope in a puffer jacket.
But beyond a bit of entertainment, is AI really useful for anything science communication related? That’s the question I set out to explore.
A Successful Experiment
Take this example: Inspired by my wife, a microbiologist working on superbugs, I decided to illustrate a “biofilm castle” where bacteria are protected from immune system attacks and antibiotics. I turned to Dall-E, one of the leading AI image generation models, and within seconds, it produced an image that blew me away.
The bacteria looked like little monsters, the walls were slimy, and the antibiotics were outside the castle walls, just as they should be.
This success made me think: could AI be a game-changer for our work? Well, yes—but…
The Limits of AI: Beware of the Details
While AI can be useful for creating simple objects or conceptual scenes like the biofilm castle, things get dicey when you ask it to generate more complex images. I’ve experimented with it for various projects, including storyboards for an animation featuring a black seabird. After much trial and error, we were able to piece together a workable concept.
However, when I asked AI to create something more intricate, like the structure of a DNA double helix, it failed spectacularly.
The problem is that AI doesn’t actually understand what it’s creating. It’s merely approximating based on reference images from its training data. This lack of understanding leads to visually striking but fundamentally flawed images. Sure, they look good at first glance, with vibrant colours and beautiful details, but upon closer inspection, you’ll find that everything starts to fall apart.
This is dangerous, especially when the images are created by people who cannot judge whether they’re anatomically or structurally accurate. The risk of spreading misinformation is real, and it’s something we should all be cautious about.
The Issue of Copyright: Who Owns AI-Generated Content?
Another critical consideration is copyright. In the Western world, the act of prompting for AI-generated content doesn’t entitle you to copyright ownership. This means that anything you create using AI is automatically in the public domain. While this might not seem like a big deal at first, it has serious implications for published research, books, and any other content where copyright matters. Interestingly, to complicate things further, in China, the situation is the opposite—prompting does entitle you to copyright ownership of the output.
So, if you’re planning to use AI-generated images in your work, remember that you can’t protect it from being used by others. This lack of control could have significant consequences for your intellectual property.
Interested in learning more about the legal side of things? We’ve interviewed a lawyer who understands AI-related issues, so be sure to check out that post for valuable insights.
Co-Creation: The Future of AI in Science Communication
So, what’s the bottom line? AI isn’t going to replace scientific illustrators. Instead, it’s more likely to enhance their capabilities. The way forward isn’t simply writing a prompt and letting the AI do all the work—where’s the satisfaction or sense of ownership in that?
Also, what separates a science illustrator from other illustrators isn’t just aesthetics; it’s precision, clarity, and a deep understanding of the subject matter—three skills that AI is completely incapable of mastering.
What I think will be more fun and lead to high quality outputs is to co-create with AI. For example, there’s software like KREA AI that allows you to combine your sketches with a written prompt.
This collaborative approach could be the future, though it’s still not perfect. AI struggles with fine details and can make critical mistakes, especially when it doesn’t understand the subject matter.
Embrace AI, But with Caution
My final message to researchers, scientific illustrators, and science communication professionals is to embrace AI. Like it or not, it’s here to stay. But as you incorporate it into your work, be mindful of its limitations and pitfalls. You don’t want to end up in the spotlight for all the wrong reasons—like the researcher with the endowed rat.
Thank you for reading, and stay tuned for more on this topic. This is just the first in a series of articles I’m planning to write about AI and science communication. If you’re interested in learning more, I also run workshops on AI for science communication. Whether in person or online, I’d love to bring this workshop to your institution. Just follow this link to get in touch.
Stay curious, and keep creating!
Comments