Together they made me think harder than I expected about a question that gets asked too loosely: what is the impact of AI on the field of scientific visualization?

The wrong question

I’ll be honest: the debate whether scientists and designers should use AI tools or not isn’t that interesting to me. AI tools are already assisting researchers with illustration, and that’s not going to reverse. The interesting question isn’t whether, but where, and how, and how much oversight is needed.

Kuukua Wilson, a medical illustrator interviewed in the piece, recalls a professor’s line from her master’s degree:

“People used to worry that illustrators would be replaced by the camera. But if a camera takes a photo of a surgery, you don’t see anything, just blood. You still need a medical illustrator to draw everything else.”

Illustration was never actually competing with photography. It does something photography can’t: show what’s obscured, too abstract, or not visible.

I think that’s the right lens for AI, too. Not “will it replace illustrators” but “what is illustration actually for, and where does that purpose hold up against a tool that’s extremely convincing, and occasionally wrong”.

Proof versus explanation

Here’s my take. Some scientific images exist to prove something — a photograph, a microscopy image, a western blot, a data graph. Their entire job is faithful representation of something that happened or was measured. Other images exist to explain — a graphical abstract, a flowchart, a timeline, a schematic. Their job is to make something understood, not to serve as evidence for it.

I’d never use AI to touch a proof-image. The risk of distortion, even the most subtle, is exactly what we can’t afford when the image is the evidence. For explanatory images, though, such as flowcharts or timelines or illustrations, I think AI can be a genuinely good assistant. In those situations, a human is making deliberate design decisions and oversight is straightforward.

“When talking about AI-generated images, people usually think of art and illustrations. But scientific graphics can also include schematics, visual abstracts, figures that depict a study’s data and images used as evidence. The ethical issues are very different, if you’re using images as evidence or using them to explain things.”
— Ethicist Sebastian Porsdam Mann

This is close to what the Nature technology piece calls out directly: the ethicist Sebastian Porsdam Mann notes that the ethical questions differ sharply depending on whether an image is being used as evidence or used to explain. The clearest cautionary tale for the evidence side is well-known: in April 2026 a biologist generated a full set of convincing-looking western blots from a single ChatGPT prompt, just to demonstrate how easy fabrication has become. Another paper was retracted from the New England Journal of Medicine after an AI-adjusted image left behind a telltale error in a tape measure’s numbering.

So, what should you do? Are AI tools allowable when creating scientific visuals? It’s good to know that publishers don’t agree on any of this. PLOS allows AI-generated images with disclosure. Cell Reports bans AI graphical abstracts outright. Springer Nature (Nature’s own publisher) prohibits generative AI images except when an article is specifically about AI. So whatever your own ethical line is, check your target journal’s policy before you touch a figure with an AI tool.

And what about graphs?

The one place I disagree with the article is graphs, of course my favorite subject. Marc-Oliver Gewaltig, co-founder of the academic-writing consultancy Thesify, suggests plotting your data yourself, then asking AI to make the figure more polished. His argument is that you already know what the chart should look like, so validation is easy.