How the AI ​​fantasy ran wild with a rat’s penis

How the AI ​​fantasy ran wild with a rat’s penis

A picture of an erect rat with an erect penis that literally transcends the animal, right through the roof of the illustration, and a scrotum that relates to its body like a skippy ball relates to a human – that can’t be true, can it?

No that is not possible. And it’s not allowed either. Perhaps in an educational picture book, in which a disproportionate magnification can help to provide insight into the reproductive apparatus. But it is not without reason that pictures in scientific publications are often so boring. Exaggeration and science simply don’t go well together.

The grotesque penis was not the only hallucinatory illustration in the article about rat sperm stem cells that was published in mid-February in Frontiers in Cell and Developmental Biology and had to be withdrawn again a few days later. Pictures of cookies with blue M&M’s (cells), and a kind of goose plate of signaling molecules and receptors further brightened up the literature review.

With the naked eye

Scientific illustrations often deviate from reality. Think of the coronavirus that is printed in all sizes and colors and cannot be seen in real life with the naked eye. But the three Chinese authors made it very colorful. Or rather, they let their AI program run wild.

They were not secretive: they neatly wrote down that the images were generated by Midjourney. Using Artificial Intelligence (AI) is also allowed Frontiers, but then there shouldn’t be a storyboard for a fantasy film. And it is wise to check the spelling, so that no ‘sigal’ molecules and ‘iollotte sserotgomar’ cells roll out. (And what would a ‘dck’ mean in this context?)

You would think that pictures this imaginative would immediately arouse the suspicion of the independent experts who checked the article before publication. But online magazine Vice quoted one of them last week as saying that he “only looks at the scientific aspects”. He believed that images created by AI were the publisher’s responsibility.

Take action immediately

Frontiers quickly withdrew the article and tried to turn it into good news: thanks to the public’s involvement in ‘open science’, they could take immediate action. According to Frontiers one reviewer had expressed his concerns, but the authors had done nothing about them. “We are investigating how our process failed.”

Frontiers doesn’t seem to be planning to ban AI just yet, unlike a number of other titles. So wrote Nature that there is no integrity and transparency as long as AI programs do not provide access to sources and which therefore cannot be verified. The use of AI to create images is therefore not permitted.

The question is: how do you control it? Image tampering was so common before AI that magazines themselves started using AI to detect it. But although simple tampering comes to light more often, writes Natureit is not always easy to recognize AI fakery with AI.




SCIENCE