Absurd image of rat with XXL penis sows confusion among scientists

Artificial Intelligence – This would be funny (and a little embarrassing) if it wasn’t the emerging surface of a bigger problem for the global scientific community. On February 13, research results on fertility were published in a specialized journal, Frontiers in Cell Development and Biology. A completely serious study, validated by a reputable publication… but which apparently doesn’t look closely at the pictures with the research.

Thus the main representation is of a mouse whose genitalia we want to show… and which ends up with a penis larger than that, the image has no logical meaning. Worst of all, the terms that usually define parts are illiterate, a series of letters that don’t quite make up a word. what happened ? The answer is not mysterious, as it was confirmed by the authors of the study: the image was created by artificial intelligence.

Testomcells, Retat, DCK

The researchers actually left it to Midjourney, the most famous graphics AI program (which creates images based on cues given by text commands), to create the illustrations for their study. Not just this rat with a disproportionate penis: all the drawings, if you look a little closer, are a strange assembly of visuals that, from a distance, pass for scientific illustrations, and closer as obscure puzzles.

funny? Undoubtedly. A scientific embarrassment for the journal that published these pictures? absolutely. It is also worrying that such errors could pass before the reading committee, and many scientists who have responded to this publication have made this clear. On his blog, Dutch expert on scientific integrity A “Tragic Example” The naivety of journals towards “AI Generated Shit”.

It must be said that this anecdote vividly raises a problem which the scientific world is grappling with as acutely, if not more, than elsewhere. The mass influx of AI-generated images must be controlled, and scientific journals do not all have the same policy… nor the same rigor in identifying them. When reviewed Nature

It is already known that it will not accept any content created by AI, Frontiers in (which is a complete list of journals specialized by field) allows it, provided the nature of the image is clearly shared with the reviewers and specified in the study.

The case of our virilized mice shows above all that this rule rarely works because of insufficient attention to the material involved in the study. A flaw that can significantly damage the credibility of these scientific journals: if the latter, whose accuracy and rigor are their core business, allow such flaws to pass, what should we expect from the studies they publish themselves? ? No response from Frontiers in Cell Development and Biology, But the top of the study mentions that improvements are ongoing…

See also on The HuffPost: