Nikon disqualifies its Small World in Motion winner over generative AI use

[Robin13]

We have followed this one for a couple of weeks, and now it has an ending. Nikon has ruled that the winning video in its 16th Small World in Motion competition broke the contest rules on generative AI, and it has removed the entry from the winners page. The video, credited to Ning Xu, showed cilia moving in airway tissue from a child with primary ciliary dyskinesia. Red, blue and purple structures appeared beneath the cilia, and those were what caught scientists' attention

The doubts built up quickly. Edward Phelps of the University of Florida said on LinkedIn that purple structures seemed to pop in and out of existence. He added that a structure made of mitochondria the size of cell nuclei does not occur in biology. Ian Donovan, an MD/PhD student at UT Southwestern, ran the video through Google Gemini, which reported a SynthID watermark, Google DeepMind's invisible marker for AI generated content

Xu's position was that the motion in the video was real. He said AI was only used on reconstructed greyscale data to tell apart and colour structures with similar shapes. Nikon's own blog later said an unsupervised neural network method had been used as part of the post processing and visualisation workflow. The contest rules say plainly that AI generated videos are not permitted

After re-evaluating the video and supporting materials and consulting its judging panel, Nikon decided the entry did not comply with the rules. It stressed that the decision is based solely on the video's eligibility under the competition rules. The new winner is Nguyen Nam Nhat of Vietnam, whose video shows a tiny roundworm and a single celled Dileptus. Nikon also says it will revisit its rules and evaluation procedures

I think Nikon made the right call, even if it took a while. Scientific imaging competitions only mean something if people can trust that what they see came from a real measurement. Melanie White of the University of Queensland put it well, saying scientists need to be able to trust that what they see is grounded in the underlying measurement. The interesting question now is where the line sits between normal image processing and AI generation

Did Nikon get this right? And where would you draw the line on AI tools in scientific images?

AI disqualification yields new Nikon Small World in Motion winnerarstechnica.com