Nikon is re-examining a winning entry from its 16th annual Small World in Motion contest. The submission, by National University of Singapore engineer Ning Xu, faces scrutiny over potential AI involvement following intense social media debate.
The "impossible" colors in Ning Xu’s lung tissue video
Edward Phelps, a bioengineering researcher at the University of Florida in Gainesville, has raised significant concerns regarding the biological accuracy of the winning entry. The video, which depicts the movement of cilia in lung tissue from a child with primary ciliary dyskinesia, contains red, blue, and purple structures that Phelps argues are biologically impossible.
According to the report, Phelps noted that while the blue nodules resemble nuclei, they do not behave like them , and the purple structures resemble mitochondria in a way that does not occur in nature. These unidentified structures have led some observers to question whether the visual data represents actual biological specimens or AI-generated fabrications.
A Google DeepMind SynthID watermark triggers scrutiny
A Google DeepMind SynthID watermark has becoome a central piece of evidence in the debate surrounding Ning Xu's video. Ian Donovan, a PhD student at UT Southwestern Medical Center, identified this digital signature within the footage, which is a technology used by various AI firms to denote synthetic content. As the report notes, detecting AI-generated imagery is often nearly impossible without such specific watermarks or equivalent technological markers.
Ning Xu’s defense of post-processing grayscale data
In response to the backlash, Ning Xu told Nature that the fundamental motion of the cilia in the video is real and captured through microscopy. Xu maintains that artificial intelligence was only utilized during the post-processing stage to colorize and distinguish structures within reconstructed grayscale data.
Nikon has stated that Xu is cooperating with their investigation by providing technical documentation that details his specific imaging methods and the microscopy equipment used. The company previously clarified that an unsupervised neural-network method was used as part of the visualization workflow to make the grayscale data more visually engaging.
The struggle between artistic colorization and the 2010 contest rules
The dispute highlights a growing tension between artistic colorization and the historical rules of the Small World in Motion contest. Former judge Andrew Moore pointed out that while the contest allows for colorized images , it strictly prohibits the use of generative AI to produce content. Melanie White, a developmental biologist at the University of Queensland, emphasized that scientific images must remain grounded in actual measurements to maintain professional trust.
Who will decide the final verdict for Nikon?
Nikon's investigation remains ongoing as several critical questions linger regarding the legitimacy of the winning entry. It is currently unclear whether the neural-network methods used by Xu's team crossed the line from visualization into prohibited generative creation.. As reported by Gizmodo, contest organizers have not yet provided a definitive update on whether the entry will retain its title.
Comments 0