For a long time, scientific photography was based on a simple idea: the microscope would capture what was happening in front of its lens, and then the researcher would try to process the image to make it clearer and easier to read. But the introduction of artificial intelligence into this process has begun to change the equation, after algorithms became capable not only of reducing noise and improving details, but also of reconstructing parts of the image and inferring what might be in it.
Hot Topics
These new boundaries became evident in the recent controversy surrounding a winning video from Nikon's Small World in Motion competition, after researchers raised questions about how accurately it represented what the microscope actually captured.
After the result was announced, microscopy researchers raised questions about some of the details shown in the video, saying that structures appeared with shapes and behaviors that are difficult to explain biologically, while others pointed out that some elements seemed to appear and disappear within the scene.
But that doesn't mean the video was proven to have been AI-generated. Shaw explained that AI wasn't used to generate the original video, the eyelashes, or their movement, but was used later to process grayscale microscopic data, highlighting and coloring similar structures to improve the visual presentation.
Where does improvement begin and where does it end?
The problem here is not with the use of artificial intelligence itself. Algorithms have become an important tool in scientific and medical imaging, and can be used to reduce noise, improve accuracy, correct motion, and reconstruct images from incomplete data.
But there is a fundamental difference between showing information that is already present in the data and adding details that were not present in the original measurement. When the algorithm raises the quality of a low-resolution image, the output may appear clearer than the original, but this clarity does not necessarily mean that all visible details were actually captured by the microscope. The problem becomes more complicated when the algorithm is able to produce compositions that appear convincing to the human eye, but are not based on sufficient experimental data.
Recent research in the field of medical imaging warns of this type of visual hallucination, as image restoration algorithms can add non-existent structures or delete real structures without the difference being easily noticeable to a human observer.
Tags:
technology