For more than 30 years, science photographer Felis Frankle has helped MIT professors, researchers and students to visually communicate their work. During that time, he has seen the development of various tools to support the construction of hypnotic images: some auxiliary, and some opponents for the effort to produce a reliable and complete representation of research. Published in an opinion recently published Nature The magazine, Frankle, discussed the use of generic artificial intelligence (Genai) in images and discussed this challenges and implications to communicate research. On more individual notes, she asks whether the research community will still have a place for a science photographer.
Why: You mentioned that as soon as a picture is taken, the image can be considered as “manipulation”. There are methods that you have manipulated to your own images to create a scene that transforms the more successfully desired message. Where is the line between acceptable and unacceptable manipulation?
A: In a broad sense, the decisions made on the form of frame and composition of an image content, as well as the devices used to create image are already manipulating reality. We need to remember that the image is only represented by that thing and not that thing. Decisions have to be taken while creating an image. The significant problem is not to manipulate the data, and in the case of most images, the data structure is. For example, for an image that I made some time ago, I digitally removed the Petri dish in which a yeast colony was growing, to pay attention to the stunning morphology of the colony. The image has a morphology of the data colony. I did not manipulate that data. However, I always indicate in the text whether I have done something for an image. I discuss the idea of image enhancement in my handbook, “The Visual Elements, Photography”.
Why: What can researchers do to ensure that their research is communicated correctly and morally?
A: With the arrival of AI, I look at the three main issues related to visual representation: the difference between illustration and documentation, the morality around the digital manipulation, and the researchers are constantly needed to be trained in visual communication. Over the years, I am trying to develop a visual literacy program for the current and upcoming sections of science and engineering researchers. MIT has a communication requirement that mostly addresses writing, but what about the scene, which is no longer a tangent to present a magazine? I will bet that after reading most of the readers of scientific articles, the figures are correct.
We need to learn students how to see a published graph or image seriously and decide if something strange is with it. We need to discuss the morality of an image “nudging” to see in a certain predetermined manner. I describe an event in the article when a student replaced one of my images (without asking me) that the student wanted to communicate blindly. I did not allow it, of course, and disappointed that the morality of such changes was not considered. We need to make a visual literacy with a very least, at least, interaction in the premises and even better, with writing requirement.
Why: Generic AI is not going away. What do you see as a future to visually communicate science?
A: For Nature The article, I decided that a powerful way of questioning the use of AI in generating images was for example. I used one of the spread models to create an image using the following prompt:
“When excited with UV lights, make a picture of the nano crystal of Maungi Bavendi in vials against a black background, depending on their size, on different wavelengths.”
The results of my AI experiment were often images like cartoon-like images that can hardly pass as reality-give the documents-but there will be a time when they will be. In interaction with colleagues in research and computer science communities, everyone agrees that we should have clear standards and do not allow. And the most important thing is that a Genai scene should never be allowed as documentation.
But the AI-borne view, in fact, will be useful for the illustration purposes. If an AI-related scene is to be presented to a magazine (or, for that case, shown in a presentation), I believe that the researcher should be
- Clearly label if an image was created by an AI model;
- Indicate what the model was used;
- Include which prompt was used; And
- Include the image, if there is any, it was used to help in the signal.
