A photograph should not become a guessing game
AI-assisted tools are quietly becoming part of the software photographers and editors already use. The important question is less whether photojournalism ‘uses AI’ and more whether we can still say exactly what has happened to a photograph.
I spoke with Aikakausmedia’s Anniina Nissinen about the limits of AI-assisted image editing and what those limits mean for trust in journalistic photography.
The conversation followed a case at Helsingin Sanomat in August. The newspaper had published an Instagram carousel about students’ clothing on the first day of the school year. One of the photographs had been processed with a tool meant to partially cut subjects out from their background for the social-media layout.
In the published image, however, two people had lost their shoes. Shadows had disappeared from the asphalt. A climbing frame and a clock on the school wall were gone. Windows had changed, new asphalt had appeared and people in the background had been distorted.
After questions arose on social media, Helsingin Sanomat removed the carousel and republished it with an unaltered photograph. The newspaper later explained that the cut-out tool had gained a new generative AI feature over the summer. The employee using it had expected it to work as it had before and had not realised that its behaviour had changed.
Nobody sat down and decided to generate a fictional version of the scene. A production tool had changed underneath its user, and the result crossed a very basic line in journalistic photography.
My basic position is simple: a journalistic photograph should not gain things that were not there or lose things that were. Context matters, though. Extending a studio backdrop by a few centimetres is different from changing someone’s appearance, or from altering a photograph that might serve as evidence of a war crime. In the latter case, even seemingly harmless editing will reduce the photograph’s evidentiary value.
Generative tools make this particularly difficult because their interventions can be opaque. The more convenient, fast and automated we make the editing process, the easier it becomes to miss what has actually changed. If a tool used to remove sensor dust also quietly smooths spots from someone’s skin, the photographer may no longer fully know what has happened to the photograph.
Simply saying that a human remains responsible for the final image is therefore necessary, but not sufficient. If software can quietly move from selecting or masking existing pixels to generating new ones, newsrooms need to know when that happens. They need clear rules for AI-assisted image editing, control over the tools entering their workflows and processes for checking what has actually happened to an image before publication.
I don’t think increasingly convincing synthetic imagery makes documentary photography less relevant. Quite the opposite. When almost anyone can generate a plausible-looking image, there is greater value in being able to say that a photographer was actually there, saw what happened and made the photograph.
Aikakausmedia’s full interview, in Finnish, looks at AI-assisted editing, newsroom guidelines, transparency and what all of this means for the audience’s trust in photography.
