How AI Perceives War Imagery

Artificial intelligence presents challenges for photojournalists concerning everything from copyright to disinformation. However, Emmy-winning visual narrator Kira Pollack perceives a chance for this technology to effectively assist photographers in safeguarding their legacy and documenting what the world resembled prior to AI. 

Pollack, the Walter Shorenstein Media & Democracy Fellow at the Shorenstein Center on Media, Politics and Public Policy, engaged in conversation with the Gazette in this edited discussion. 

---

You were the creative director at Vanity Fair when AI began to permeate mainstream culture. What were your initial impressions regarding AI’s potential influence on photography as both an art form and an industry? 

My initial response was that we were embarking on an entirely new chapter of image creation — one that presented a fundamental threat to photography as we understand it. On one end, the complexity of manufactured images was alarming; on the other, it heralded an outburst of creativity. While at Vanity Fair, I was overseeing the commissioning of elaborately produced fashion imagery of celebrities and raw journalistic images from the field. With the aggressive evolution of generative AI tools and large language models, and the wide-scale circulation of fake images, the photojournalism community was set on high alert. My experience has shown that each time new technology disrupts photography, the most crucial aspect is to fully grasp it — comprehend how it functions, how it can be utilized positively, and how to guard against its potential dangers.

Your fellowship at the Shorenstein Center targets a very specific dilemma for photojournalists, so can you elaborate on the issue you recognized? 

A significant challenge confronting photojournalism today is the destiny of its archives. When individuals hear the term “archive,” they often envision dusty boxes — and their interest diminishes. However, to me, archives are living, dynamic collections that narrate the visual history of our planet. Having been at the forefront of overseeing that work — as director of photography at Time, deputy photo editor at The New York Times Magazine, and more recently at Vanity Fair — I am acutely aware of the remarkable content these archives hold. Throughout their careers, photojournalists accumulate hundreds of thousands of images, with an estimated 95 percent never making it to public view or publication. A limited number of professionals have documented the pivotal events of our epoch, and their images frequently represent the sole visual record we possess. Many of the globe’s leading photojournalists are still active and capable of contextualizing their life’s work — yet we face a significant risk of losing it.

At a time when it is critically important to safeguard these photographs before AI continues to distort our perception of visual authenticity, we must also consider how AI itself could assist. Can these tools facilitate the cataloging, organization, and contextualization of this essential body of work to enhance its discoverability? Can we achieve this in an ethical manner, without subjecting these images to unauthorized training or misuse? What are the limitations of AI? That is the heart of my research: how to leverage AI to aid our understanding — on a broad scale — without undermining the authenticity of our perception.

How have you begun testing this so far? 

In collaboration with photojournalist Christopher Morris and engineer Gregor Hochmuth, we’ve conducted nearly a dozen case studies utilizing images from Morris’ archive. For one study, we selected a variety of narratives — including the U.S. invasion of Iraq, the events of January 6, and the Yugoslav wars — and tasked AI to analyze the images. While AI can easily recognize straightforward visuals — a feline, a vehicle, a person — the genuine challenge lies in its ability to interpret the intricate complexity of conflict photography. In one notable example, the AI scrutinized a photo from the U.S. invasion of Iraq and accurately identified the scenario as a house raid, the environment as a domestic setting, the type of firearms carried by the soldiers, and the emotional expressions on the civilians’ faces — employing nuanced terminology like “appears tense” or “seems wary.” It even evaluated the composition, lighting, and symbolism. We were amazed that it could extract such detailed and precise insights with minimal context.

We are also investigating issues of authorship and legacy. Archives should be more interactive and vibrant — there’s a richness of narrative and intent from the photographer that surpasses simple captions or keywords. We are testing how AI might assist in highlighting that storytelling aspect.

You foresee an opportunity for AI to assist photographers in safeguarding their extensive archives. How do you reconcile that application of AI with the contrasting concerns surrounding the degradation of trust in what is genuine? 

I view these as two separate discussions with some convergence — akin to a Venn diagram. One aspect centers on generative AI’s capability to produce photorealistic imagery without a camera or lens. In today’s fast-paced news environment, where images proliferate quickly on social media platforms without oversight, this can create a precarious intersection that undermines public trust. Another concern pertains to copyright — specifically, the potential for photographers’ work to be scraped and utilized to train AI models without consent. This raises urgent issues regarding ownership, authorship, and protection.

The work I am undertaking occupies a third segment of that Venn diagram. It emphasizes employing AI not to fabricate or exploit images, but to preserve, organize, and highlight genuine photojournalism. These archives are extensive, often hard to access, and face threats — both physically and digitally. I am examining whether AI can help responsibly unlock this content on a large scale while preserving the photographer’s vision, rights, and legacy. It is about leveraging technology to reinforce visual truth — not supplant it.

What are your aspirations for the Shorenstein Fellowship?

My aspiration is to utilize this period not merely to scrutinize the technology itself, but to engage profoundly with the broader questions it evokes for photography and journalism. What makes the Shorenstein Center exceptional is the chance to interact with individuals from various fields — technologists, ethicists, journalists, policymakers — who are contemplating critically about the future and the values we choose to uphold.

I approach this endeavor not as a technologist. My background lies in journalism, having worked intimately with photojournalists globally and led teams that influenced how the public perceives history in the making. Rather than getting swept away by the rapid pace of the tech sector, I seek to comprehend the trajectory of this technology, recognize its shortcomings, and determine whether it can uphold the fundamental values of photography: truth, authorship, and memory. Ultimately, I hope to relay these insights back to the photojournalism community — to contribute to ensuring that we are not merely reacting to transformation but actively participating in shaping it.