Ai Visuals and Metadata
A Journalist’s Diary
Ai is the inescapable technological marvel that impacts photography as much as other activities of life. Aside from asking where technological advancements are actually beneficial to society or humanity, the fledgling technology already proves challenging for journalism. It’s become clear that we need some ground rules for using Ai to limit the technology’s threat to accurate and authentic information.
There are many aspects to the complex world of Ai visuals, from dealing with Ai “operating” cameras and controlling the post-processing stage to how we “train” image generators and treat their outputs.
Last year, I wrote about Ai in cameras and journalism, followed up with thoughts on what some see as the computational photography future, and finished a three-part Ai photography series with some words on the ethics of it.
Today, I look at metadata, the crucial post-processing tool that ensures truthful pictures can be accessed, verified, and archived.
When I wrote about metadata before, it was a general explainer about how I use it to attach relevant information about what and who is in a photograph and where and when it was made.
Metadata might not be the most exciting element of photography, but society needs to know what it can do for them.
What’s relevant for photography is especially true for Ai photography and visuals.
We can already see the dangers of this new technology: a political party using an Ai-generated ad to tell a narrative that fits the party’s message with the goal of influencing people’s decisions. That’s especially slippery territory if the note that the content is fictional and artificially created is barely readable.
Software allows for ever more convincing manipulation. What some might see as harmless fun for “vacation pictures,” (where is the point, one might ask…) others might see as the perfect tool to influence the public’s perception of historical events.
Some see visual Ai tools as a means to bring their creativity to life or access inaccessible stories.
That last example opens up its own debate: where do Ai illustrations fit into journalism, and how do we handle them? Photojournalists, like broadcast or writer-journalists, should follow a code of ethics that guides their reporting and editing decisions.
Kenneth Kobre’s guide to photojournalism “Photojournalism: the professionals’ approach” includes a chapter on illustrations and the need to “eliminate the docudrama.” By this, he means pictures that mimic reality and can confuse viewers even if they are labeled as photo illustrations.
How is that “90 Miles” piece linked above ethical sound and not risking to confuse viewers (not to speak of the big problem that current Ai generators rely on scraped data and thus generate content that’s unethical from a copyright perspective alone)?
As part of that big bundle of Ai joy readings and thought processes, I found a discussion on the future of the relationship between metadata and Ai images. The video from the 2022 IPTC Photo Metadata Conference is a 38-minute segment that provides good stimulus for one of the most important questions about Ai visuals and journalism: how can we ensure we can identify content that is not a direct photographic representation of what human eyes witnessed?
You’ll hear about “synthetic assets,” the ethical problems behind “training” Ai algorithms, the difference between static and dynamic output, and much more.
Discussions like this might not be everyone’s cup of tea, but we all will inevitably make contact with Ai visuals and need to become literate on the issue. We need to begin to understand how we can use Ai as one of the tools for the visual future.
And while we’ve already heard that the camera and software side of the industry has begun to work on technology that helps identify authentic visuals, this discussion looks at the value of metadata for the same purpose.
They are both important pieces of the Ai literacy puzzle. So, I hope you consider spending time with metadata and its role in visual content, whether it’s traditionally made or Ai-generated.
That’s all for today, next week I’ll share fresh reading recommendations, conclude my four-part photography series, and I might finally get to that training update.
