The worries of generative AI and journalism

Generative AI, a form of technology that uses artificial intelligence to create new text, photographs, artistic images, audio and video in seconds, is getting a lot of attention.

In a nutshell, the technology employs an algorithm that uses existing internet content, such as photos, and the algorithm scrapes countless images as part of its learning. Later, a person using the online tool can simply type word prompts and out pops a new, unique, high-quality image that the machine has created from its learning.

A “mash-up” or compilation of existing images, if you will.

To be clear, the Star doesn’t use this technology and doesn’t have plans to, says Star editor in chief Anne Marie Owens.

The emergence of generative AI doesn’t change our commitment, set out in our standards guide, to accuracy and transparency in our photojournalism, she added.

What types of images can generative AI produce? One that has gone viral in recent days on Twitter shows Donald Trump being arrested by the police. It’s a fake.

I’ve seen many others. One person created a compelling image of an astronaut riding a white horse in outer space. Another fake, of course.

This technology can take you where your imagination can go — a mechanical dove with a jacket made of gears.

Garnet Fraser, a veteran team editor at the Star, brought up the topic of generative AI after experimenting with Midjourney, an AI program developed by an independent research lab in San Francisco. The image generating platform first came on the scene last summer.

Fraser used a 17-word text prompt beginning with “Gretzky as a zombie” which produced an amusing rendering of what the Canadian hockey great might look like as a ghoulish character. It was more a piece of art than a photograph.

Fraser and Taras Slawnych, the Star’s visuals editor, both pointed out that they don’t see generative AI having any immediate usefulness in terms of photos or art for what we call breaking news stories — a hostage taking, a gunman’s mass shooting, a natural disaster.

Where the new technology might come in handy down the road, they said, is for feature-type stories that might not be easily illustrated with news photography — “stories that are abstract or in some way hypothetical,” as Fraser noted in a follow up email to me.

The Star’s journalistic standards guide prohibits photo manipulation except to improve quality, such as lightening or darkening an image. “While digital manipulation is permissible to improve technical quality, any alteration or enhancement that renders a photograph inaccurate or misleading is forbidden,” the guidelines state.

“You literally can’t tell what part started as a photo and what was generated by a computer. For me it’s a hard and fast no, let’s not use this, let’s not go down that route,” Becky Guthrie, art director for the print version of the Star, told me.

But upon reflection she later said she could perhaps imagine a role for the technology in the future. “News media is so starved for cheaper ways of producing interesting visuals,” she added.

But in that scenario, labelling would be a key concern, Guthrie and Slawnych said.

“As long as we tell the reader this is an AI shot. But for something as complex as this should the note be in the cutline and photo credit for the image?” Slawnych mused.

The issue of copyright infringement has become a major sticking point for generative AI.

“Fair dealing” exceptions in the Copyright Act of Canada do permit third-party use of copyrighted content in certain circumstances such as for private study, research, parody, education, satire, review or criticism and news reporting.

But Pina D’Agostino, an associate professor at Osgoode Hall Law School and an expert in copyright and AI, says when an artist’s work is wholesale copied without their permission, “I don’t support that.

“Open AI can use this content from artists, but at the same time there has to be due reward and protection for the creators of that content,” she said.

“With all this tech there are ways to trace the provenance, the origin, of this work and rewarding the creators. It’s inexcusable to say we don’t know who the author is. In large part you can find out,” D’Agostino added.

Among the possible solutions are legislative intervention and AI developers coming to the table to “do the right thing to ensure while they innovate and progress with their tech, it shouldn’t be on the backs of the creators who are feeding the content, the data needed for these new systems to thrive,” she said.

Angela Misri, an assistant professor in the school of journalism at Toronto Metropolitan University, says for photo and image creators, there are some protections. There are emerging technologies that allow photographers and artists producing original work to ‘opt out’ of the data sets that are used to create AI-generated imagery.

For me, as public editor, another key area to consider is what generative AI means for public trust in the media, an area that has eroded in recent years due to accusations of fake news.

Misri gave me some compelling insights on that point: “Why are we handing our social capital/authenticity capital over to AI? Do we as an industry want to become the fact-checkers for AI generated content? Is that what the journalists of the future can look forward to?”

Misri said she does use basic AI for tasks such as transcribing interviews, suggesting structural changes to her writing and helping her cite content properly.

But adds: “I don’t want AI to be my creative crutch. And I don’t think that’s what the audience wants either.”

Donovan Vincent is the Star’s Public Editor and based in Toronto. Reach him by email at publiced@thestar.ca or follow him on Twitter: @donovanvincent