Tackling an AI contribution carefully

TARLETON – A publisher called the center last week with a problem. Always happy to help, we listened. The question started out short, but the answer took some time to unpack.

By AUSTIN LEWTER, Texas Center for Community Journalism director

“What is your view on articles sent by community contributors that use AI to write it?” they asked. “I just received an article from a teacher that is 100% AI.”

The publisher said the person who submitted the piece had not disclosed using AI. The paper hadn’t questioned them, but a few online AI checkers predicted it was 100% AI generated.

There are multiple platforms popping up to check for AI usage. Some of the more popular ones include Copyleaks and ZeroGPT. 

The proofreading site Grammarly has also developed an AI detection tool. Our publisher ran the story submission through all three and the consensus was that the work was 100% AI generated.

To understand the nuances here, it helps to understand how these detection sites work.

All generative AI content is produced when a chatbot is prompted to create it. Nothing that comes out of ChatGPT (or any other Generative AI platform) has ever been published before— anywhere. And, if given the exact same prompt to the exact same platform again, the next response will be equally as original.

It will not copy and paste from previous work. So traditional plagiarism checkers don’t work with AI generated content. 

Since that exact work has never been produced before— as is the case with most traditional plagiarism when someone steals someone else’s work— there is no previous work out there to compare it to.

A traditional plagiarism detector will not catch AI generated copy. I can’t, either. The advent of AI detecting platforms like the three mentioned above aim to curtail this, but they aren’t perfect.

ZeroGPT is interesting. It detects text that is “likely not human generated” — by using (wait for it) artificial intelligence. Ironic indeed. 

It looks at word patterns and syntax and guesses what percentage of the work is, again, “not likely human generated.” 

A low percentage should not raise flags. In the case of our publisher, a 100% rating is something to worry about.

I encourage you to start playing with one of these detectors— especially with submitted copy. There are free versions for each of them.

The story submitted to our publisher was about a local high school group going to a state contest. 

There was nothing factually inaccurate, but the detector scored a 100% likelihood that a human did not write it. 

I told them that those AI checkers aren’t always completely accurate. According to the latest research, the AI checkers are wrong 7% of the time. 

So, if the publisher confronted the contributor, there is a small chance they’d be wrong. 

But when asked, folks most often fess up to AI use— at least in my experience.

The publisher said they, themselves, use Grammarly to check for misspellings and grammar errors.

“Is that the same ethical gray area as ChatGPT?” they asked. “I am still trying to figure out where the line is and how to navigate this.”

Absolutely not, I said.

Grammarly, spell check and other copyediting tools employ artificial intelligence— just like your car’s GPS system— but they don’t employ generative artificial intelligence. Grammarly and spell check do not create the work from scratch.

When the rubber hits the road, all this depends upon your newsroom’s AI policy. If you have a policy, you should stick to it. 

If you don’t have a policy, you should create one. 

Generally, I think it’s okay to use Generative AI for ideas, brainstorming and sourcing information. Many newsroom AI policies adhere to this. 

Yes, you should always triple check the sourcing. That’s just good journalism. But using AI to write any part of a story is plagiarism. 

For our publisher, the sticky question is this: since the story was sent from a contributor and the paper is not claiming to have produced it, are you in the wrong ethically for running it?

To me, the answer is simple— if you know (or even suspect) they didn’t produce story, you can’t run it in good faith with a byline that claims authorship. “If you didn’t know AI produced the story, that’s one thing,” I said. “In this case you do, so I think you know what not to do.” 

In the end, their options were to confront the contributor— which was an uncomfortable thought— or reject the submission and have a newspaper staffer tackle the story on their own. They could write something original and claim ownership of that story.

That’s what they did, and I agree it was the best choice. 

A staff-generated story trumps community contributors and chatbots any day of the week.