My colleagues and I are engaged in the convoluted, ever-shifting process of figuring out how to use artificial intelligence in journalism in ways that are both productive and ethical. Somewhere between “Let students use AI to write their stories” and “We should forbid all uses of AI,” there is a reasonable approach, and we’re all trying to figure out what that is.
Our students learn from us. We learn from our students. Keep in mind, though, that we have not yet seen what you might call “AI natives” in our classrooms. Young people in their late teens and early 20s were part of the before times. In the not-too-distant future, though, we’ll start seeing students who can’t remember a world without ChatGPT, Claude and the rest.
Recently I devoted a class to AI in my graduate ethics seminar. It’s a small group of five students, one of whom is an advanced undergrad. I was surprised to learn that they are as skeptical of AI as I am — maybe more so, since I regularly use Claude to find sources and background information, to brainstorm interview questions and, experimentally, to produce summaries of the “What Works” podcast on local news that Ellen Clegg and I host. (Here’s a recent example.) I thought I would share the results of our AI class in the hopes that others might find it useful, and that I might get some advice on how to make it better.
I began with a lecture on AI and journalism. I pulled together some anecdotes about the good, the bad and the ugly, the ugliest being a shady local news site that published an AI-hallucinated story about a murder in New Jersey a few years ago. Then we moved on to a class exercise using Claude, which I chose because Northeastern has a deal with Anthropic to provide all of us with the enterprise version. I wanted to see what we could do with an interview of Tracy Baim, a journalist who’s the head of the LGBTQ+ Media Mapping Project and the Chicago chapter of Press Forward.
Baim was a guest on “What Works” last fall, so I prepared for the class by downloading that episode and transcribing our conversation with Otter, an online service that is itself powered by AI. I spent a little bit of time cleaning it up, but the results are not publication-ready. Still, I thought it was good enough for our purposes, and you can see the transcript here.
Next I divided the class into one team of two and another team of three. I asked one team to tell Claude to produce five bullet points from the transcript, along with a few sentences of explanation, and the other to produce a 600-word summary. I asked each team to look over the results and think about four questions:
- Is it accurate, or do you see errors?
- If you were writing a story about this interview, what use would you make of the AI summary?
- Do you find it helpful, or would you be able to work more efficiently without bothering with this step?
- Do you think it would be ethically necessary to disclose that you used AI to assist you in finding the most important points?
Refine and repeat
The students who had Claude produce bullet points observed that the results were too long to be useful. So we refined it, telling Claude to restrict each point to one sentence. You can see both results in the document I’ve linked to.
After that, we switched the teams around. This time I asked one team to tell Claude to produce a 600-word news story and the other to write an eight- to 12-word headline and a social media post. Here are the discussion questions for the news story:
- Is it accurate, or do you see errors?
- Do you think AI did a better or worse job than you would have done?
- Given that a human writer could produce something more nuanced and interesting, is it worth using AI in order to save time and let the reporter move on to more productive tasks?
- We would all agree that the news organization would need to disclose that the story was written by AI. What should that disclosure say?
And here are the discussion questions for the headline and social media post:
- Are they accurate?
- What would you use the results for? To help you, as a human editor, write a better headline and social media post? Or would you just go ahead and run them?
- Do you think disclosure is necessary for this use?
What I hadn’t expected was that the command to write a news story would automatically produce a headline, so we ended up with two. I think the specific request for a headline resulted in something better (“LGBTQ+ News Outlets Struggle to Survive Amid Advertising Collapse and Staff Shortages”) than the one that was automatically added to the news story (“LGBTQ+ Media Outlets Face Structural Crisis, New Mapping Project Finds”), but your mileage may vary.
We had a good discussion in class, but I wanted something more lasting. So I put together an online discussion topic asking for a 200- to 300-word reflection on the ethical use of AI in journalism. I asked my students to refer to the documents that Claude produced in class and to come up with one more: an outline that could be used to organize a 600-word story about the interview. Here is the outline Claude produced at my request; I don’t imagine the ones that the students asked for are much different. And here are the questions I asked my students to respond to:
- Would you find the outline useful, or do you think you could have done a better job of organizing it on your own? Why? Did the outline cover all of the key points? Did it leave out something important, or get nuances and emphases wrong?
- Do you think it would be ethically necessary for your news organization to disclose that AI was used in helping to organize your story?
- Think about Chris Quinn’s policy at Cleveland.com and The Plain Dealer, which I lectured about. (Here is what he wrote.) He is assigning reporters to do nothing but report; they turn over their notes to AI, which writes a story — checked, he says, by human editors before it’s published. As long as this practice is disclosed to readers, do you think that is an ethical way to practice journalism? Would you want to work in such an environment?
The students’ answers were thoughtful, nuanced and more dubious about AI than I might have expected. I would love to reproduce everything they had to say, but here are some excerpts.
“I do believe that the outline is useful because the structure of the outline detailed separate sections and bullet points to talk about in each section,” wrote one student, who nevertheless added that she thought even the use of an AI-generated outline should be disclosed. As for Quinn’s use of AI to write stories based on his reporters’ notes, she said she regarded it as unethical, explaining, “I believe this because although he claims that it is checked by human editors the humanistic side of journalism and the use of skills in journalism is lost. It feels lazy.”
Another student with a particularly sharp eye said the outline did not appear to be an entirely accurate gloss on the transcript. “I see how it could be helpful for someone looking to organize their thoughts or outline an article. Possibly,” she wrote. “However, I noticed a quote that was reversed and certain things I would have included that were not included. Therefore, for me, it would not be helpful.”
A conscientious objector
Perhaps my favorite comment went considerably farther than I would go, but I appreciated his reasoning.
“I’m terrified of generative AI, and I have yet to open any of them,” he wrote, adding: “I didn’t create an outline, and I don’t see how the exercises we did in class would be genuinely helpful. If you don’t have the time or creativity to put together an article or come up with a tweet on your own, then this might not be the field for you.”
He also said, “Something will always be lost when a robot writes. Its creativity is learned from people, and stolen ideas are not creative. Something will always be lost. And second, when someone is replaced with a computer, that’s a hard-working individual who’s spent their life getting to a point to be replaced.”
Here’s another student’s assessment: “Overall, I find the outline to be useful. I might have emphasized different details, but it provided a fair assessment of the central points of the conversation. Given that I was not in the room when this discussion took place, it takes a great amount of brain power to make sense of the central points of the transcript. Although Baim is eloquent, a transcript of a conversation is never as succinct as writing.” She added, “Journalists are meant to be on the ground, talking to people. … AI should be used sparingly, and only when acutely necessary. Its main use should be for cleaning data or coding, and it should always come with a disclaimer if published. Generative AI has no place in written articles, in my opinion.”
And finally: “When facing a time crunch or writer’s block, an outline created by AI can certainly be a helpful tool for synthesizing and organizing ideas. In my experience, however, it is most effective only to provide a few bullet points here and there. Usually, a mention of something in an AI-generated outline will spark a new idea in my head, or a sentence will bring up a point I hadn’t thought of before. It is imperative to use AI as a supplementary assistant rather than a tool that replaces all else. Many nuances were missed in the outline I received, and I found myself thinking that I would do things a little differently. Personally, if I’m going to change so much about what an AI chat gives me, then I’d rather just do it myself.”
Obviously this is not an issue that’s going away. On Sunday, I took part in a welcome event for high school seniors who’ve been admitted to Northeastern. Not surprisingly, one of the questions we got was from a father wanting to know about AI. My colleague Jeff Howe and I told them about the university’s deal with Claude, and how we’re trying to help our students learn what’s ethical and what isn’t. Like me, Jeff has found that his students are more skeptical than we are, especially given that one of the reasons they enrolled in journalism school was to learn how to write.
Generative AI is better than it was a year ago, and much better than two years ago. It will continue to improve. The temptation for news organizations to use AI in aggressive and even unethical ways isn’t going away, either. I’m not a fan of what Chris Quinn is doing in Cleveland, but at least he’s using actual reporters and disclosing how his outlet is using AI.
What I really don’t like is the burgeoning practice of using AI to write stories based on transcripts of governmental meetings grabbed from YouTube. Ethics aside, the essence of journalism — and especially local journalism — is to connect with the public, to foster community while also providing people with the news and information they need to govern themselves in a democracy.
You can’t rebuild those ties with stories that are reported and written by AI. And my students are amazing.
