<Tamuna Chkareuli / Midjourney>
If you ever opened up a ChatGPT and blankly stared at it, contemplating the endless possibilities, and then sighed and closed it, you’re not alone. Generative AI hasn’t left the daily discourse for over a year now, but with that unexciting chat window opened, you might still be wondering – what can it actually do for me, that I already don’t know how to do?
At this point though, we all have that one colleague who gleefully shares their experience of performing some AI-boosted magic. Making a mental note-to-self to try it out later, you put it off for indefinite time. Apart from having skepticism as a second nature, and seeing potential harms everywhere we look, as journalists we also have a mistrust of everything that comes too easy. If you haven’t worked hard, it’s probably not good enough. And sometimes it’s scary how easily chatbot summarizes a full page of text or throws out paragraph after paragraph on virtually any topic. I talk to my colleagues about generative AI all the time, most of them are either straight up against, or half-embarrassed by using it.
While acknowledging all the harms of the misinformation that AI has already done, I also blame the word “generative” here. We’re afraid that our craft will be distilled down to mediocre writing that machine spits out on a whim.The privacy issue and the pride we take in our work are arguments strong enough to stay away from generative AI chatbots.
But what if we looked at it as a tool, good as any of the hundreds we are already using? If that’s what you’re wondering about, this blog is for you.
I’ve talked to three journalists and ASU professors who closely follow the relationship between journalism and AI to find out if there’s a room for cautious optimism, and tips day-to-day work.
One of them is Sarah Cohen, ASU Knight Chair in journalism and data specialist.
Data storytelling always has been on the forefront of new technology – what are your thoughts on using AI in your work?
– Data journalism, particularly in reporting, is a function of working in a lab, then going to the field, and then coming back to the lab, on repeat, until you understand how the world looks in three dimensions, compared to how it looks in two dimensions. And if we take AI, we’re just sticking in that two dimensional world, without any backup from the field.
The other part is the verification part. A lot of what we do is making sure that what we hear or see is true. And now we have another source, a chat bot, where we have to check what they say is true – so there’s no sense of truth in AI.
I don’t think most of what we do that we like to do is going to be replaced, but the promise of technology in our newsroom has always been replacing the boring stuff. Let a machine do it – that would be awesome.
Are you optimistic?
– Yes, I’m very optimistic. Greedy publishers are always going to try to turn it into a “can we get rid of all the people” – But there was this really smart man who, when I was in grad school, talked about the future of journalists or reporting and writing, and he always put our work into four buckets. One was the Great Explainer, the expert who can lead you through a really complicated topic. Second was the Investigator who would hold the government accountable. Then there was the Observer who can go places we can’t go. And the last one was the Shoveler who is just taking stuff from other people and shoveling it out so other people can use it.
And his idea was you never want to be the Shoveler, right? And I just feel like the more that AI can pick up that job of shoveling, we can keep the other three jobs.
If there is one thing that makes you worried about AI, what would it be?
The equity issue. AI reinforces stereotypes and historical disparities. If AI is all about prediction based on the past, then we’re going to keep doing the past.
–
I also spoke to Henk Van Ess, an expert on following the stories in data, and an OSINT specialist, who regularly shares insights on using AI in his blog Digital Digging, on building a working relationship with a Large Language Model (LLM) – an AI like ChatGPT, Gemini or Bing that was trained on datasets to understand and generate text.
How can we work with a model that “hallucinates” and spits out nonsense?
– Large language model is just looking in the mirror of society. It stole all the information you can find on the web. The moment you ask questions, which are too complicated or too fast or too generic or too precise, you will confuse the model.
I have a great tip for all of you. When you have an answer that is pleasing to you, ask it, “what should I have typed in, in the very first place to get that answer?”. And then you get the nerdy instructions, what you should have asked in the very first place. When you do it a few times, you’ll grasp the workflow.
Why do you think journalists are refraining from using AI at work?
– Right now there is some hesitation in journalism because of the fear for it killing our jobs. Maybe it’s not objective and hallucinating. Maybe we shouldn’t use it because it’s unethical. Why would I embrace something which is frightening? So that’s basically my mission to show and tell my peers, look, it’s just a stupid computer. It’s garbage in and garbage out.If you do the right things, it will improve your work.
What about the AI-produced fakes?
– There is no AI police at the moment. You have now a set of rules in Europe, which are probably one of the most interesting set of rules that you can think of, but there’s still no police. So there the deepfakes will be more and more sophisticated.
But in the end, I think it’s good for journalism. If we can’t trust video or audio anymore, who are our gatekeepers? Who are the people who can say, you can trust us? Things that are already happening, these big verification departments of the Washington Post, of the New York Times become more and more important. And I have this little hope – maybe a big hope- that journalism will benefit from it.
—
So how do you approach AI on a practical level while working with text?– I asked this question to Djordje Padejski, Associate Director at Knight Journalism Fellowships at Stanford, who works at the intersection of journalism and artificial intelligence.
– ChatGPT and large language models are really good in converting your content into other medium, like a long form text into a summary or a tweet. I could also recommend using ChatGPT to review your writing, asking “Hey, here’s my story. What are your thoughts on the, you know, a number of, or diversity of sources being used?” – questions like grammar mistakes, style mistakes. And while I’m not saying you should learn AP style from chat GPT, but there’s AP style already embedded in it, too. I would never, ever use AI for fact checking.
What is a constructive way to approach AI?
– I’m often asked where I stand on it, but I see lots of opportunities and lots of risks and challenges, so I don’t like to declare if I’m for or against it. However, the discourse we’ve been seeing in the media for the last two yeast was not telling the real story, it often ascribed identity or even humanity to AI. I feel like the obligation that we have to the public, the very core of our profession, is to tell that story about AI in a way that will help our public understand, first of all, what are we dealing with, but also then bring the scrutiny to AI itself. We are part of that public discourse and we are affecting policymakers that should be thinking about these topics.
Any messages for the journalists overwhelmed with the pace of things? There’s a new development every other day.
– There’s a hype around AI, but guess what? You don’t have to use these AI tools. Nothing, nothing dramatic will change if you don’t use them. There’s going to be way more going on about AI in the next couple of years. So this frenzy should not make people feel like they are missing the train.
