How Can We Use AI Safely and Responsibly?

Branislava Lovre: Welcome to AImpactful. My guest is Yumi Wilson. She started as a reporter, later worked in Silicon Valley, and for years she has been teaching journalism students. Today she runs a podcast and a newsletter about AI and journalism, and she’s writing a book about AI. In this episode we talk about how AI is changing everyday newsroom work, how to preserve your own writing voice, what small newsrooms and freelancers can do with almost no budget, when it’s better to stop using a tool, and what audiences expect from a journalist. Welcome, Yumi.

Yumi Wilson: Thank you so much for having me.

Branislava Lovre: You started at the AP, then worked in Silicon Valley, and now you teach. In each of those places, you’ve seen journalism change. Now everyone is talking about AI. When you think about your students and other young people coming into the profession right now, how do you feel? Are you more worried about them, or more excited about what they’ll be able to do?

Yumi Wilson: That’s a great question. I’m in the middle of writing a book for a publisher, and the research is raising a lot of questions for me. To be honest, I’m a little worried that we are grappling with a lot of changes very quickly. We’re entering this phase of agentic AI and automation and the possibilities are just endless for newsrooms and companies, many of them for-profit companies. Journalism is the fourth estate: we uphold democracy and question the world around us. I wonder whether the automation of certain tasks might homogenize or flatten the voices and unique qualities that individual journalists have brought to the field.

I think we have to guard against that. As an educator, I’m on sabbatical this fall. But when I return to the classroom in spring 2027, hopefully with at least a draft of the book finished, I think I’ll introduce a new set of parameters for using AI. I use AI. I see the value in it, but I also see the dangers of it.

Branislava Lovre: You said you use AI, and you see both the good side and the dangers. Let’s start with the practical side. Which newsrooms, in your opinion, use AI effectively, and what do they use it for? And what about a journalist who works alone? Where does AI help, and where does it bring new questions?

Yumi Wilson: Let me start with newsrooms in general. Look at The New York Times, The Washington Post, the Financial Times, the BBC, or the Associated Press. The list goes on globally: news and media companies have integrated AI-powered systems into their workflows. For example, we see AI summaries on Google, but many news companies are also creating automated summaries for their readers and staff. When I interviewed Ole Reissmann from Der Spiegel, a large publication, about a year ago, he explained that AI wasn’t replacing fact-checkers, editors, or writers. It was being used to assist them.

For example, fact-checkers have access to AI-powered tools that can help them do their jobs better. We’ve also seen award-winning articles produced with the help of AI-powered tools, whether machine learning, task automation, or generative AI. There are different types of AI, of course. But what’s important is that companies are asking, ‘How can we perform our existing functions better and faster?’ For example, they can analyze enormous amounts of data that might take a human a lifetime to work through.

I think individual journalists are asking themselves, ‘What can I use while still remaining myself as a journalist?’ There’s a lot of discussion about whether AI can help you pressure-test an idea or identify potential sources, much as Google has done. Google also has AI summaries that often appear in search results. Studies have shown that even if you don’t think you’re using AI, you should take a closer look. It’s becoming part of almost every aspect of our world.

Traffic maps, weather apps, search engines. Getting back to the point, there are many things individual journalists could, and perhaps should, do with AI, depending on their company policies and ethical considerations. But I think people draw the line at a key question: are the words your own? When you sit down to write your final copy, is it your voice?

Branislava Lovre: What would you advise, and how do you work with these tools yourself?

Yumi Wilson: I’ve been grappling with this myself. I started using generative AI models, I think, in early 2024, after ChatGPT came out in late 2023. One of my students may have used it because her lead was so great, or at least that’s what some friends thought. Anyway, at first I thought, ‘Wow, these tools can write. They can produce a draft, and I can take that draft and fix it.’ But my approach has evolved. Now I say to myself and others: write the original draft yourself. Write what comes to mind.

Be Virginia Woolf. Use a stream of consciousness. Let it come out. Don’t worry about grammar. Don’t worry about where the commas should go. Don’t worry if a clause is out of place. Write what comes from your heart, mind, and soul, and from your reporting and interviews. That’s what an AI tool can help you improve. For example, after you’ve written a draft, you might run it through Grammarly’s AI-powered tool, which costs about 20 dollars a month in the US. You might discover that your sentences are rather long and complex.

Perhaps the subjects don’t agree with the verbs, or you’ve misspelled someone’s name. Catching those errors is extremely helpful. In large newsrooms, we used to have fact-checkers and copy editors, a whole team giving us one-on-one support, helping us become better writers and catch mistakes. Many content creators today, including independent freelance writers and students, don’t have those teams. That’s one way AI tools can help.

From Grammarly to ChatGPT, Claude, and Perplexity, there are so many tools. You might tell a tool, ‘I’ve just written this, and I’m pretty happy with it.’ You can also prompt it: ‘Don’t change my voice. I’m a simple, direct writer. I don’t like big, ten-dollar words.’ People can give these kinds of instructions to a powerful tool, and it will respond to them. But the tool shouldn’t be the one telling you, ‘This is how you should write it.’

‘This is how it should look, based on other people’s writing. This is how you should sound.’ That’s where the tension comes in. Sometimes we look at someone else’s writing and think, ‘Oh, that’s really good. We couldn’t make it any better.’ But that doesn’t mean it’s the way we write.

Branislava Lovre: In a newsroom, many people are involved, and everyone may have a different opinion about AI. Some want to try everything, and some don’t want to use AI at all. If an editor asked you where to begin, what would you tell them?

Yumi Wilson: The first thing you should do is sit down with your staff, everyone who’s going to be involved, and ask for input: as a company, as a newsroom, as a team, where do we see AI being useful to us? Where do we see AI not being useful to us? Come up with an AI governance policy. It’s not just a matter of saying, ‘You can use AI,’ or ‘You can use this particular tool,’ because some people will say, ‘That tool isn’t very good. I’m going to try four others.’

Again, have that conversation. Newsrooms and companies have to move forward. They have to turn a profit, survive, and pay their bills. I’m not arguing with any of that. I’d say: sit down with your team, and invite experts like you into the conversation. Ask, ‘We want to come up with an AI governance policy. What should we be looking at?’

Branislava Lovre: Can you give an example of a task where AI really does a good job? Perhaps from a newsroom you know from the inside?

Yumi Wilson: I’m thinking of the Associated Press, for example. It was my first job back in Los Angeles. I graduated from USC, interned at AP, and then got hired. To me, the Associated Press is a role model for news that is as fair and balanced as possible. Without emotion or subjectivity, always aiming for accuracy. But AP is also committed to incorporating AI into its systems and workflows. They’ve done a number of things, though I can’t pull up the articles right now.

There are numerous articles about this. I’d invite anyone to look up the Associated Press. You’ll find case study after case study of how they’ve used AI, including for traffic information. I remember that when I was there, one of my jobs was to take weather information from the National Weather Service and simplify it so we could send it to our members, the people who pay for AP content. That’s a perfect function for a machine because you can give it a simple task.

You can control the output: ‘Here’s the task I want you to do, and that’s all.’ There’s no subjectivity. You’re not asking it, ‘Tell me what you think.’ So I think AI is well suited to functions where you’re not asking a machine for advice. You’re asking it to perform specific tasks. The AP… I’m also thinking of Heliograf, which I believe was with The Washington Post. There are all these case studies that I read and think, ‘That makes sense.’

Branislava Lovre: Big media companies can build their own systems and pay for expensive tools, but many colleagues work in small newsrooms or as freelancers, often with very little money. How can they use AI without putting their sources or their material at risk?

Yumi Wilson: Several years ago, I moderated a panel for the Online News Association featuring three small newsrooms that were using AI in different ways. They didn’t have large budgets or teams, and they couldn’t afford enterprise solutions promising things like, ‘We’ll keep your content safe in your own cloud.’ They didn’t have access to any of that. Listening to them, and reading more case studies, made me think about how smaller newsrooms and independent journalists can use AI without those budgets.

If you put content into an LLM, say ChatGPT, one best practice is to ask yourself: would you be comfortable if that person’s name, or anything else you’re sharing, became available to others? My advice is that if you’re comfortable sharing the content because it’s already public, that may be less of a concern. Another tip is to remove names and personal information, such as addresses or phone numbers, if you can.

Another step is changing your settings, for example, telling the service not to train its model on your content. These are small steps people can take when they don’t have budgets.

Branislava Lovre: For journalists, interviews are often the most sensitive material. They contain names, phone numbers, and things people told us off the record. Many of us now use AI to turn these recordings into text. Is there a tool you trust with that?

Yumi Wilson: One of my show’s sponsors is Good Tape, which is based in Denmark. What I like about the company is that Good Tape grew out of Zetland Media. They talked to journalists and said, ‘We know you have to transcribe hours and hours of interviews. Yes, you could use Rev.com, Otter, or many other services, but is your content safe?’ So they created Good Tape, which lets journalists around the world transcribe their interviews.

They say, ‘Your content won’t be shared with the public or in the cloud. It will be yours. You own the copy.’ One thing Lasse Finderup told me was to read the rules. When we sign up for a service, we often skip the fine print. We’re tired, we don’t have time, and we’re not legal experts. Sometimes we can’t make sense of it. But when you’re signing up for a tool, you can copy and paste the fine print into an LLM and ask, ‘What does it mean?’

What are the dangers and red flags? Do I own my content, or does the company? What happens to it? Are the rights granted in perpetuity? These are things we may not understand. One thing I like about AI tools is that I can ask questions about things I don’t understand, then double-check and fact-check the answers. That gives me more clarity about the tools I’m using. I’m also an independent journalist and an educator; I don’t have access to enterprise solutions. So I think you can use AI tools even with a small budget or no budget.

You just have to read the fine print, change your settings, and be aware of what might become public. For example, if there’s a lawsuit and someone requests all the data on a particular matter, they might be able to access your content. So you have to take all of those things into consideration.

Branislava Lovre: You mentioned reading the fine print and checking the settings. For me, that’s where AI literacy starts. What does AI literacy mean to you, and how much do journalists really need to know?

Yumi Wilson: Everyone used to talk about media literacy, and it’s still extremely important. But I think if you’re not talking about AI as part of media literacy, you’re missing the boat. As we’ve said, AI is built into so much of what we do and what we receive as consumers. If journalists, media practitioners, and students don’t understand how AI works, what tokens mean, its environmental impact, or the different phases of its development and where we’re heading, then we can’t fully understand how media is changing and how it will be affected.

Branislava Lovre: You talk about all of this on your podcast and in your newsletter, which I really like. How did the podcast start, and why did you choose AI as the topic?

Yumi Wilson: I’d always wanted to start a podcast. In 2023, I hired a coach, originally thinking she’d help me break through my writer’s block because I’m always trying to work on a book or something like that. But over 12 weeks of coaching, I realized that my dream was to start a podcast. The big question was: what would the podcast be about? The timing couldn’t have been more perfect. I was starting to hear about AI, how students were using it, and how teachers and educators didn’t understand it or what impact it might have. I teach journalism, but at that time no one was talking about it.

When I searched for the term ‘AI in Journalism,’ the original name of my podcast, I found that it was available and wasn’t being widely discussed. I realized that, as a journalist and educator, this was a space ripe for discovery and exploration. So I jumped into it. That’s how the show started: I was curious, and it seemed like a great way to learn about AI, because I’m not an expert. I’m not an engineer, a coder, or a legal professional.

I’m just a regular person trying to understand what’s happening with AI. That’s how it started, and then the newsletter followed. Because of that, I’ve had opportunities to travel and speak at conferences. Recently, I gave a talk to people from the Solutions Journalism Network and the Metcalf Institute, who are concerned about the environmental consequences of AI. For that presentation, I needed to research the topic, find statistics, and discuss AI use and what people might be able to do with prompting.

I usually start by saying that it’s up to you. You decide whether to use these tools and how to use them, unless your company says otherwise. I think we’re in a situation where you’re not necessarily forced to use them. You can choose to do so, and the important thing is having a conversation about what that would look like.

Branislava Lovre: AI tools promise to save us time, but sometimes we spend more time fixing the result, for example a video, than we would spend doing the work ourselves. Where is the line? When does it make sense to say, “This tool is not for us”?

Yumi Wilson: Here’s the thing, though. If you haven’t been trained in what makes a good video, the rule of thirds, good lighting, and the authenticity of good filmmaking or media production, but then trust an AI tool to do it for you, I don’t think you’re going to get a high-quality product. I think the work still needs to be driven by experts and professionals, whether it’s graphics, images, or video.

The public is saying, ‘We don’t like this. It looks like AI slop. It looks terrible. It looks cheesy.’ That’s the first thing. Feedback from the public will help determine whether using the AI tool was really worth your time. One of the first AI classes I took was in graphic design. It was an online class taught by a graphic designer. She’s really good with Canva and art, and she’s also a big AI user.

What she showed in the course was that she had the skills and knew what a good graphic should look like. AI could assist her in creating or finishing it, but she had the vision. To me, if you don’t have the skills yourself and are asking AI tools to do the work, you probably won’t get great quality. Maybe you shouldn’t use them until you gain that experience or work with people who have it, paying for their expertise if you need to. And if you try these tools and the output isn’t very good, I think you have to walk away from them.

For example, I don’t usually ask AI tools to generate images. Once in a while, the woman I work with creates AI images for my social media posts or website. But if I were writing an article for a magazine, I’d want to go out and take the photo myself. Unless it’s a composite or a graphic, of course. Graphics are great, but you do have to check every single number. And in my own experience, if you ask an AI tool to create an image of me, it’s usually not very accurate.

I think they’re getting better, but there’s another issue: LLMs are trained on large datasets drawn from information that’s already out there. For women, people of color, and others often left out of mainstream media stories, our images and experiences may not be accurately reflected back to us when we use LLMs. So if these tools aren’t working for you, don’t use them. Maybe in two years, or five years, as we move toward AGI, which is apparently going to be as smart as a human, things will be different.

That’s what they say, right? We can have that conversation again then. But right now, if a tool isn’t producing the results you want, go back to the books, hire people, or do it yourself.

Branislava Lovre: You said the audience is often the first to notice. People today don’t just trust a media outlet; they also decide whether they trust a specific journalist.

Yumi Wilson: I think the public wants to know you. People want to know that you’re real. They want to know your foibles and flaws, as well as your strengths. I think they want that more than ever because AI output can sound so perfect. Everything looks perfect. Everything sounds perfect. But when you dig a little deeper and look more closely, it’s too polished, right? As human beings, we’re not perfect. And when you think about why a Hollywood movie becomes a blockbuster, it’s often because we can relate to the main character.

That character makes mistakes, has to overcome them, isn’t perfect, and struggles, just as we do. I think we want to reflect that as writers, content creators, photographers, and biographers: things aren’t always the same, and they aren’t always great. How do we capture and share that? If the public can feel that you’re coming from an authentic place, saying, ‘Hey, I’m not perfect,’ that’s when people connect.

Branislava Lovre: Yumi, we started this conversation with young journalists. What would you like to say to them, and to anyone who is still trying to find the best way to use AI in their work?

Yumi Wilson: We’re in a world where AI dominates conversations everywhere. I was in Canada recently, and so many news stories, especially those about stocks, were about AI. The advertising was about AI, too. We know that AI is part of our lives around the world. People are talking about it at the corporate, company, and team levels. We can’t avoid the fact that AI is being discussed. We also have to recognize that some people are pushing back against AI.

There have been commencement ceremonies where speakers were booed or graduating students walked out. We can see the tension: AI is arriving and being used, yet some people are very worried. Movies like The Matrix and The Terminator come to mind. People wonder, ‘Are we heading into a world where AI will take over?’ That fear is real. But I’d say we have to understand what AI is and how it’s being used in media.

It goes right back to media literacy. Once we understand AI, its applications, dangers, pitfalls, and best practices, we can make a choice. Do we want to use it and for what purpose? My advice is to gain that knowledge first and make your decision based on it, not simply on what you’ve heard or seen, or because you think it sounds better to say, ‘I’m not using AI,’ or ‘I use AI.’ It’s not a popularity contest.

It’s not an either-or choice. As human beings, we each need to learn about AI’s evolution, how it’s used, and where it’s going. By becoming aware of what’s happening, we gain more power. I think that’s really important as we move forward.

Branislava Lovre: Yumi, thank you. To everyone watching, Yumi advises you to write the first draft yourself, to talk to your team before choosing any tool, to read the terms before uploading anything, and to stop using a tool if the results aren’t good. And as she said, the point is not to be for or against AI, but to understand it well enough to make your own decision. If this episode was useful to you, send it to a colleague and subscribe to AImpactful.