AI and journalism: What will young journalists have to know?

An interview with Steffen Leidel, senior consultant and AI expert, and Ramón Garcia-Ziemsen, head of journalism training at DW Akademie.  

DW Akademie: How has AI already changed journalism? 

Ramón Garcia-Ziemsen: Basically, everything and nothing has changed as a result of AI. Of course, all areas are affected, and this is not only about Large Language Models (LLMs), but also about other important AI systems, such as for verification. Nevertheless, DW's journalism traineeship is still first and foremost about being able to do research without making mistakes - and to master everything analog. It is also about continuing to train the endangered cultural technique of writing. Beyond its use in everyday journalistic work, we must of course also talk more than before about the effects AI has on politics, democracy and society. 

Steffen Leidel: Journalism is knowledge work - and therefore is also influenced by AI systems in the entire value chain. It is always about data that are collected, processed, formatted and then disseminated. AI can be used in all these steps. The question is, should you really do that. AI is a probabilistic system, i.e. it does not generate reliable knowledge, but strictly speaking probabilities. To put it very simply, it is based on statistics and therefore prone to error. At the same time, journalism itself is undergoing major changes as people increasingly consume information mediated by AI-based platforms and model providers. 

What new skills do journalists need today? 

Leidel: It is important to understand in principle that all of these systems that we use in journalism were not built according to journalistic logic. When I have a text summarized, I don't initially know the criteria. Even a good prompt does not secure the result, especially if we put in long texts, the systems weaken. Journalists need to learn where AI is a black box and how they can improve output in terms of reliability and other basic journalistic criteria, for example, by learning how to make data records more machine-readable or through more professional prompting, or by - the more appropriate term - context engineering.

Garcia-Ziemsen: That's right, dealing with AI is central here. This is not necessarily about doing an insane number of AI workshops now. Ultimately, it is a transversal topic that must be considered in all trainings. Deutsche Welle is already investing a lot in this area: there are more training courses and mandatory online training courses. At the same time, I believe that the analogue continues to play a role: people still talk to people, and it will stay that way. That's why we need AI and journalistic training to dovetail.  

What can that look like? 

Leidel: We already do such trainings in media development and trying things out is extremely important. There are always mentors who accompany the whole thing with journalistic and technical expertise. In our Kenya project, we have set up a so-called "AI sandbox". This is a safe learning environment in which we have various AI systems and tools so that people can compare and try out more complex use cases, such as linking and automating several tasks.  

Together we then consider: When does it make sense to use AI? And what are the weaknesses? For example, there is the so-called Context Rot: If you give the system too much data, some will no longer be read – the more context, the worse the result. This is a huge problem for journalism. Understanding such things and dealing with the risk is something you only learn by trial and error. 

Garcia-Ziemsen: We also have to ask ourselves crucial questions: Will the organic thought process in which ideas and thoughts arise still exist in the future? This is central to journalistic training, because a lot has worked here so far through analogue experience. That means meeting people, dealing with these people, portraying them, letting them tell things, listening to them as experts, and only then bringing AI into play as a partner.  

The danger is that many will go and use AI for a complete product. We also see this in applications, people who send us a podcast that they didn't even record themselves. But that's exactly where the problem is: the creative moment when you initially sit in front of a blank sheet of paper is no longer there. And how do we actually deal with it? 

Does that mean that the entire journalism industry has to reinvent itself? 

Leidel: I believe that we are facing a fundamental change in journalistic thinking. A central model of thought here is "liquid content," i.e. the idea of preparing content in such a way that it can be used by both humans and AI systems. AI is now an audience itself, if we think of the outputs of chatbots or automated summaries that process journalistic content. Here, information is played out multimodally, i.e. in different formats whether text, audio or video, depending on the user's needs, which requires an understanding of content as a "data set” that must be structured in such a way that it can be played out as reliably and flexibly as possible. You have to think more like a data scientist, not like a journalist. Understanding this change and then transferring it to journalism is a challenge, but also the opportunity for new creative ideas.  

Garcia-Ziemsen: I also believe that newsrooms will change in this regard in the future. In the past, you needed people for design and graphics, science journalists, and today we also need expertise in dealing with AI. And our goal in the recruitment process must be to find exactly these people and then give them a chance, even if they don't have any relevant experience in journalism yet. Maybe we even have to think about offering specialized apprenticeships.  

And how does journalistic training have to change? 

Leidel: I think you can learn a lot from the sandbox project, and you could also incorporate the development of a concrete use case into the training: I want to edit and analyze a specific data set. In doing so, you should deliberately rely on tandems, on someone who has a journalistic background, i.e. a trainee, but together with this competence from data science and computer science.  

Garcia-Ziemsen: Excitingly, in the end it's all about fantasy again, i.e. creativity and the imagination of what could be. We should consider how people want to communicate and how they will communicate in five or ten years. This includes simply imagining and developing a communication economy of the future and deriving from it what this also means for the training of journalists. I think the training should also call for creating more space to try things out without prescribing the outcome.   

Leidel: I also believe that human creativity is required right now when dealing with AI. many media are already using it and have established certain things, but we will always find new use cases, all of which will take a lot of time. Teachers and students, we all have to live with these technical systems for a certain time and then find out what we can use them for. And if you can create spaces in journalism training to find out exactly that, that's of course brilliant. 

Interview by Nina Molter