It is a strange time to begin a career in journalism: Vice - the media company I grew up with - just went out of business, layoffs have been gripping the industry for years and now AI is further set to decimate my career.
I have trained as a journalist with German broadcaster Deutsche Welle for the past 17 months, switching over from a career as an economist. Will AI kill my career before I even start it?
I think not, if I am able to push for a career as a reporter. To come to this conclusion, I need to make a couple of assumptions on where the future is headed - some of which may be wrong.
I think there are two pathways into this question: One is to look at AI as an innovation that will transform the technological foundations of journalism. The other is on its impact on the editorial process.
This pathway of looking at AI is a simple function of how technology has shaped production processes in the past. Enhancements in typewriters, later computer technology, online publishing, and most recently mobile live streaming technology has been a simple trend: An incremental removal of barriers between reporting and publishing.
As recently as 2005, multiple TV stations would share a relay truck to report on the devastations of hurricane Katrina in New Orleans. The relay truck would allow for live reporting via satellite, accesed through a huge dish on the truck’s roof. A the time, reporters, producers and camera people would edit TV pieces in a car, plugging a mobile editing station into the car’s 12 Volt lighter outlet, to edit the tapes they had used to record scenes of the hurricane aftermath.
Today, mobile service permitting, we can do lives through the LiveU app on an iPhone and technically edit on a smartphone and publish on social media directly.
I’m not advertising that these technological changes have made journalism neccessarily better. The point is rather, to showcase how technology has cut out pieces of hardware and human labor between reporting and publishing.
So how may AI change this technological foundation? I have for example brought up a thought experiment, that in 20 years from now, autonomous drones will film on the ground reporters and relay live video streams that may be edited, cut, packaged, and delivered through an AI platform. In this thought experiment, AI mostly changes the editing process of video: If there is a constant stream of video feeding an AI editing model, this model could edit various versions of output from the same material: I would walk through the wreckage of the next natural catastrophe, a swarm of drones would capture video on this and other scenes around it and the model would recombine the video into packages following user preferences.
Jane from Nebraska may like to get a written report of what happened, Tejesh from Bengalore may enjoy an imersive virtual reality experience of the catastrophe. It seems to me very much within the scope of the technological trajectory that within the next 20 years such a process is automated.
This doesn’t mean that there won’t be any jobs for video editors or cutters anymore. But I would expect them to rather be in commercial spaces that have more funding from leasure spending, such as cinema or whatever comes out of the whole VR/metaverse trend.
If technology has mostly chipped away at the barriers between reporting and publishing, jobs in between will be affected. This includes camera people, editors and what I guess we call content editors now - people who basically publish content on various platforms.
As a young reporter that means, learning to film is smart because we’ll be doing everything ourselves anyway. But don’t get too caught up in it. Because machines will take those parts over.
From what I understand of recent AI developments, they congregate around three functions: Accessing, structuring and recombining available data.
This is true for both Large Language Models (LLM) like ChatGPT and visual generators such as DALL-E.
My basic assumption about how AI processes will develop over the next two decades is a shortening of what “available data” is. ChatGPT now is trained with data until September 2021. Let’s assume that over the course of 10 years, it has access to not just what happened until yesterday but also to Reuters, Bloomberg or AP feeds.
Now, this is an huge assumption for two reasons: On the one hand, acccess to information is extremely costly and the way I thing about systems in general is, that they bifurcate. This means: I would expect access to information to increasingly separate into walled-off silos. As corporations try to squeeze margins, you’ll get a Bloomberg feed in your Terminal, but nothing else. Maybe your operating system may just be compatible with one data provider because Microsoft bought Thomson Reuters.
On the other hand, the required computing power to make information accessible the above cannot be overstated from today’s point of view. The current cost estimate for a single request to ChatGPT stands at up to $14. This will obviously decrease with increased computing power, but I think it is fair to say that it will be quite hard to have information instantly accessible.
But even given these caveats, I think it is within reason to expect LLMs to compete with each other over what information they have available and in turn decrease the time between an event and a request being possible on this event.
This will bring new business models with it. But why would I have subscription to a newspaper if my AI curated app could just provide the information it gathers from various agencies. And experiments to this effect are already in place.
Put differently, I think it is within reason that over the next decades media organisations that rely on agency footage or information are cut out of the competition. Because technology will simply take over the processing of agency material and either directly serve it to customers or have a handful of curators do that.
But I do think there is huge space for media organisations, yet.
Current AI models provide access, structuring and recombination of available information.
Journalism additionally comprises of, accessing previously unavailable data and analysis and the ability to provide emotional anchors. Three lessons draw from this perspective.
After I had come up with the above classification of journalism, I had asked Chat-GPT-3 about what journalism is and was pleased to see that the internet on average seemed to define the field as »gathering, analyzing, and presenting news and information«. With regards to analysis, AI models can and will fundamentally change journalism.
The ability to crawl through troves of data will make AI-tools a new friend for investigative journalists. Recently, German outlets Der Spiegel and ZDF published a story on German fighter pilots who after their flying days ended, found new jobs as instructors for the Chinese military. The journalists apparently had found this story by taking a new dive into the Panama Papers leaked more than seven years ago. This shows, how many thus far uncovered stories may sit in these datasets. AI capabilities could help sift through these large datasets.
But especially in political journalism it is the ability of a seasoned reporter to know and understand trajectories of decision making, maybe the character of decision makers, the historical and cultural aspects of events, that allows for superior analysis. This knowledge coupled with access to people and events allows for a deep explanation how and why President Biden’s economic policy is fundamentally different from previous democratic presidents. The author has a historic overview and is thus able to ask the right questions. An LLM may be able to trace how certain think tankers moved into government and how suddenly its all about racial job equity with Biden. But it would have either be prompted to these questions or the respective analysis would have to already be published elsewhere.
But doing is what the human reporter can do. They can ask questions at press conferences, receive an envelope in a shady garage, meet an environmentalist doomer for debate or put sensors on hundreds of Berliner’s bicycles.
Human reporters’ agency, their ability to access information that was previously unavailable is what I think will set them appart form AI curated information for the next decades. Reporting, the act of gathering information, has always been what made good journalism. It is what young journalists should do exclusively.
There is one more aspect that I’m not quite sure about. All forms of outlets, be it media networks such as CNN, FOX, BBC or DW or TikToks, Youtubes, be it influencers or social media journalists, all of them, are filled to the brim with talking heads. People who talk about something, point at something or ask something.
They say people like seeing other people. And I personally think that this won’t change. It is absolutely within range of imagination that within the next decade a machine generated “deep-fake” image of a person could read the news, “report” from somewhere in a live “cross” or give an “expert opinion” through funneling data into a model.
But I think people like watching reality. They like to see that something could possibly go wrong, they like knowing that the person on screen or in their VR goggles is an actual person. I call this emotional anchors. An audience anchors their emotions in this person on screen. I think people would feel like a computer generated person would feel cheaper than an actual person, even if its impossible to tell them appart. I think market competition will crowd out media outlets that put their bets into a ai-generated person basket. But this is just me rambling.
I think for young journalists there is only hope if we can work as reporters and possibly as emotional anchors. The first thing to note is, that this has probably always been true. If a young journalist too quickly went into editorial roles or too quickly went into roles of news packaging or research, they probably struggled later on to get back into reporter roles.
But I think with AI generated content on the rise, this is more true than ever. For young journalists this means, we have two not mutually exclusive options: One is becoming emotional anchors. This pathway means as much on camera or public-facing work as possible. This means building a reputation with a community of your audience. Being the witty one, being funny, being entertaining. This is easier to do in visual journalism than in writing, I think, but it is inherently a very crowded field with thin ways to climb a ladder.
The other way is to gear our early careers very clearly onto the gathering of previously unavailable information. This can mean many different paths but I do think they have one thing in common: Asking the right questions in the right way.
Most of this is a matter of experience, hence gearing a career towards being a reporter will mean as much doing as possible, as much reporting as possible.
Asking the right questions is a matter of education. This will mean specialization of topics, constant education and learning, this will mean asking your employer to go to research seminars, to conferences, to get another degree. I personally don’t think that this will involve too much education on new skills because I believe technology will make access to tools easier (see above).
Asking in the right way comes through what I’d call reporter stamina. Being able to not cave in an interview but solicit information. Being able to emotionally connect with a person who is struggling to tell their story. I think this is again a function of experience but I also think that this is something that can be best learned from colleagues. Watch and observe and then ask accordingliy.
An AI won’t ever be able to do that.