Fünf Fragen an... Olle Zachrison

Olle Zhalf 600x400

Olle Zachrison ist Leiter der Entwicklung digitaler Nachrichten beim Schwedischen Rundfunk (SR). SR ist der öffentlich-rechtliche Sender Schwedens und landesweit das führende Audiounternehmen. Olle Zachrison ist auch Vorsitzender des EBU Investigative Projects & Network (EIPN). Bevor er seine neue Rolle bei SR übernahm, war er vier Jahre lang Leiter der Abteilung News & Current Affairs. Er hat einen Hintergrund im Zeitungsjournalismus und war geschäftsführender Redakteur und Wirtschaftsredakteur der überregionalen Tageszeitung Svenska Dagbladet.

You are working for Sweden’s national public broadcaster SR. One of your projects at SR is to create an editorial algorithm for news from scratch. What is the purpose of this new system?
The purpose is to create a new model that will enhance the quality of our news output. In a traditional newsroom, the assessment of the importance of a particular story in relation to others was stored in the head of the editor (or the gut). But in a digital environment it’s important to have a transparent model that also translate these editorial values into data. So the purpose is to create a computational “language” that expresses the journalistic values that are a such a strong part of our public service mission. Another important purpose is to standardize our news evaluation across our big organization. For example, SR’s 26 local stations can’t have completely different ideas on what constitutes a really important story. We need a common model. So it’s much more than a digital system, it’s the manifestation of our editorial values.

In your blog post you are writing: “The algorithm is powered by ‘news values.’” What does that mean?

Every news story at SR is done as a stand-alone news clip. Every clip is rated in three dimensions: the magnitude of the story, if it contains unique public service values (we call them “SR values”) and the life-span. Behind these three parameters we have checklists. For example: when we assess the general news values we ask ourselves: does it affect a lot of people? Are we first to report it? Gauging the SR values involves questions like: does the story contain unique voices? Are we out reporting from communities? When we rate a specific piece according to these news values we get an algorithmic score that decides where to surface the news item in our app or website.

How does the editorial algorithm work?

Basically, the rated news pieces all get an individual score. So a large story, that contains unique SR values and has a long life-span will have a more prominent placing in our playlists than an average news piece that lacks unique public service features and has a short life-span. A great SR-investigation will simply get a higher score than a quick news flash about a traffic disturbance. But the second a story is published, the algorithmic score starts to decrease, so a couple of hours later a newly published piece will outrank the first one, even if the first one had a higher score to start with. This algorithm gives special prominence to stories that signify great public service journalism, that’s why we included the SR values. Now, all our local stations have started to automate their websites and news playlist using this system, which gives them more time to focus on the journalistic content and the audio-visual presentation.

Implementing the algorithm will change the work of journalists in which way? Will they still be in editorial control?

Yes, absolutely. The algorithm is what I call “tech-supported editing”. It’s just a tool, a model that is there to enhance the human editing and make it more transparent, reliable and efficient. If our editors think that the algorithmic listing is not optimal, it’s very easy to change it. Perhaps the magnitude of the story was greater than we first thought? Perhaps we have included a new component in the story? It’s no problem then to adjust the news values and refresh the article/audio story.

Most conversations about AI in journalism tend to revolve around the robot-journalist and the question, if and when machines will replace human writers. Why do you believe algorithms and computational power can increase the quality of journalism?

When I look around, I see very few examples of serious newsrooms that want to abandon editorial control and let bots do the work without human involvement. The mainstream view is that semi-automated processes is the best way forward. Take translation for example. We think AI can play a significant role in getting out an important Swedish news story in English or Arabic, but we would never dream of doing it without an experienced journalist going through the translation. The audience is today – and rightly so – expecting a better and more personalized experience from us than the “one size fits all-model” that signified traditional media. And to serve millions of user with their relevant mix of news, podcasts and live radio is impossible without computational automation. We don’t have a million editors that can do that job!