News publishers have experienced a massive digital transformation in the past 20 years. Now, we are on the cusp of a new major wave of digitisation driven by artificial intelligence (AI). This wave brings both vast opportunities and challenges for news publishers.

Ekstra Bladet identified the transformative potential of AI in 2020 and decided to invest massively in AI to begin our journey toward becoming an AI-driven news brand in the 2020s. More specifically, we launched the Platform Intelligence in News project (PIN) which aimed to:

1. Create editorial and commercial content by using AI to create a wider (with recommender systems), deeper (with natural language processing), and richer (with natural language generation) news experience.
2. Realise which AI systems aligned with editorial values and over which we had control.
3. Finally, by sharing our experiences in both these areas, to contribute to a healthy norm-setting for the use of AI in media.
   

You can read more about our recommender systems here. But read on to learn how we have been introducing GenAI in the newsroom at Ekstra Bladet with MAGNA, our key platform and toolbox for Monitoring and Assisted Generation of News Artefacts.

A richer news experience

With MAGNA, we aim to simplify work processes and broaden journalistic capabilities, enabling journalists to allocate time more efficiently to high-quality journalism by reducing time spent on repetitive tasks. By automating routine responsibilities, MAGNA empowers journalists to concentrate on creative and analytical aspects of their profession, ultimately elevating the overall standard of media production.

MAGNA was rolled out in two phases in 2023:

1. Auto-generated articles

In the first half of 2023, we rolled out a version of MAGNA that enabled the automatic generation of a news feed containing local news stories. Readers who informed us of their area of residence were exposed to these stories. The purpose was to produce a local news feed of value to our readers without adding to the workload of our journalists.

The local news feed included auto-generated news stories containing local business news, local house sales, and stories based on restaurant inspection reports mixed with human-written stories from Ekstra Bladet’s editorial room. Initially, the auto-generated news stories were created with traditional robot journalism (rule-based methods or “good old fashioned AI”), combining structured data with text bits using rules that we configured in an editor (AX Semantics).

However, as generative AI matured, we augmented the auto-generated stories with fact boxes that gave them context. These fact boxes were produced automatically with GenAI, first using davinci-003, then GPT-3.5, and later GPT-4 to summarise unstructured data from sources such as statements from management in annual reports or comments from auditors in food inspection reports — always with a human in the loop.

2. Editorial tools

In the second half of 2023, we introduced a new version of MAGNA that included generative editorial tools for all journalists in Ekstra Bladet’s newsroom. The purpose was to assist journalists in their workflows, saving them time and enriching their stories with new content elements.

The new version of MAGNA included tools in three categories:

• In the “Articles” category, MAGNA assists journalists with eight features developed to meet the specific needs identified for daily tasks: shortening articles, proofreading, bullet point summary, headline suggestions, fact box, improving article, rewriting draft to final article, and find similar articles.
• For our live coverage, MAGNA generates bullet point summaries and headline suggestions and identifies reader question themes, creating a dynamic interaction between readers and journalists.
• By leveraging semantic search and Retrieval Augmented Generation (RAG), we have revolutionised search in our article archive, empowering reporters to discover articles and establish connections within the archive effortlessly. Furthermore, upon retrieving search results, our system generates fact boxes, case summaries, draft articles, or even allows for completing free-form tasks.

Just beginning the journey

With this version of MAGNA, we have taken the first major steps in a long process that aims to bring AI into the newsroom in ways that align with the values of the news publishers. However, MAGNA has no end date, and we will continue adding new functions in MAGNA to enable journalists to work more efficiently and to enrich their news stories with new content elements in text, image, and sound. 

PIN’s accomplishments, particularly through the development of MAGNA, have established it as the core AI unit within JP/Politikens Hus. This recognition has created new opportunities to expand and enhance PIN’s AI technology across the various media brands within the organisation.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .