A few quick tidbits on AI and journalism, based on a recent seminar arranged by The Norwegian Online News Association (NONA) on Norwegian media and AI.

New AI-powered title generator

On the day of the debate, Labrador CMS announced the launch of its advanced AI-powered title generator at the PPA Festival held on Tuesday, April 25, at The Brewery in London – a UK version of the story, Journalisten.no on the launch (Labrador is lead by key former NONA-folks)

“A few thousand reporters will soon have direct access to machine learning writing support in CMS. This is one of the coolest things by delivering a SaaS Publishing Platform. When we roll out new functions, we do it to all customers. The latest function is to integrate directly with Open AI and suggest better titles to the story and then automatically send them to A/B test,” said Jan Thoresen, CEO, Labrador CMS.

“Journalists won’t be replaced by AI, but by journalists who know how to utilize AI,” laughed Inga Strümke, Associate professor in AI at NTNU and author of a popular, recently published, and currently sold out, book on AI “Maskiner som tenker”, at the end of her talk at the NONA-event.

Though it seems from the debate (which I watched via live stream) this might have been a conclusion derived from, or shared with, other key people in the audience (such as Torgeir Waterhouse) before the debate. You can still see the live stream on Facebook here, though the debate is in Norwegian (starts roughly 15-16min into the recording)

She also said the recent call to pause AI development is an ineffective way of dealing with the issue, but that we are in dire need of proper regulation of AI. She said the EU’s proposed AI act is an interesting piece of proposed legislation to follow in this respect.

As of last Thursday, Strümke's non-fiction book on AI was in fact Norway's most selling book.

Another interesting thought I made a note of from the seminar was this:

“I think there’s a lack of wider journalism about AI. Now it's being covered as a cool new thing. But I’m concerned with how it is already affecting us – e. g. Snapchat's My AI is in the pockets of very many people now, also very many young people. What kind of effects does it have?,” said Janni Frederiksen Kalafatis, UX lead in Schibsted-owned VG. Among other things, he’s been central to the development of the transcription tool JOJO and AI-generated article summaries.

As a former tech reporter I always thought covering tech should be about a lot more than just the cool new, new thing/gadget and its specifications: technology and the way it is used has so many ramifications for society as a whole (not only new tech, but also old and bad IT solutions that we’ve seen get in the way of effective legislation or even effective health care).

So it will be vital to see journalism about AI move beyond the new cool thing phase. In fact, I have read about international media hiring AI reporters. Googling, I find Muck Rack has this list of 10 top AI in 2023 – and it seems like a good list to explore. Speaking for myself, I am only familiar with the work of Madhumita Murgia, AI editor at Financial Times, as of writing.

Inga Strümke, Photo: Mona Haugli / Kagge Forlag

Schibsted-owned Aftenposten has used AI to «clone» the voice of one of its journalists and podcasters.

The project started when the newspaper was working on Aftenposten Junior, a digital news service specially developed for students in primary school where, due to strict accessibility requirements, text to speech was business critical.

The result is an AI voice with unlimited possibilities. Product director Karl Oskar Teien and product manager Lena Beate Hamborg Pedersen talked about the project at a recent meeting of the Norwegian Online News Association (NONA). Journalisten.no has an article about the project here (in Norwegian). Some quick notes:

• 5-8% in Norway have reading difficulties, so we believe this is an important democratic project
• Video streaming eats up more and more of the market, and the battle for users' attention year by year
• Scarcity of information: The newspaper used to be portal to the world. Now: Information overload – lots of sources, too little time.
• We’ve gone from from local physical distribution to global digital distribution
• Visual attention - reading on paper and screen
• Audio attention – screen-free news consumption
• More people are listening and each individual is listening more to online audio (all audio that is not radio).
• More and more people are using AirPods and other earplugs
• We are seeing more and more audio startups - one of the biggest augmented reality offerings available today. Sound facilitates more multitasking and less screen time
• Correlation between listening and customer loyalty: Aftenposten has a strong position in audio and podcasts. 200k users now pay for podcast.
• Challenges in scaling up sound production - e.g. costs, time, prediction, updates (news is fresh)
• Text to speech can be a solution (opposite of speech to text) = unlimited output and available everywhere
• Our ambition is for all of Aftenposten's journalism to be made available in audio format
• The technology we use: BeyondWords
• Used Anne Lindholm's voice, at the time she was the presenter of "Explained"
• We have had a robot trainer who has worked with this - a student who sat and listened
• Sound resonates better with younger target groups
• Users who listen complete more of the article
• Possibilities going forward: playback from the front page, playlists for different needs, play my saved articles, etc
• This is groundbreaking work both legally and technically.
• Has, among other things, a contract with Anne Lindholm which states that the voice should only be used in editorial and not commercial products.

The debate about artificial intelligence (AI) and how it will reshape various industries has exploded as of late, especially after the launch ChatGPT. Here are some ways AI is being put to use by Norwegian media, as discussed during a recent meeting of the Norwegian Online News Association (NONA).

Teknisk Ukeblad (TU) Media has developed its own AI-powered tool to transcribe audio recordings and to  subtitle videos based on Whisper from OpenAI, supported by programming in Java and Phyton.

Every day, the tool saves hours of work in the newsroom.

TU is currently focusing on how to improve the tool's speed, which of course is critical when people are out working in the field and video and/or audio needs to be published quickly. Eirik Helland Urke, responsible for photo, video and multimedia at TU and developer Sebastian Hagemo gave an interesting introduction on this.

Schibsted-owned VG has also developed an AI-based transcription tool – and will soon make it available in AppStore. Johannes Gorset, director of engineering at VG, told the meeting how the tool, “JOJO”, saves journalists hours of work – and promised that VG soon will make it available for everyone to download via AppStore and run on ones own servers.

Interestingly, in the early stage of developing this tool the programmers experimented with using Amazon Sweden cloud storage, which of course didn’t provide the necessary security for protecting sources. This is essentially why VG created an app: to ensure that the solution runs locally and is secure.

This is, of course, is of vital importance when working with major investigative podcasts on issues such as true crime or other controversial topics such as VG has had great success with.

Journalisten.no has an interesting article on JOJO, written with the help of JOJO.

At financial daily Dagens Næringsliv (DN), language models are used to strengthen DN Investor, the newspaper's stock exchange and market service.

Martin Kermit, Head of Data Science at DN-owner DN Media Group, told the audience how large amounts of text are analyzed using natural language technology (NLP) to strengthen DN Investor, the media house's stock exchange and market service - which combines share prices with journalism.

Users get access to share prices on the Oslo Stock Exchange together with news about developments in the markets.

The service also contains a number of useful tools, such as notifications when something special happens on topics the users follow.

AI is used to (quick notes):

• Automatically identify companies on the Oslo Stock Exchange mentioned in a DN article (and only relevant companies). This is based on Natural language processing (NLP). It’s a challenge that NLP is not necessarily so steady in Norwegian, so DN has collaborated with Nowegian media company Amedia on developing language models based on the framework spaCy, as well as language information from Nome and NDT. The NLP language model is trained using the content of these two media groups. Most important component: Named entity recognition (NER). Practical use: integration in the CMS - it's implemented as a plugin in Dr. Publish (the CMS DN.no runs on). The company is added automatically, the reporter has the option of editing.
• DN has several projects with NLP language technology: Automatic categorization of news, article recommendations for further reading etc.

Interestingly: The language model must be constantly updated with new words, e.g., corona - when the pandemic hit, the tool was just unable to decipher that word.

Disclosures: I’ve co-founded and was the founding president of NONA and am of course a big fan of their work even though I no longer have any official role in the organization. I’ve also worked for VG and Journalisten