In Sweden, like in many other European countries, there is a growing democratic dilemma spawned from something I like to call the increase-decrease gap: an information increase coupled with a journalist decrease.

There is a lot of useful, local information that we are missing each day: news, sports results, events, ads, alerts and warnings, information from the local municipality, etc. This information bypasses us not because we have actively chosen to ignore it, but because it is broadcasted to us in old-fashioned ways.

Instead of automatically reaching the right person at the right moment, all this information resides in different content silos waiting for local citizens to stumble upon it.

Media companies utilising Artificial Intelligence have the opportunity to create more content to satiate reader needs.
Media companies utilising Artificial Intelligence have the opportunity to create more content to satiate reader needs.

This is not a natural behaviour on the Internet anymore. We no longer start our workday by visiting our eight bookmarked favourite pages. We don’t even open our eight favourite apps every day, and having your new app become a part of someone’s life is bloody impossible.

Meanwhile, local news companies struggle with shrinking newsrooms because one inevitable consequence of print is that revenue is declining faster than its costs. As a result, fewer reporters need to cover more subjects and larger geographical areas. Thus the volume of unidentified information increases and widens the gap even more.

At MittMedia, Sweden’s largest publisher of local news, we launched several efforts over the last two years to narrow the gap, and we’ll soon initiate an interesting project to reduce the increase-decrease gap. The project is called The AI Journalist. This post is about that project — what we’ve done so far and the next step.

What is “breaking news” in a personalised, digital world?

As I concluded in the previous section, there is an abundance of local information waiting to be turned into news for someone. For a piece of information to become a news article, we need journalists to discover it, refine it, and publish it.

When local news companies have to prioritise their diminishing resources, what gets cut first are the really small, really local news stories like minor league sports. Ironically, such material is what distinguishes local media from national media and attracts local readers and advertisers.

This resource prioritisation makes sense according to print logic; we need our remaining staff to focus on big important stories and breaking news.

But what is breaking news in a global, digital world, especially when local breaking news faces competition from an instant push notification about the latest NHL draft or one of your friends announcing marriage plans on social media?

And what about the news we are forced to cut, like the minor league sports team winning against all odds against its bitter rival or the municipality refusing to fix that pothole down the street?

The tech giants have a tough head start when it comes to closing the gap and making the right information reach the right person at the right moment. By combining user insight with smart algorithms, companies like Facebook, Google, Apple, and Netflix have given us personalised devices and services, and we can’t seem to get enough.

With the next wave of deep learning Artificial Intelligence washing over us and the rise of intelligent assistants and bots, our expectations on digital services will rise even higher. In that world, but a few years away, how will we appreciate one-size-fits-all broadcasters like news sites or the municipality Web site?

General AI to narrow the gap

In 2015, MittMedia became pioneers in the Nordics when we launched the robot text start-up United Robots, automatically turning API data into intelligible sports and weather articles. In the AI community this is called Artificial Narrow Intelligence (ANI). That means our text robot, Rosalinda, can write intelligible sports articles but does a terrible job at playing chess or recommending a good sushi restaurant in downtown Los Angeles.

By following your local sports team in our app, you get Rosalinda’s article on the latest game within an hour after the game is finished.

In addition to making sure this article reaches the right user, it’s important to highlight exceptional games for our reporters, like that local team victory against its ancient foe. But Rosalinda writes several thousand articles each week, and, as we previously stated, the number of journalists are decreasing.

Needless to say, we need algorithms to sift through the thousands of articles and find the gems of journalistic value. And since we also have text robots writing weather pieces, articles based on traffic alerts, weather alerts, etc., these algorithms must leave the realm of ANI and step into the realm of Artificial General Intelligence (AGI).

It’s time to introduce the AI journalist.

The AI journalist

Right now we have access to an abundance of information, namely all the local news articles written for 28 local newspapers over the course of 10 to 15 years, as well as all the robot-written content of the latter days.

To train our AI journalist, first we will apply unsupervised machine-learning. Using cluster algorithms, the machine will learn how to categorise any article and equip it with fundamental meta-data like category, places, people, etc.

Combined with our versatile meta-data model for describing any type of text, we can then move on to processing any external content source like blogs, the municipality’s site, local sports teams, etc. This is the first job of the AI journalist: to gather, analyse, and organise vast amounts of content.

The second job is to write short summaries. Yes, I said summaries. We’re not talking Pulitzer-winning, digging exposés here. We’re talking about an intelligible, grammatically correct sentence that describes the essence of the news story. Think of it as a tweet or push notification.

OK, so now we have an automatically written text equipped with meta-data and a short summary. This short summary can instantly be pushed to someone who, through personal settings or machine-learning-derived segmentation, will likely want to read that notification.

Man versus machine

To conclude and circle back to the increase-decrease gap, with the AI journalist project we will have created a machine that can scan the Internet and databases for interesting stories, summarise and organise them, send them directly to our readers, and alert human journalists about pieces of potential interest.

By combining man and machine, we have created an intricate set-up to narrow the increase-decrease gap.

The question is, will we stop here?

In his excellent book Superintelligence: Paths, Dangers, Strategies, Dr. Nick Bostrom suggests that the next wave of AI will be the first technological leap in history that leads to fewer human job opportunities instead of more, because naturally we won’t stop here.

We will also let the AI journalist monitor what happens with each news piece after a human journalist has been notified. How did he act? What did she write? And, most importantly, the machine will monitor what happened after publication. How did the article perform against set key metrics? What were the effects in social media?

If Google can create a robot that teaches itself to master the complex game of Go, how unlikely is it that we can create a robot that outperforms human journalists when it comes to finding the real breaking news stories in a digital world of endless information and personalised user experiences?

Not that unlikely, if you ask me.