Developing AI-tools for journalists, a scientist’s opinion
Interview with David Graus, AI-researcher and data scientist working for FD Mediagroep in The Netherlands.
Which AI-tools for journalists are you in your company developing?
We are working on two project. One is called SMART Journalism, the other is called SMART radio. Both projects are funded by Google’s Digital News Innovation Fund. In Smart Journalism we use AI to get the right article to the right person. In recent years we presented the online-readers of our financial newspaper every Sunday with a newsletter that lists the five best read articles of the week before. Everybody received the same list. Based on AI we have developed a tool that now presents every online-reader with a personalised list of five articles, based on what they have read in the past. In this way we can present also the less-read articles to readers who find them interesting.
How well does it work?
Implementing this AI-tool has not only resulted in more readership engagement in an absolute sense, but also to a broader variety of articles that have been read.
Are you already working on the next step in SMART Journalism?
“The next step is to not just get the right article to the right person, but to get the right information within an article to the right person. Automatically making personalised summaries is our Moonshot-project. We have developed a working system and over the next months we hope to turn it into a working product.
Some people might worry that personalising information leads to a filter bubble. What is your opinion on this?
We don’t see a filter bubble in practice. Personalisation is just another way to get information to users. But there are many other ways, like ordinary newsletters, tags or following files on certain topics. All these ways complement each other. Nobody only depends on personalised information.
What is the idea of the second project you mentioned: SMART Radio?
We have created a tool that automatically selects, cuts and tags radio fragments based on the preferences of a journalist or an ordinary listener. You type a keyword, and our system automatically offers you the radio fragments that match your keyword. So, you don’t have to listen to complete programs to find out if there is something that particularly interests you. I especially like this tool, because it can not exist without AI. It takes nobody’s work. No human would be able to manually do the same for the 24 hours of radio produced every day, 7 days per week.
How willing are journalists to innovate with AI-tools?
Often they are skeptical. Some find AI even a bit creepy. But we talk with them and show what we do. About the personalisation they are very enthusiastic. And they see that they can save time with simple AI-tools.
What are your thoughts on fully autonomous robot journalists?
Robot journalists that make human journalists superfluous will not arrive in the foreseeable future. The bottleneck is that present-day AI lacks knowledge about the world and lacks the ability to interpret information in its context. Computers can not make a connection with the real world. Only in very limited domains robot journalists can produce brief news stories autonomously.