What exactly are ‘personalised news recommendations’?

‘At the Amsterdam Law School, people immediately think of social media. But it’s much broader than that: it’s a system with access to a vast pool of data that picks things out and organises them into a sequence. On a news site, for example, you see this under headings such as “Read also”, because if you are reading one article, you may well find similar content interesting too.’

How far does AI’s reach extend in the ranking of news?

‘You don’t yet see a truly personalised news offering very often yet. Editorial teams don’t want to simply hand over their standards and values to a computer system, especially as such a system currently takes little or no account of journalistic values. The NOS, for example, has a “Chosen for you” tab in its app, and you can indicate whether you want to see news from your own region. But when you open the app, you’re taken to the general homepage first.’

Do you think it’s a good idea to use news personalisation?

‘There is so much news out there that it’s easy to miss something that is actually relevant to you. That’s where the opportunities for a personalised system lie. But I would never say that we should replace editors entirely. What makes a news article newsworthy or important is a very nuanced matter. Such a recommendation system cannot yet predict this sufficiently, so before you know it, you’ll only be shown entertainment news. If we have technology designed for newsrooms and aimed at keeping you well-informed, that can add value. We want to develop technology that strengthens the position of the news organisations. But trying to create the perfect solution can be paralysing. That’s why it’s important to start small, with something everyone understands and that you can verify actually works.’

How do you prevent AI from creating a news bubble?

‘The idea behind news personalisation is often that you’re shown more of the same, end up in a bubble and are never again confronted with other opinions. But you can also use technology to do something that actually gets you out of that bubble. It’s very easy to say what you don’t want: no filter bubble, for example. But what would a good filtering system look like? We’re not used to saying what we do want. In my research, I’m anticipating that question and trying to bridge the gap between abstract journalistic values, such as diversity, and concrete technical implementations.’

‘To give a small example: on a news website, you could create a section titled “Political views you don’t yet know”. Or you could create a recommendation system that shows readers all political parties equally. But you could also argue that the parties with the most votes should receive the most attention. There is a case to be made for both positions. It is important that editorial teams first work out exactly what a diverse news offering looks like to them; then the techies can get to work.’

Are editorial teams keen to get involved in news personalisation?

‘Surprisingly, yes. Most media organisations feel the pressure from social media and chatbots breathing down their necks. They know they need to keep developing, but they want to get it right. The reality is that editorial teams don’t always have the manpower to push ahead. It’s crucial for development that journalists talk directly to the tech teams, so they know exactly what to build. That still doesn’t happen often enough.’

Are you hopeful about where things are heading in practice?

‘I’m hopeful about the personalisation of news if we do it the right way. It’s just that it’s very easy not to do it the right way. By that I mean: optimising purely and simply for clicks and for the short term. That’s essentially how social media platforms work. But news personalisation doesn’t have to work that way. I see lots of people who really want to do it right. That’s what I like to focus on.’