1. How would you define news personalization and what elements/variables are involved in the process?

I would define news personalization as the move from an environment in which the news was made available in a single form through a single channel that any member of the public could in principle access or (as) a number of individual single channels  to a world in which at this stage primarily the channels themselves will personalize the content to the individual users on the basis of a number of different machine readable signals that are both active and passive forms of personalization, and in the future I´d expect that we would see that the content itself might also be personalized; not just the distribution and curation but also the actual content itself and again it´s done on the basis of automation and quantified data signals…

Active and passive personalization:

Active personalization is a decision by user to change the settings of a digital interface in a way that changes the display decisions that are being made…this can be from the very symbol to telling a newspaper app for example which sections you´re interested in …do you want sports or not, do you want business or not and this will change what you see when you open the app…to far more sophisticated forms of personalization where you might indicate your level of interest in individual topics or stories or individuals and so on and so forth in ways where you can manipulate quite carefully as a user what exactly you see when you open an app

Active personalization can also be not the selection of content or topics but the deselection of them, right, so this is something that I think is at this stage more developed in social media environments that news and publishing environments ..but it could be things like snoozing individual voices or simply buttons that says you know see less like this and then leaves it to the machine learning algorithms to determine what this is if you will…these are choices made by users where they may not have full control over the consequences of the choices that they make but they can make choices that influence what they see …

Passive personalization on the other hand are personalized curation decisions that are made on the basis of data that is collected without the user actively expressing or influencing preferences….done in line with terms and services…these are the basic principles of using both machine learning and AI to add front pages to the news sites….

Passive personalization does not imply that this is in any way nefarious or indeed invisible to the user, though of course it is important to mention that in contrast to active personalization, there will be more cases of passive personalization where the user is not aware that this is happening.

2. How do you see journalists’ perspective when it comes to adopting Artificial Intelligence or when applying news personalisation in their work?

I think there is a lot of fear and uncertainty among many journalists about the implications of the rise of various forms of automation, machine learning, personalization and the like in the news…and some of this fear and uncertainty is general…about sort of the disruptive potential of large shifts in the technologies of the media environment, some of them are associated with the fact that these technological shifts right now are coinciding  with structural contraction in the news industry which means that they are often accompanied by cost cutting and people are worried about their livelihoods and their jobs ….

There is also a fear that is specific to the unknown.. .that many working journalists have limited technical expertise  and competence …and are thus poorly positioned or equipped to understand the real and likely future consequences of this technology in their work.

3. To what extent is AI applied in journalism and what do you think are news organizations’ expected outcomes?

I think we are seeing a number of market leading news organizations …including many important European newspapers like The Times of London, ..the Global Mail in Canada, so sort of midsize organizations , as well as public service organizations like the BBC and their counterparts  are investing in the use of automation, machine learning, personalization in AI, and I think it’s clear in all cases that these organizations are seeking both short term, specified, operational goals but also view that it’s critical for their long term success  to develop their capabilities and understanding in these areas so that they’re able to integrate these technologies in their work as this technology matures

4. According to a study you have done in 2016 people prefer content personalized based on their preference, so the concern that arises here is that users might not be exposed to diverse quality content or stories that they need to know because of their personalized preferences, giving room to the risk of filter bubbles, echo chambers and polarization as consequences of personalized content consumption. How would you comment on that?

There are two sides to the answer to that: First, empirically demonstrated consequences of a move to a media environment where a greater and greater proportion of news consumption is shaped by automated algorithmic selection; for example reliance on search engines and social media.

The second part…how we understand the way in which the audience thinks about and engages with that and what the likely outcomes are.

Personalization, filter bubbles and polarization:

On the first part, there are a lot of unsubstantiated understandings, assertions and opinions in this space, some of them I think are based on plausible hypothesis, like the fear of filter bubbles leading to echo chambers or the fear that…intense debate on social media might contribute to political polarization. As far as I can tell, the best available evidence suggest that neither of these things are happening, that in fact, search engines and social media use are both associated with more  diverse and more news use online than not relying on these algorithmic forms of selection; so this is the opposite of the filter bubble hypothesis…

In terms of polarization, the same technologies that rely on the same systems are being widely deployed and used in very different countries around the world. If these technologies directly drove polarization, you would expect to see the same outcomes in every context; we demonstrably do not see the same outcomes in every context. We see a few very prominent countries with very high levels of polarization that coincide with the rise of social media; the United States is the poster child of this development. But, it’s also obvious that there are a number of countries in Europe where this is not at all happening to the same extent.

Information needs of the public

This one, I think, is a more significant worry and one where I think the evidence is stronger…I will respond in two parts:

The first part is that there is no question that we are moving towards an environment characterized by greater information inequality between those people who seek out a lot of news and are very interested in news and consume a lot of news, and those people who are less interested in news and don’t seek out a lot of news and consume less news. The differences in interest are not substantially different than the way they were in the past but there is a very significant change in the amount of choice people have, which means that the different interests have very different outcomes …because in the past where people had fewer channels to choose from, if you wanted to watch TV, the news was the only thing on…now if you come home and you want to watch something you just click on Netflix or whatever you want to see and you see that!!

So in this environment, there is greater information inequality between news lovers and…news avoiders…These differences are heavily correlated with socioeconomic differences in terms of class, education and income. So there is a tendency towards a Mathew effect where the most privileged are more informed and the less privileged are less informed.   

The second thing…I think that this is a …really difficult situation where to intervene in because, the fundamental thing we have to confront is that this is driven by choice, and that one of the underlying mechanics here, it seems, is that a very large part of the public in many countries – the majority of the public – do not find news, as it’s produced by professional journalists, to be valuable, trustworthy, relevant, informative or helpful. And if that is so, and they choose not to use it, it’s hard to see on what basis one would…make arguments about what’s  in their real interest….

People make decisions in this space, and right now a lot of these decisions are leading them away from the news, and I think some of the reasons why has to do with technology and needs to be examined, but I think it’s also clear that journalists themselves need to look at why is it that the a very large part of the public seems to have come to the conclusion that journalism is not really relevant and valuable for them.

5. Moving to the ethical dimension of adopting AI and news personalization, how far do you think algorithmic transparency is important?

We tend to use transparency as a general term…

Transparency in itself, is a very weak vehicle for the ultimate aims that I think we are often looking for, which is accountability and intelligibility…Accountability being that people who make important decisions can be held to account for them, and intelligibility being that citizens have a chance to understand the world  in which they live and the forces that impact their lives …Transparency is a very weak vehicle for achieving those two goals, for a simple reason that a lot of the technology involved is too complex and simply being able to access it would not actually help with accountability or intelligibility in those cases, because no one would know what…to do with it …

6. Has Reuters institute been involved in research on news personalization or the application of AI to journalism?

Existing research:

We have large amounts of audience research and how people from different countries use news media and how they see the news in the media and of course issues of personalization will come up in that; so for example, our research looks at things like the reliance on search engines and social media and aggregators for news….

Research on people’s levels of literacy about how display decisions are made in environments like social media…all of this is published in the Digital News Report

Regarding AI specifically:

…we have a line of research under the program of Misinformation, Science and the Media , where we have looked at, amongst other things, how the news media itself covers AI in news coverage of new technologies and shown that it is highly driven by industry sources…that’s a line of research that deals more with how journalism covers AI than how AI is used

Future research:

…we don’t currently have new research planned in this area, but we are working through our journalism fellowship programs  and in our leadership development programs on helping individual journalists as well as editors and executives in the news industry think about the choices that they have to make in terms of how they respond to the use of automation and also how they cover it and how they make decisions about which of these technologies they would want to use in their own work , whether that work is reporting, editing or other operations in the news.