We used to be told that we live in a media-saturated world. True as this may be, today’s is also data saturated. Johannes Bjerling (JB), editor of Nordicom Review, discusses datafication in the Nordic welfare societies with Rikke Andreassen, Anne Kaun, and Kaarina Nikunen.
Johannes Bjerling: Today I’m talking with Rikke Andreassen, Anne Kaun, and Kaarina Nikunen, the authors behind an article that was published by Nordicom Review late last year. The article, “Fostering the data welfare state: A Nordic perspective on datafication”, deals with how data is collected and used by public authorities and institutions in the Nordics, and some rather critical concerns are being raised. However, before digging deeper into this, I think we should say a few words about the overall setting. The Nordic region is known to be a digitalised region and digitalisation is, to some extent, a prerequisite for large-scale datafication?
Anne Kaun: It’s true that digitalisation enables datafication as we today discuss and think of it, but datafication has a long history in the Nordic countries – extensive data collection lay behind attempts at social engineering and policy-making – so it would be wrong to suggest that without digitalisation there would be no datafication. You are, however, right in that the form we discuss in the article is enabled by digitalisation.
Johannes Bjerling: Okay, so how should we understand datafication? How do you define it?
Anne Kaun: In the article, we rely on a definition that was originally put forth by José van Dijck. Datafication, in this view, refers to the process of using data to understand social life and behaviour. When we interact with each other through digital devices, how and when we communicate and search for information, it’s all being datafied and, in this way, rendered measurable. Rather than the content, what we say to each other, it is the metadata – how and when we commute, how and when use our phones – that are of interest to the companies that track us.
Johannes Bjerling: And since we spend more and more time with digital devices, there are more and more traces to collect. But in itself, the collection and quantification of data is not very problematic, is it? So, what’s the problem?
Anne Kaun: One problematic aspect is the underlying assumption that data are objective, neutral, and provide an unbiased image of “the world as it is”. This assumption – the belief in unfiltered data – is often referred to as dataism, and central to dataism is the belief that data on the aggregated level efficiently predict the behaviours and actions of individuals.
One problematic aspect is the underlying assumption that data are objective, neutral, and provide an unbiased image of “the world as it is”.
Johannes Bjerling: And in this way, the categorisation and grouping of individuals has become increasingly important? Rikke?
Rikke Andreassen: Yes, because data needs to be processed in order make sense, it is termed “raw data”, but it is still processed, among others ways, according to categorisation and grouping. Instead of accepting the notion of pure and neutral data – data that merely reflect who we “are”, what we “do” – we question the very logic behind this mapping. To put it somewhat differently, we question the assumption that categorical belonging is an appropriate way to predict what individuals will do. Age, gender, ethnicity, whether you’re married or not, whether you were married when your kids were born – this mapping, based on a set of predetermined parameters, doesn’t say very much about you as an individual; what it does is group you according to certain principles. And while this logic (it is how your typical recommender systems work) may not be conceived of as problematic when it comes to services from which we can choose to opt-out, it becomes, we argue, problematic when it is applied by public authorities and welfare services. When automated decision-making, ADM, is used to calculate the risk of, for instance, long-term unemployment, it is more problematic than when ADM is used for online shopping.
Johannes Bjerling: That’s one of the cases that you discuss in the article: how automated decision-making is planned to be used by unemployment services in Denmark…
Rikke Andreassen: That’s right. Since there is, on the aggregated level, a correlation between certain factors and long-term unemployment, these factors are used to calculate – predict – the risk of unemployment on the individual level. In practice, if you score high on these factors, you are more likely than a person who scores low to be long-term unemployed.
Johannes Bjerling: And this way of thinking is, as you see it, wrong?
Anne Kaun: There are, we say, several reasons as to why this is problematic, and one of them is the risk of correlations replacing explanations. Should this happen, we don’t develop any deep or thorough understanding; crime rates, to use an often-discussed example, may be connected to certain neighbourhoods and areas, but without taking more complex issues – such as segregation and living standards – into account.
Rikke Andreassen: Another problematic aspect is related to the lack of transparency. In relation to public authorities, we should always be able to ask why a certain decision was reached – why our social benefits were cut, why our unemployment benefits were reduced – but when decisions are automated, there is the risk that the arguments underlying decisions are neither clear nor explicit. Anne used the term “understanding”, and the term highlights a crucial difference between humans and machines: Whereas humans have the ability to see “the big picture”, machines can calculate. If citizens are unhappy with a decision, they can ask for explanations of why and how that decision was made. In case of the “black box”, citizens and welfare workers do not know the rationale behind a particular decision, and it is not possible to get to know it.
Another problematic aspect is related to the lack of transparency. In relation to public authorities, we should always be able to ask why a certain decision was reached – why our social benefits were cut, why our unemployment benefits were reduced – but when decisions are automated, there is the risk that the arguments underlying decisions are neither clear nor explicit.
Johannes Bjerling: I guess “efficiency” is an argument behind automated decision-making? Machines need no coffee breaks, they are never sick, and replacing social workers with machines is therefore believed to be more cost-efficient?
Rikke Andreassen: Efficiency is a key term here, and with efficiency, reduction of costs. We need to keep in mind that one of the main reasons for implementing ADM is the belief that it will reduce costs. But so far, experiences with public digitalisation have not been cost-effective; rather, they have been very expensive.
Johannes Bjerling: Alright, let’s move on. We have talked about datafication in the social service sector – another case in your article is the corrections sector. I would, however, like to hear more about your third case, datafication within public service media. Kaarina, I’ve understood that YLE has come quite far when it comes to datafication?
Kaarina Nikunen: Well, datafication – or algorithmic decision-making – is probably a reality in the public service institutions of all Nordic countries, not only in Finland. But datafication within public service organisations is, we think, particularly interesting, not least since it brings about some specific challenges. The very model of public service media – media funded by public money, and there to serve the public – makes datafication within this institution a case of particular interest.
Johannes Bjerling: Okay. What you mean is, I guess, that the very idea behind public service media – the roles these media are to play within Nordic welfare societies – is what makes them particularly interesting to study. So, what about them? Or, more specifically, what about datafication at Nordic public service institutions?
The four pillars of the Nordic media systems
Kaarina Nikunen: In The Media Welfare State, by Syvertsen, Enli, Mjøs, and Moe, four pillars of the Nordic media systems were discussed. The pillars they discussed were universal access, editorial freedom, content diversity, and policies based on deliberation and consensus. However – and as mentioned earlier – the Nordic region is a region where digitalisation has now come very far. The question we raise is whether the media welfare state is about to be replaced by the data welfare state. More specifically, to what extent are the pillars of the media welfare state valid also for the data welfare state? Do the same pillars hold – or would other pillars be needed for there to be a data welfare state?
Johannes Bjerling: Right. And relating what should be central to the data welfare state to datafication as we know it today, you also provide us with an answer to the question of whether the data welfare state is here. So, what about the pillars? Are there any differences between the pillars of the media welfare state and the data welfare state?
Kaarina Nikunen: The first pillar of the media welfare state, the one related to universalism and public service media as a meeting place – a public sphere, so to speak – is in a straightforward way challenged by datafication and algorithmic services. Through data-driven services, we today receive different media and content; personalisation and recommender systems have, in this way, led to fragmentation. Consequently, the first pillar that we discuss for the data welfare state is – instead of universalism – related to fairness and justice. For there to be a data welfare state to speak of, processes of datafication must be unbiased.
Johannes Bjerling: And this is, as we have discussed, at odds with how datafication works today. Anne mentioned the example of crime, and how datafication in relation to criminality can give rise to negative feedback-loops: Since crimes rates are known to be high in certain areas, more policing is carried out there, whereby more crimes are documented and more polices are sent.
Rikke Andreassen: Yes – and a consequence for those who live in these areas is that they, to some extent, lose their individuality. Instead of being conceived of as individuals – with different dreams and ambitions – they are conceived of as residents of a problematic neighbourhood.
Johannes Bjerling: And the second pillar of the data welfare society?
Kaarina Nikunen: The second pillar that we discuss is decommodification. Today, news media organisations all over the world depend on software and platforms that are provided by large, commercial companies. While this may be fine for the news media that are privately owned, the blurring of private and public interests is more problematic for public service media. When private companies gather data, the underlying reason is, of course, to make money. Data gathered by public service organisations should, if gathered at all, be gathered with the public interest in mind.
Johannes Bjerling: And the third pillar?
Kaarina Nikunen: Whereas the third pillar of the media welfare state was content diversity, we say that the third pillar of the data welfare state should be data diversity. There are similarities between this pillar and the first, but this pillar departs from the fact that our contemporary societies are marked by heterogeneity rather than homogeneity. While processes of datafication must be fair and unbiased (that’s the first pillar), they must also be inclusive and aiming to serve different people’s needs. In essence, the third pillar relates to principles of equality – the notion that data is equally important no matter age, gender, income, etcetera, etcetera.
Anne Kaun: Discussing this pillar, we can also go back to the example of automated decision-making in relation to social benefits. Here, the procedures are all very standardised; it’s the same kind of data – or parameters – that determine decisions and outcomes in each and every case. But if we conceive of welfare provision in terms of supporting capabilities, where different citizens may need different support, it becomes evident that standardisation is problematic.
Johannes Bjerling: One size does not fit all – or, in this case, one set of data does not serve everybody’s needs?
Anne Kaun: Right. And this means that we’re back at the important difference between computers and professional social service workers: Computers may be good at counting, but they can’t take into consideration what has not been put into the system.
This means that we’re back at the important difference between computers and professional social service workers: Computers may be good at counting, but they can’t take into consideration what has not been put into the system.
Johannes Bjerling: Only humans can see the bigger picture?
Anne Kaun: Yes.
Johannes Bjerling: Then, what’s the fourth pillar of the data welfare state? You’ve mentioned the need of fair and unbiased datafication processes; following this we discussed decommodification and data diversity – finally, what is the fourth pillar that you have in mind?
Kaarina Nikunen: We discuss it in terms of transparency and sustainability, and whereas we have touched upon the lack of transparency as problematic in relation to the provision of social benefits – why the “black box” says “no” can’t always be figured out – we haven’t discussed the sustainability aspect. In essence, this pillar shall be seen against how we know the big tech companies of today work: The environment in which they operate is one where short term goals and short term profits are always prioritised above long-term solutions and sustainability. There is always a new product to sell, always a new update to promote.
Rikke Andreassen: And this tendency of always looking ahead, never settling with what we have, can also be related to the transparency issue: Since these companies are driven by the quest for market shares and profit, opening-up the black box is not in their interest. How their systems work, why their algorithms rank the way they do, well, from the perspective of big tech firms, there is rationality in keeping the answers a secret – giving them away would be doing their competitors a favour.
Johannes Bjerling: Given what you’ve said, and the four pillars that you discuss in the article, it seems as if the question of whether we today live in data welfare societies can only be answered by a “no”?
Rikke Andreassen: That’s right. The data welfare society – or “state” – is not yet here; certain aspects of datafication – as we know it today – are actually in conflict with central values of the welfare model.
Johannes Bjerling: I’ve learnt a lot from talking to you, there are many aspects to discuss when it comes to datafication, and I hope your article will find many readers in the Nordics, and elsewhere. So, thank you – for talking to me as well as for submitting your paper to Nordicom Review!
Rikke Andreassen: Thank you!