By Emilie Mørch Groth, Junior Researcher
In marketing there is a saying along the lines; “if you can find the brand in Netto, then you know it has left the trendsetting sphere and entered the realm of mainstream” – for better or worse.
I will concoct a similar saying for technology; “if you can find a technology used internally in your local municipality, then you know it is no longer emergent tech, but rather mainstream technology”.
AI is one such technology; it’s becoming widespread throughout the public institutions with a great political focus and quite the budget backing it.
The centrepiece in this project is the 40 Signature Projects, which are administered by the Agency for Digital Government under the Ministry of Finance (Digitaliseringsstyrelsen) and financed by the Danish National Uptake Fund for New Technologies. This investment fund is backed by taxpayer money and has between 2020 and 2022 distributed 180 something million DKK to 40 different AI projects around the country led by either municipality or region – all described on Digitaliseringsstyrelsen site.
Through these projects (and those predating them), we see AI playing a bigger role in public society as it is made to handle and/or assist a growing part of workload of the welfare state; from auto-archiving emails, to optimising ultra-scans of pregnant people and from enabling CO2 saving scheduling of vehicles to flagging cases with children at risk.
It is not everywhere, but it is definitely more there.
AI is not without controversy, and due to the complex nature of the technology, I identify a potentially damaging knowledge gap between the people who design, build, and deploy AI and people who are affected by AI. This made me want to take a closer look at the Signature Projects to understand them better, and to understand their impact on the Danish society and its citizens.
I initially started with the questions: What are the implications of the knowledge gap between AI-driven solutions used in the state, the employees working with them, and the citizens affected by them? Could bridging these gaps influence the process of how the state employs AI-driven solutions?
This soon turned into a mapping exercise, where I wanted to dive into these projects and map them in order to bring nuance into the debate of public AI – to map a snippet of current landscape of deployed AI in the Danish state.
But that journey was off to a rough start.
The journey
In order to map the Signature Projects; my initial step was to build a registry. So, I went to look for information on them; datasets used, areas of deployment, project owners and partners, budgets and so on and so forth.
I read through the project descriptions at Digitaliseringsstyrelsens page, and I started plotting the information into a spreadsheet. It quickly became clear that some information was hard to categorise while other was plain out missing. This became an obstruction, and it altered my process greatly. Let me show you; this was my imagined journey – naĂŻve as it might be:
The orange dot represents where I initially imagined the project to be at the time of writing this blogpost. I ventured into this believing that I would spend most of my time fiddling with an interactive map – curating the experience to optimise the value of it. An iterative process of tinkering with the data I would collect, map, test and change it. But alas; my route turned out quite different. Something like this:
As you can see, I can’t even keep it to one dot; a part of the project is very much still in the “find info” stage, while I have also started to play with the mapping part of the project in order to feel progress. A dissatisfying process – being so far from the target? No, not really. This straying from the path has proven to be the most interesting part about the project, and it has raised the most interesting question; why is the information so hard to come by?
I took on the role as detective, digging around the internet for the information not listed in the overview from Digitaliseringsstyrelsen. There were patterns in the vagueness, but I have yet to uncover the reasons behind this. I am not contemplating ill intend, I am more considering arguments of protection of intellectual property, business secrets, simple differences in institutional procedures, NDAs, an unwillingness to disclose private companies by name, as to not brand them, etc.
In alignment with this last bit of speculation; one pattern in particular was that stakeholders were hard to pinpoint. Which prompted my interest even further, and it also asked me to take a step back to not only question the anonymity of certain stakeholders, but also question the baseline argumentation of these projects to begin with.
The new questions
Through the process and the obstructed journey, my research questions morphed, and I started to look to beyond my initial thoughts: Why are these projects set assail in the first place?
Is it to save money in the name of automation? To brand a certain region as extraordinarily digital? Or to foster relationships between the private and public sectors? Who are the drivers? Who benefits? What did we learn – if we learnt anything?
There are many questions to be posed, and I will let a piece of poetry guide the way. Hang with me: I was at a casual lecture at my old højskole the other night with Anders Kjærulff and Jesper Balslev from Analogiseringsstyrelsen, and through a poetic reading of their newly published book, my mind got obsessed with one sentence in particular: “AI hylder det middelmådige og det forudsigelige”, I am paraphrasing as my memory is not limitless, and an English version: “AI salutes the mediocre and the predictable”.
There is a discomforting truth to this, and it sheds a very different light on the “endless promises” of AI. It is not limitless and only ever boundary-breaking, in this perspective it is quite the opposite; it is an enabler and reinforcer of the mediocre and predictable. Within this lures the risk of reinforcing stereotypes, excluding and mistreating marginal cases and creating data cascades, where the negative and erroneous consequences only surface way down the line – and often too late. Hence, why we need data accountability, not only here and now, but trailing all the way back to the initial data fed into the algorithms and stretching further out to unforeseen future implications.
So, I will pose a new question; why are we so obsessed (at least politically) about digitising our society and all processes within? Should we maybe think twice before deploying AI in all cracks and crevasses throughout society and our shared institutions? And should we not demand accountability throughout these processes?
With these questions in mind, I turn my gaze towards the stakeholders, and through them, a trail of algorithmic accountability.
The preliminary result
The information gaps hindered my exploration into the mapping exercise itself. Instead of mapping away, I have created a baseline for it, by building a vector graphic based map of Denmark outlining all municipalities – and yes, it took an extensive amount of time, but while clicking away in Illustrator I also had the opportunity to learn the names and locations of all 98 Danish municipalities. Neat, right?
So far, I have pinned down 81 stakeholders in total (explicitly mentioned in project descriptions), and 9 are still to be named. When these are identified, it will be time to utilise the template.
Preliminary map, snapshot, municipalities only, colour-intensity by no. of projects
The next step
The lack of transparency came as a surprise. Taking into consideration that these projects are funded by the investment fund – established with taxpayer money – I assumed all information would be readily available for someone like me – a curious abiding taxpaying citizen with research in mind.
The next step will be to fill the most hollowing gaps of my spreadsheet; to locate the still anonymous stakeholders of the 40 signature projects.
My method will change, and I will leave the desk-research and reach out into the world. I will contact individuals somehow involved in the signature projects; to ask them directly if they know about other stakeholders or can tell me explicitly what entity is involved with the AI and data sets. If this cannot provide me with the answers I seek, I might turn to seeking access to project documents trough public sources – but I know this can a longsome process.
And then of course I will progress in my map making – trying out different variations and playing with the different data options, exploring how to bring different nuances into play.
The afterthought
I am not a technophobe, I genuinely find tech and data interesting, but at times also frightening. I see lack of attention to detail, failure to take future consequences into account and a general lack of care for the individuals and groups implicated in the algorithmic mess – and it scares me. This is my motivation to keep on digging; to add more care and afterthought into the mix. To ask questions and to push for a different future approach with emphasis on transparency and accountability.
The sources
Digitaliseringsstyrelsen (no date). Signaturprojekter med kunstig intelligens i kommuner og regioner. Digitaliseringsstyrelsen: https://digst.dk/digital-transformation/signaturprojekter/ (Accessed 14.12.22)
Analogiseringsstyrelsen (no date). Om Analogiseringsstyrelsen. https://analogist.dk/ (Accessed 14.12.22)