PhD defence: The Myth of Algorithmic Regulation

On Friday 17 May at 10.15 hrs., Lukas Lorenz will defend his PhD thesis The Myth of Algorithmic Regulation. An ethnographic exploration of algorithms, actors, and institutions.

The use of algorithms in public organisations comes with risks (as we know from scandals such as the toeslagenaffaire) and the need to better regulate the use of algorithms therefore is evident. This was also made clear by the AI Act that just passed the EU parliament. Lukas Lorenz’ PhD research shows that it is an organisational-institutional process through which the promises and risks of algorithms are being translated into organisational reality and that public organisations should make deliberate decisions about how to organise that process.

The myth of effectiveness and efficiency

A growing number of public organisations in general and regulatory agencies in particular adopt algorithms in the hope that organisational practices will become more effective and efficient. Regulatory agencies are tasked with the supervision of large numbers of potential inspectees. In some regulatory domains, such as food safety, regulatory agencies in the Netherlands need to supervise up to 100,000 businesses with a limited number of inspectors. The agencies are therefore highly reliant on identifying businesses with a high risk of violating rules. Regulatory agencies hope that algorithms may improve the assessment of risks and thereby make regulatory practices more effective and efficient. Yet, realizing the potentials of algorithms often does not result in desired outcomes.

In his dissertation, Lukas Lorenz has explored how regulatory agencies attempt to realize the potentials of algorithms mainly through ethnographic fieldwork at two regulatory agencies in the Netherlands. To make sense of the results, this dissertation distinguishes between the myth of algorithmic regulation and the organisational reality. This enables us to understand the adoption of algorithms not as a technical process but as an organisational-institutional process through which the algorithmic myth is translated into organisational reality.

Institutional patterns of adopting algorithms

That practice differs from the myth needs to be acknowledged and remembered especially with all the hype around algorithms and AI, Lorenz states. Furthermore, it is crucial to look at the translation process from myth to reality.

His research shows that the process of adopting algorithms involves three specific patterns:

• First, conflicting understandings of meanings, norms, and power relations among actors involved in the adoption of algorithms may prevent changes and decouple algorithms from regulatory practices.
   
• Second, learning describes two ways of realizing change in organisations that adopt algorithms: single-loop learning when data science is being connected to other forms of expertise (for example directly to inspectors or domain experts) and double-loop learning when institutional mechanisms are being established for algorithmization (because the exchange between inspectors and data scientists, for example, is not yet institutionalised).
   
• Third, integration refers to how public sector data scientists as professionals react to varying and potentially conflicting institutional logics: they integrate a technological logic with domain and political-administrative logics in their work.

These patterns reveal mechanisms of change in organisations and help to answer the overarching research question: How do actors and institutions shape and are shaped by the adoption of algorithms in public organisations? Actors and institutions are being transformed because of the interactions between actors and their norms, understandings, and power relations on the one hand and algorithms and the powerful qualities attributed to them on the other hand.

Critical value choices

This means that in organisations in which actors attach great value to transparency and accountability, these norms are likely to shape algorithmization processes and may make the adoption of algorithms more difficult than in organisations that mainly promote effectiveness and efficiency, which align with the myth of algorithmic regulation. Adopting algorithms involves making critical value choices beyond effectiveness and efficiency. Safeguarding values, such as transparency, accountability, and non-discrimination, needs to be institutionally enabled. Thus, to benefit from the promises of algorithmic regulation, creating these institutional conditions is a key challenge for public organisations.

Lukas Lorenz is a consultant at PD (the inhouse consultancy of the public sector in Germany) and a PhD student at the Utrecht University School of Governance (USG).