In collaboration with the Department of Informatics of Universität Hamburg, the Hans-Bredow-Institut hosted the symposium “The Rule of Algorithms? Formen, Einfluss und Regulierung algorithmischer Entscheidungen [The Rule of Algorithms? Forms, Influence and Regulation of Algorithmic Decisions]” on October 4th, 2016. In the run-up to the international conference of the Association of Internet Researchers AoIR 2016, which started a day later in Berlin as well, German and international experts came together in the Representation of Hamburg in Berlin to approach the topic “algorithmic decisions”.
Algorithms and their assumed or actual power is an important topic of public and scientific debate. Algorithm-based decisions are attributed with structuring, influencing or manipulating various aspects of our lives, such as algorithms deciding on newsfeed content in social media, on deleting hate speech or on our creditworthiness. In social debate on these developments, it is repeatedly criticised that algorithms are non-transparent and, thus, defy control by users. This presents politics and the civil society but also scientific community with challenges.
To begin with, there is the question of what we mean if we speak about algorithms and what risks can be described. Then further questions can be connected to this: How can complex parameter and process of algorithmic decisions be designed in order to understand their activities? Are there any starting points for the regulation of such systems if risks occur – should the software be regulated or the programmers who develop them or the people who use the results? And: how can we research algorithms and their social impacts in a meaningful way? In any case, scientific disciplines have to work together that have barely cooperated so far.
These and other questions were debated by experts from the fields of informatics, social sciences, law and journalism. It was the objective to promote an interdisciplinary understanding of algorithms and to discuss the practical dealings with algorithmic decision-making systems and their possible regulation. Algorithmic decision-making system in journalism served as an example.
The event consisted of an internal, small circle of experts in the morning and a public, German-speaking part of the event with panel discussions and about 150 participants from research, practice and politics.
Researchers from different disciplines (informatics, law and social science) had the opportunity to exchange ideas on this topic in a closed roundtable. The invited participants, including communication researcher Tarleton Gillespie (Microsoft Research, Cornell University), lawyer Niva Elkin-Koren (University of Haifa) and computer scientist and communication researcher Bernhard Rieder (University of Amsterdam), deal with algorithmic decisions, their development processes and effects for a while now within their respective disciplines. Based on topics and questions agreed on in advance, they formulated three statements each that served as food for thoughts in the discussion chaired by Jan-Hinrik Schmidt.
The experts agreed that algorithms as a research topic cannot be seen as something isolated. The development and manufacturing process, the models behind it along with the data that feed into the (often self-learning) algorithms have to taken into account. In addition, the scope of the algorithm cannot be ignored either, since cultures and norms can be very different, such as regarding data protection or the transparency of decisions in usage contents like medicine, public communication or autonomous driving of cars. The access to (mostly private) companies that develop algorithms is immensely important for a meaningful research into this topic. Moreover, there is the danger that the common denominator mainly determines the decisions of algorithms, which are often based on quantitative data. Thus, it is important that developers of algorithms deal with such outliers consciously in order to ensure a variety.
After three thematic blocks, Prof. Dr. Maalej from the Department of Informatics and Prof. Dr. Schulz from the Hans-Bredow-Institut summarised the morning. For Prof. Maalej it was especially important that, besides average values, outliers remain traceable in algorithmic decision processes that are based on big data. Furthermore, he encouraged the participants to not use the term ‘algorithm’ light-minded and to point out about what they are talking. Prof. Schulz concluded that the research on algorithmic decisions concentrates or should concentrate on the decisions and less on algorithms themselves. Moreover, it is important to him that research focuses on the fields of application in which algorithmic decisions cause social consequences.