Paper short abstract:

We collaborate on the problem of presuppositions in user modelling from an ethical and a developer´s perspective. In our work, we try to establish a VSD-orientated approach of the integration of ethical reflection on implicit assumptions and values into the technical implementation of user models.

Paper long abstract:

Our collaboration is set in the context of the SFB/TRR 318 "Constructing Explainability" that deals with the challenges of explainable AI.

In this context, user modelling emerges as a central issue for tailored communication between humans and machines. However, it raises certain ethical concerns as it inherently relies on strong presuppositions about human communication and personality. For example, models from psychology are frequently used and assessed via machine learning techniques, a practice accelerated by technological advancements, yet without sufficient contextual reflection. Consequently, these presuppositions, embedded in the development process, can manifest not only as implicit but also as unexamined and tacit assumptions. Tackling implicit value systems is therefore a central point in an ethical reflection in the development and implementation of user models. To make technical design ethical, the role of the ethicist is to provide ethical expertise and skills (e. g. making implicit assumptions explicit, initiating a change of perspective) whereas the part of the developers is to integrate those insights into practice, by acknowledging the ethically challenging role of tacit assumptions as well as by using them responsibly, i.e. in consideration of possible risks. The approach we choose for addressing these challenges is value-sensitive design (VSD).

Martina Philippi wrote her dissertation in philosophy on the phenomenology of tacit assumptions. Before joining the TRR318, she did ethical accompanying research in a project on rescue robotics with VSD.

Dimitry Mindlin is writing his dissertation in computer science on explaining black box machine learning models in co-constructive dialogues in the TRR318.