Flow describes a state of complete absorption in a task, which occurs when an individual's perceived skills match the perceived challenge of the task. Because flow is related to increased performance and well-being, it has been studied across multiple domains such as knowledge work, sports, and gaming. However, research indicates that the flow experience differs depending on the characteristics of the task. Yet no framework systematically classifies tasks according to their flow-relevant characteristics.

Thus, the aim of this thesis is to develop a framework that classifies tasks according to characteristics relevant to the flow theory. Students derive flow-relevant task characteristics from existing research and consolidate them into a classification framework.

Flow describes a state of complete absorption in a task, which can be experienced across various activities such as learning, music, or gaming. This state can be captured through questionnaires or physiological measurements such as heart rate or brain activity. Because flow is related to increased performance and well-being, it is of high interest to coaches and institutions in competitive e-sports to support individual and team performance. A system capable of detecting flow in e-sports players could provide coaches and institutions with insights to help players enter or remain in this state. Yet no such flow-supportive system exists in the context of competitive e-sports.

Thus, the aim of this thesis is to develop an artefact for a physiological flow system in a competitive esports environment. Students identify requirements through a systematic literature review and interviews with coaches and representatives of esports institutions, and translate their findings into corresponding design guidelines. Finally, the resulting requirements inform the development of a first prototype system.

Artificial intelligence (AI) in medicine shows potential for various clinical tasks such as imaging, diagnosis, prognosis and treatment planning. Because one treatment may show different effects when applied to different patients, predicting individual treatment effects (ITE) could aid doctors in selecting the right treatment for each patient. This ITE describes the difference between the expected outcome if a patient receives treatment, and the expected outcome if no treatment is prescribed. However, using AI-based ITE predictions in practice requires that doctors understand the system’s reasoning and are able to detect potential errors. This is not the case for many modern, highly performant AI systems, so-called black-box models. Many established methods for explaining AI outputs to promote understanding of black-box AI models suffer from the “consumer-creator gap” (Ehsan et al., 2024), as they were not designed with doctors’ explanation needs in mind. Hence, other approaches to making AI-based predictions of ITE understandable are necessary.

In a previous study, we collected visualizations of medical ITE predictions and conducted interviews with doctors to assess their explanatory potential. Therefore, a qualitative analysis of these interviews based on Grounded Theory Methodology (Wiesche et al., 2017) will be conducted in this thesis. In addition to the interviews with doctors from the domains of oncology, gastroenterology, and rheumatology, interviews with experts in medical didactics will be analyzed in the thesis.

• Ehsan, U., Passi, S., Liao, Q. V., Chan, L., Lee, I.‑H., Muller, M. & Riedl, M. O. (2024). The Who in XAI: How AI Background Shapes Perceptions of AI Explanations. In Proceedings of the CHI Conference on Human Factors in Computing Systems (S. 1–32). ACM. 10.1145/3613904.3642474

• Wiesche, M., Jurisch, M. C., Yetton, P. W., & Krcmar, H. (2017). Grounded Theory Methodology in Information Systems Research1. MIS quarterly, 41(3), 685-701.

Research into Explainable Artificial Intelligence (XAI) relies on distinct families of methods to evaluate the proposed methods. Doshi-Velez and Kim (2017), for example, differentiate between functional-level evaluation that tests properties of XAI methods without human involvement, and human- as well as application-level evaluation, which require humans of different levels of expertise, and respective tasks to match the humans. While evaluation of XAI methods has become more prevalent in recent years (Nauta et al., 2023), methods from the different evaluation levels are seldom combined, especially between the two levels that work with human study participants, and the functional evaluations that rely on quantitative metrics. This leaves a knowledge gap about mechanically quantifiable properties of XAI methods and their influence on human assessment. 

In this thesis, the student will investigate this problem in the domain of diabetes care. First, a systematic literature review (SLR) on functional-level XAI evaluation will be conducted. In parallel, requirements elicitation through interviews with diabetes patients and diabetologists will ensue. Based on these interviews and a functional-level evaluation of representative XAI methods, a prototype for diabetes self-management will be developed. Finally, this prototype will be tested in a think-aloud interview study (Eccles & Arsal, 2018) with a distinct sample of patients and diabetologists. Additionally, lawyers will be included to explore the potential of functional-level evaluation to satisfy regulatory demands for explanations in systems using artificial intelligence.

•  Doshi-Velez, F., & Kim, B. (2017). Towards a rigorous science of interpretable machine learning. arXiv preprint arXiv:1702.08608.

• Nauta, M., Trienes, J., Pathak, S., Nguyen, E., Peters, M., Schmitt, Y., Schlötterer, J., van Keulen, M. & Seifert, C. (2023). From anecdotal evidence to quantitative evaluation methods: A systematic review on evaluating explainable ai. ACM Computing Surveys, 55(13s), 1-42.

• Eccles, D. W., & Arsal, G. (2017). The think aloud method: what is it and how do I use it?. Qualitative Research in Sport, Exercise and Health, 9(4), 514-531.

Although artificial intelligence (AI), and especially Large Language Models (LLM), increasingly show the capability to aid medical decision-making, several obstacles to practical adoption remain. Since medical guidelines provide a framework that indicates which treatments doctors should prescribe under which conditions, AI advice might fall short whenever it recommends a treatment plan that does not conform with the guidelines (Sivamaran et al., 2023). This becomes especially relevant in cases where guidelines do not provide enough detail to cover individual patient circumstance, or have not recently been updated. As guidelines provide legal safety for the doctors, the resulting tendency to adhere to legally undisputable treatments can negatively affect treatment quality and the effect of AI use (Mansi & Riedl, 2025). However, while physicians can deviate from guideline prescriptions in specific cases, little is known about how physicians react to AI advice when it does not match guidelines. It is neither clear, what leads them to consider alternative treatment suggestions, nor what information they require to properly investigate the merit of such a suggestion. Furthermore, while explainable AI (XAI) has been investigated in medical contexts as a means to make AI systems understandable and improve decision quality, there is a lack of research that deals with XAI as a link between data and knowledge in hybrid medical decision support systems (MDSS) that draw both on established medical knowledge and on AI advice.

This thesis addresses these issues through a design science research (DSR) approach. Requirements are derived through a systematic literature review (SLR) on hybrid MDSS and through expert interviews with clinical staff. This phase also includes the development of patient vignettes that represent cases which induce ambiguous decision-making situations for the evaluation phase. To evaluate the final prototype, a round of think-aloud sessions will be conducted to evaluate the prototype hybrid MDSS using those vignettes. All interviews will be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA). 

• Mansi, G., & Riedl, M. (2025). Understanding the Impact of Physicians' Legal Considerations on XAI Systems. arXiv preprint arXiv:2507.15996.
• Sivaraman, V., Bukowski, L. A., Levin, J., Kahn, J. M., & Perer, A. (2023). Ignore, trust, or negotiate: understanding clinician acceptance of AI-based treatment recommendations in health care. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (pp. 1-18).

The increased availability of observational data in the form of electronic health records (EHR) and the rise of advanced deep learning (DL) models have turned clinical decision support based on artificial intelligence (AI) from a distant long-term goal to a tangible one that is already becoming reality. Especially in image-based diagnostics, e.g., in polyp identification and removal in gastroenterology, AI has already found its way to clinical application. For prognostic and prescriptive tasks however, the use of AI lags behind. Training AI to predict individual treatment effects (ITE) promises to enable precision medicine, meaning that treatments are tailored to individual patients by estimating not only the outcome under treatment, but also the outcome under control for a given individual (Shalit et al., 2017). However, similar to other approaches that rely on complex models, the black-box nature of these may cause problems with interpretability and understanding in practice. This can cause adverse effects like faulty treatment decision due to AI errors, when doctors cannot properly assess the reliability of the AI’s advice. Because traditional techniques to explain AI outputs have been shown to be too abstract for lay users, other forms of visualization may be necessary to allow doctors to scrutinize an AI system. In the endeavor to find the right level and mode of explainability for doctors, other perspectives besides their own must be considered to avoid one-sided solutions. Experts in visualization, explainable AI (XAI) and clinical decision support systems (CDSS) may contribute valuable insights on how AI can be made explainable in ways that are approachable and beneficial for doctors.

In this thesis project, the student will conduct a systematic literature review (SLR) to collect a dataset of visualizations that show and explain ITE predictions. In addition, the student will transcribe and analyze a series of interviews we conducted with visualization, XAI, and CDSS experts using grounded theory methodology (Wiesche et al., 2017) (e.g., via tools like MAXQDA).

• Shalit, U., Johansson, F. D., & Sontag, D. (2017). Estimating individual treatment effect: generalization bounds and algorithms. In International conference on machine learning (pp. 3076-3085). PMLR.

• Wiesche, M., Jurisch, M. C., Yetton, P. W., & Krcmar, H. (2017). Grounded Theory Methodology in Information Systems Research1. MIS quarterly, 41(3), 685-701.

This thesis investigates the automatic detection of flow states using machine learning techniques applied to webcam data. Building upon existing research, a classifier is developed and trained on video recordings collected in a controlled experience sampling field study. The trained model is then evaluated on its ability to recognize flow states from unobtrusive webcam input, offering a feasible alternative to traditional measurement approaches. The thesis aims to advance the understanding of flow detection using webcam data.

Treatment Effect Prediction (TEP) emphasizes that merely predicting the outcome of a treatment decision instead of both scenarios – under treatment and under control – is a critical flaw in many machine learning (ML) systems. Although true individual treatment effects (ITE) cannot be directly evaluated, as only one decision outcome can be observed for each patient, TEP has sought to develop techniques that allow for such predictions and their evaluation, e.g., through aggregated benefits of medical outcomes when following model recommendations (Pan et al., 2024). While explainable artificial intelligence (XAI) methods are being used to interpret these models, XAI sees little use when it comes to improving them. For an image classification model, e.g., Ribeiro et al. (2016) showed that Local Interpretable Model-Agnostic Explanations (LIME) were suitable to identify spurious correlations in their training data. The explanation showed, that background information, snow in this case, had been the classifier’s focus instead of the animals shown on the images. To the best of our knowledge, no comparable efforts related to TEP have been made so far. Another possible approach to model improvement could be to remove less important features from the training data (Nauta et al., 2023). On top of attempting to improve older and more recent TEP-based ML models, applying XAI may also help us to understand differences between the different models’ internal reasoning.

In this thesis project, the student will develop three TEP-based ML models (Pan et al., 2024) for cancer patients using data from the Surveillance, Epidemiology, and End Results (SEER) database. The models should include the established causal forest (Athey & Wager, 2019) and the more recent SNB model (Pan et al., 2024). First, a scoping review on ML models for TEP and their evaluation will be conducted. After the implementation and training of the models, the student will apply three different XAI methods to generate explanations for each model and analyze these to find potential for adjustments to the training pipeline. These XAI methods should include Shapley Additive Explanations (SHAP) (Lundberg & Lee, 2017) and Diverse Counterfactual Explanations (DiCE) (Mothilal et al., 2020). Interviews with oncologists will be conducted to find flaws of the models through their explanations and improve upon the model by accounting for their input. The interviews will be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

• Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). " Why should i trust you?" Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining (pp. 1135-1144).

• Pan, H., Wang, J., Shi, W., Xu, Z., & Zhu, E. (2024). Quantified treatment effect at the individual level is more indicative for personalized radical prostatectomy recommendation: implications for prostate cancer treatment using deep learning. Journal of Cancer Research and Clinical Oncology, 150(2), 67.

• Athey, S., & Wager, S. (2019). Estimating treatment effects with causal forests: An application. Observational studies, 5(2), 37-51.

• Lundberg, S. M., & Lee, S. I. (2017). A unified approach to interpreting model predictions. Advances in neural information processing systems, 30.

• Mothilal, R. K., Sharma, A., & Tan, C. (2020). Explaining machine learning classifiers through diverse counterfactual explanations. In Proceedings of the 2020 conference on fairness, accountability, and transparency (pp. 607-617).

• Nauta, M., & Seifert, C. (2023). The co-12 recipe for evaluating interpretable part-prototype image classifiers. In World conference on explainable artificial intelligence (pp. 397-420). Cham: Springer Nature Switzerland.

The theory of flow, which describes the state of being completely absorbed into an activity, has gained considerable attention in recent years across a wide range of domains. As a result, numerous questionnaires have been developed to assess the experience of flow and to account for these different contexts, such as “the flow state scale for occupational tasks” or the “reading flow short scale”. However, the use of a wide variety of questionnaires across studies makes it difficult to compare and synthesize findings and therefore hinders the development of flow research. 

Therefore, the aim of this thesis is to systematically identify existing flow questionnaires and assess their methodological validity. Based on this analysis, a framework will be developed to guide the selection and application of flow questionnaires depending on the research context and purpose.

• Rosas, D. A., Padilla-Zea, N., & Burgos, D. (2023). Validated questionnaires in flow theory: a systematic review. Electronics, 12(13), 2769.

Over the past years, consumer wearables have evolved from lifestyle gadgets into advanced sensing platforms capable of continuous physiological monitoring. Ecosystems like Garmin and Apple Health provide access to health metrics (e.g., sleep score) and raw data (e.g., heart rate) for research institutions. This is of particular importance for researchers interested in predicting diseases (e.g., hypertension) or classifying mental states (e.g., stress or flow), based on physiological data. Among these, flow presents a state of deep task engagement and optimal experience, and can be assessed with physiological indicators, such as heart rate variability (HRV) or respiration rate.

Therefore, the aim of this thesis is to develop a Garmin application that records physiological data and correlates it with user-reported flow experiences. Students will receive access to the Garmin Developer Portal and will implement a mobile application capable of collecting HRV-related metrics and questionnaire-based self-reports. The application will then be evaluated in a small pilot study to explore relationships between health metrics and flow data.

• Henriksen, A., Haugen Mikalsen, M., Woldaregay, A. Z., Muzny, M., Hartvigsen, G., Hopstock, L. A., & Grimsgaard, S. (2018). Using fitness trackers and smartwatches to measure physical activity in research: analysis of consumer wrist-worn wearables. Journal of medical Internet research, 20(3), e110.

While feature attribution methods like SHAP and GradCAM see widespread application for image data, they do not constitute the only class of explanation methods for images. Another category of methods relies on learned prototypes that capture relevant patterns in the images. These prototypes are supposed to be more interpretable than traditional methods for explainable artificial intelligence (XAI) and offer better capacity to detect shortcomings in classification decisions. A typical application of prototype models is the CUB-200 dataset that contains images of 200 different bird species (Nauta et al., 2021). 

In this established setting, the student will implement and train two models for bird classification and extract prototype and feature attribution explanations. These will be evaluated through a series of expert interviews with bird enthusiasts to investigate their alignment with human explanations and their understandability for experts in that field. The interviews will be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

• Nauta, M., Van Bree, R., & Seifert, C. (2021). Neural prototype trees for interpretable fine-grained image recognition. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition (pp. 14933-14943).

Although the field of explainable artificial intelligence (XAI) has developed a plethora of explanation methods since its inception, many of these cannot easily be transferred into practical use by non-experts. Having been developed by AI researchers with little attention to how human explanations work, common methods like LIME and SHAP offer insights in a form that does not align with the general expectations and needs of practitioners (Ehsan et al., 2024). Other methods like counterfactuals or narratives however are, at least in theory, closer to human explanations. Still, it is not properly understood how practitioners interpret the various XAI methods and what information they can extract from them. To investigate this problem, this study will look at explanation methods applied to neural networks for treatment outcome prediction in oncology and rheumatology. 

The student will implement and train neural networks and three different explanation methods. To evaluate how they are perceived by medical professionals, the student will develop interview guidelines and conduct a series of expert interviews with medical professional from the fields of oncology and rheumatology. The interviews must be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

• Ehsan, U., Passi, S., Liao, Q. V., Chan, L., Lee, I. H., Muller, M., & Riedl, M. O. (2024). The who in XAI: how AI background shapes perceptions of AI explanations. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems (pp. 1-32).

The public availability of open government data (OGD) constitutes a major achievement for transparent government work and official accountability. Using OGD opens the possibility for civil society stakeholders like researchers, citizens and journalists to not only have a watchful eye on public spending, but also to use the data for other projects or to co-design solutions in cooperation with government partners. However, while OGD is indeed publicly available through designated online portals, these portals do not yet completely fulfill their purpose. They are too complex, difficult to navigate, and lack vital features like visualization and data analysis tools, leading to entry barriers for non-technical citizens. Integrating natural language interactions into these portals could be a helpful feature to make them more accessible to the general public. In addition, explanations on how to adapt the natural language queries to better match the desired data might also be useful to ensure trustworthy and understandable interactions.

In this thesis project, the student will follow a design science research (DSR) approach to develop an accessible OGD portal for financial data. First, the student will conduct a systematic literature review (SLR) on natural language features for explainable user interfaces. Then, further requirements will be elicited and refined through several formative steps involving interviews and design workshops. Finally, the developed prototype will be evaluated through a set of summative interviews. The interviews must be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

Flow is the experience of being fully focused on an activity and fluidity of action. In this state, individuals often lose track of time and perform at their optimal levels. Thus, the flow state is of particular interest to e-sport players looking to enhance their performance. Therefore, technologies that are able to detect the current flow state of the user and help them enter or maintain this state could provide a valuable benefit, leading to improved performance. This thesis employs a design science research approach to develop a flow-adaptive system.

Flow, the state of optimal experience and complete absorption in an activity, is of growing interest in information systems research. Recent studies have shown that flow states can be classified using machine learning models trained on physiological data, such as heart rate and heart rate variability (HRV). For instance, Rissler et al. (2020) trained a flow classifier using a random forest model and achieved an accuracy of 70%. Traditional machine learning approaches often rely on hand-crafted features (HCFs), such as standard HRV metrics like SDNN or RMSSD. However, these features require expert knowledge and are labor-intensive to compute. Feature learning methods, such as deep neural networks, present a promising approach to overcome these limitations due to their capability to automatically extract relevant features. Therefore, feature learning approaches may outperform HCFs, in particular when dealing with large-scale, noisy, or unstructured data. 

The aim of this thesis is to investigate the differences between HCFs and feature learning approaches for classifying flow states from physiological signals. Students working on this project will have access to a publicly available flow dataset.

Rissler, R., Nadj, M., Li, M. X., Loewe, N., Knierim, M. T., & Maedche, A. (2020). To be or not to be in flow at work: physiological classification of flow using machine learning. IEEE transactions on affective computing, 14(1), 463-474.

This thesis investigates the automatic detection of flow states in gaming using machine learning techniques. Building upon existing research that links physiological signals, such as heart rate and heart rate variability, to the experience of flow, a classifier is developed and trained on a publicly available flow dataset. The trained model is then evaluated in a laboratory study, in which participants are measured repeatedly across several gaming sessions and receive feedback on their flow states during play. The study aims to advance understanding of flow in gaming and to explore the potential impact and benefits of flow feedback for players.

Implementing artificial intelligence (AI) models for medical tasks like diagnosis or prognosis promises to support medical staff in their decision making and improve the overall quality of healthcare. However, in order to achieve effective usage of AI for decision support, some potential problems have to be resolved. These include issues of trust, overconfidence, and legal requirements. A popular approach to make AI models more trustworthy, transparent, scrutable, and generally understandable lies in AI explanations. The research community of explainable AI has developed a wide array of methods that attempt to extract valuable insight about the reasoning of any given AI model. While most scholars have developed methods that attribute importance values to individual features to indicate their significance for a given prediction or for the global model behavior, others have taken inspiration from the social sciences and tried to construct more intuitive, human-like explanations. Among these are counterfactual explanations, also known as contrastive explanations. These provide alternative sets of minimally changed inputs that lead to a different model output. Presenting diverse counterfactuals enables insight into the model’s reasoning process in a different way to attribution-based approaches.   

This Master thesis project consists of a systematic literature review (SLR) on contrastive explanations for AI models in the field of medicine. Based on this SLR, the student identifies requirements for the development of an interface that provides contrastive explanations for a specific medical task. Expert interviews with medical professionals will be conducted, recorded, transcribed, and analyzed (e.g., via tools like MAXQDA) to evaluate the developed interface.