When users integrate advice from artificial intelligence (AI) models into their decision-making processes, they usually achieve better results compared to their standalone performance. Adding explainable AI (XAI) to this mix raises user performance even more. This has led to the conclusion, that XAI enables users to better understand the underlying AI system and avoid mistakes that the AI makes. Upon closer inspection however, a different conclusion emerges. In many cases, the AI by itself still outperforms any human-AI or human-XAI team (Vaccaro et al., 2024). Hence, it may not be increased human understanding that drives the improved performance, but simply higher reliance on a better partner. As potential reasons for such overreliance on AI, the literature discusses several biases, such as confirmation bias or automation bias (Bertrand et al., 2022). Confirmation bias means that when a recommendation is already present, explanatory information is usually interpreted to support that recommendation (Wang et al., 2019). Automation bias, also termed automation complacency, means that AI-generated recommendations are perceived as more reliable, but also as more convenient than applying one’s own judgement (Romeo & Conti, 2026). In both cases it is not understanding and careful assessment that drives decision making, but cognitive biases that may lead to overreliance on the AI and thus to worse decision quality. 

To mitigate this problem of overreliance on XAI-enhanced AI advice, researchers have proposed diverse bias mitigation strategies. In this thesis project, the student will implement a prototypical interface for an exemplary bird classification task that integrates XAI and a selection of bias mitigation strategies. This implementation will be based on a scoping review of relevant literature and preliminary interviews with psychologists. After the implementation, the student will evaluate the interface with bird experts and psychologists with expert knowledge on biased decision making through think-aloud sessions. The interviews and think-alouds will be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

• Vaccaro, M., Almaatouq, A., & Malone, T. (2024). When combinations of humans and AI are useful: A systematic review and meta-analysis. Nature Human Behaviour, 8(12), 2293-2303.

• Bertrand, A., Belloum, R., Eagan, J. R., & Maxwell, W. (2022). How cognitive biases affect XAI-assisted decision-making: A systematic review. In Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society (pp. 78-91).

• Wang, D., Yang, Q., Abdul, A., & Lim, B. Y. (2019). Designing theory-driven user-centric explainable AI. In Proceedings of the 2019 CHI conference on human factors in computing systems (pp. 1-15).
• Romeo, G., & Conti, D. (2026). Exploring automation bias in human–AI collaboration: a review and implications for explainable AI. Ai & Society, 41(1), 259-278.

Marketing professionals make decisions for campaigns that cost hundreds of thousands, or even millions of euros. When artificial intelligence (AI) is used in this context, e.g., to find demographics to target in a campaign, the reasons for any final decision need to be clear and understandable. As marketing professionals have little AI expertise, methods for making model recommendations understandable are required. This is the goal of explainable AI (XAI). However, many XAI methods suffer from the “consumer-creator gap” (Ehsan et al., 2024). Although researchers have created these methods for lay users, they are fundamentally shaped by researchers’ more technical explanation needs. Hence, they can be difficult to interpret for marketing professionals. A potential solution for this is employing large language models (LLM) as explainers. Receiving information in an interactive, dialogue-based format that enables marketing professionals to ask follow-up questions could provide benefits for understanding and decision-making in customer segmentation. 

In this thesis project, the student will conduct a systematic literature review (SLR) on XAI in marketing. Subsequently, a LLM to explain customer segmentation models and the respective XAI augmentations will be trained and evaluated through a series of expert interviews. 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).

Flow is the experience of being fully focused on an activity and the fluidity of action. In this state, individuals often lose track of time and perform at their optimal levels. For software developers whose productivity relies on attention and engagement in the task, entering a flow state can improve their work efficiency and well-being. Therefore, technologies that help software developers to enter a flow state could provide a valuable benefit, leading to reduced mental workload and improved performance 
 

Artificial Intelligence (AI) is being increasingly utilized in various domains. The high performance of deep learning models in tasks such as prediction and classification has largely facilitated this trend. However, this evolution of models becoming increasingly complex has led to a problem of low model understandability, and thus a lack of trust, concerns about possible biases, and even potential regulatory obstacles. This can become a problem when marketers use AI to optimize their campaigns, e.g., by asking it for feedback on the effectiveness of their product branding or by applying AI to guide their resource allocation strategy. Using AI responsibly in this context requires understanding how the respective model reaches its conclusions. Which input features have a positive effect on the predicted buying power of potential customers? How large would the predicted size of the target population be under slightly different circumstances? Explainable AI (XAI) provides a range of methods to enhance model interpretability and promote understanding, enabling answers to these and related questions. To make better use of the available methods, research calls for the development of interactive systems that support a variety of follow-up and drill-down actions. Interactivity is designed to make explanations more human-centric, enabling users to engage in a dialogue with the AI system. 

This thesis project reviews existing XAI methods for applications in the marketing domain in a systematic literature review (SLR). Based on these insights and interviews to elicit requirements for the project, the student develops an AI-based interactive system for marketing data that utilizes a selection of XAI methods found in the SLR. The student evaluates their developed system by conducting a series of expert interviews. These will be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

Similar to other predictive tasks, the field of treatment effect prediction (TEP), which attempts to not just predict a singular outcome, but the difference between two or more different counterfactual outcomes, can also benefit from improved performance through deep learning (DL) models. The downside to this manifests in the reduced interpretability of DL models, which can impede the usability of DL-based TEP in high-stakes decision-making contexts like medicine, that require human users to understand the tools they use and be able to detect whether a prediction is based on sound reasoning and thus trustworthy. Although researchers have developed a range of Explainable Artificial Intelligence (XAI) methods, these are subject to various concerns about model faithfulness and their actual usefulness to end users. We intend to specifically address the use case of TEP for the prognosis of cancer treatment outcomes, and explore how visualizations of treatment effects found in the available literature can support user understanding. 

In a previous project, we curated a dataset of visualizations used to represent predictions of treatment effects. In this thesis project, the student will conduct a series of expert interviews with oncologists and discuss the curated visualizations concerning their helpfulness and accessibility. The interviews must be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

The field of Explainable Artificial Intelligence (XAI) has rapidly gained traction over the last years. Motivated by the promise of increased performance through using Artificial Intelligence (AI) and the associated problems of low interpretability hindering adoption in practice, methods, e.g., for generating feature-importance values or heatmaps to indicate important parts of images, have been proposed by researchers aiming to increase the usability of AI. However, these explanation methods have seen criticism about facilitating a “consumer-creator” gap (Ehsan et al., 2024), targeting the needs of data scientists and AI engineers, but not the needs of end users such as medical professionals. Researchers like Miller (2019) have pointed out similar problems with XAI being too static and not considering the nature of human explanations. 

In this thesis project, the student will conduct a systematic literature review (SLR) on the role of theories from the social sciences on explanations in XAI research. Using this knowledge as a reference, the student will prepare interview guidelines and conduct a series of semi-structured interviews with doctors to gain a better understanding of how explanations work in the medical context. These interviews will 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, May). 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).

Miller, T. (2019). Explanation in artificial intelligence: Insights from the social sciences. Artificial intelligence, 267, 1-38.

Unlike in high-stakes decision-making contexts like medicine or law enforcement, where tabular data is prevalent and commonly available, stock market analysis relies on transparent access to the associated longitudinal data. Similar to the development in different domains, researchers are also attempting to increase predictive performance through the use of artificial intelligence (AI) in the detection of anomalies in time series, thereby reducing the risk of erroneous decisions by human end users. However, the low interpretability of the underlying AI models, if not properly addressed, can also lead to problematic outcomes. If end users cannot detect erroneous reasoning within an AI model’s anomaly detection process, they either tend not to use the system due to their low confidence, or they tend to put too much trust into the system due to not being able to question its outputs. To mitigate both of these problems, researchers have developed Explainable AI (XAI) methods that aim to make AI models scrutable and understandable to human end users. A majority of these methods, though, are intended for use on tabular data.

This thesis project reviews existing XAI methods for time series data in a systematic literature review (SLR). Based on these insights and interviews to elicit requirements for the project, the student develops an AI-based anomaly detection system for stock market data that utilizes a selection of XAI methods found in the SLR. The student evaluates their developed system by conducting a series of expert interviews. These will be recorded, transcribed, and analyzed (e.g., via tools like MAXQDA).

Desk work associated with prolonged sitting and visual strain is linked to musculoskeletal complaints and reduced well-being. Existing solutions address these issues, but typically focus on a single function, such as posture reminders or micro-break prompts. The aim of this thesis is therefore to develop a webcam-based sensing system able to support the well-being of desk workers. To address this gap, an exploratory literature review to derive requirements for a modular, privacy-preserving architecture is conducted. On this basis, a system is built that can simultaneously measure posture, gaze, and blink rate, in addition to heart rate via remote photoplethysmography, to provide the user with feedback.

Contactless vital parameter estimation, such as heart rate, has gained substantial relevance in recent years, as it enables unobtrusive monitoring with widely available devices, such as webcams or smartphone cameras. Within this domain, remote photoplethysmography (rPPG) has emerged as a promising technique. However, the practical applicability of rPPG is still limited by critical sources of error, such as motion artifacts and illumination variability. Both effects reduce the reliability of parameter estimation in real-world conditions. Therefore, the thesis aims to explore how the robustness of rPPG-based heart-rate estimation can be improved under motion and illumination variability.

Modern smartwatches and smart rings increasingly rely on photoplethysmography (PPG) algorithms to measure vital signs such as heart rate, sleep patterns and stress levels. These sensors are particularly attractive because they are non-invasive and inexpensive, which makes them suitable for integration into everyday life. In recent years, advances in signal processing and machine learning have significantly improved the ability to extract meaningful insights from raw PPG data. For example, the first commercially available smartwatches are now able to measure blood pressure, offering an exciting solution for health monitoring. However, most commercial consumer devices do not allow researchers and individuals to access the raw PPG signal, significantly limiting the potential for innovation.

The aim of this thesis is to first define the potential of PPG-based sensors through a systematic literature review. Subsequently, a prototype will be developed that enables the streaming of raw PPG data in real time.

Note: A PPG sensor is provided for technical implementation.

Smart camera-based monitoring systems are capable of tracking and monitoring human activities. In recent years, systems have been developed that enable contactless monitoring of vital signs such as heart rate and oxygen saturation, capture human behavior such as posture and eye activity, and are able to detect human emotions. In contrast to other monitoring systems, cameras have the advantage of being low-cost and unobtrusive, which facilitates their implementation in different environments. Despite the technological capabilities of these systems, their use in the real world is currently limited. 

Therefore, the aim of this Bachelor thesis is to identify use cases and their potential in real-life scenarios. In order to successfully complete this Bachelor thesis, use cases will first be identified through a systematic literature review and then their potential will be explored through an online survey (e.g. LimeSurvey).

In order to properly utilize performance improvements through the adoption of artificial intelligence (AI) models, a number of conditions must be met. Since modern deep learning systems are opaque and inscrutable to human users, problems of mistrust and corresponding non-use can arise. But even if the adoption of AI technology into clinical practice is not hindered by such barriers, problems may arise due to an attitude of overconfidence and overreliance on AI results. The explainable AI (XAI) community strives to develop methods that help to create an appropriate level of trust in AI systems. Such methods are particularly important in the medical application context, as incorrect diagnostic and prognostic decisions can have significant negative consequences for the patients concerned. We intend to research XAI in the context of medical decision support systems. This includes developing an understanding of the application of XAI to different data types and diseases, and whether there has been experimental evaluation of the impact of XAI in AI-based decision support.

To develop a better understanding of XAI in the context of medical decision support systems, a systematic literature review (SLR) is carried out in this Bachelor’s thesis. To collect additional data and enhance the knowledge about XAI use cases in medical practice, the student conducts a series of expert interviews for requirements elicitation. 

Notifications from instant messaging applications can interrupt employees' productive time. While there are different ways to influence the notification behavior of instant messengers, such as turning off the application or muting notifications for certain periods of time, these measures require self-discipline and/or often result in missing notifications when not in flow. We have developed an adaptive instant messaging blocker that aims to solve this problem by recognizing the user's flow state at predefined intervals, based on their physiological data and using machine learning methods. As soon as a flow state is recognized, the “do not disturb” status is automatically activated for the duration of the flow state.

We conducted interviews with knowledge workers to evaluate the developed system. Therefore, a qualitative analysis (with MAXQDA) is to be carried out in this Master's thesis in order to evaluate the system on the basis of the interviews conducted.