Open Access

While persons across the globe are increasingly leading longer lives, they are not necessarily leading healthier lives, particularly in old age. To support this, addressing the health and care needs of older adults and promoting healthy aging is vital - particularly in the wake of aging populations and the shortage of health care workers in many countries worldwide. Promoting both health literacy and self-care can support this, by empowering older adults to take a proactive stance regarding their health. Assistive technologies can provide support through health monitoring and facilitating self-management, and thus leveraging evidence-based strategies to directly support the well-being of older adults. For a comprehensive and holistic assessment of health among older adults, a multidimensional approach which considers the physical, mental, social, and environmental health domains is crucial. Here, subjective data can provide insight into a person's well-being from their own point of view, and contribute context regarding their health status, while objective health data can be measured, observed, or self-reported and are based on facts or numbers. In this dissertation, we present a health monitoring system combining subjective and objective data related to the health and well-being of older adults. Such a system can encompass various health information technologies and assistive technologies to achieve comprehensive self-health monitoring: A social robot or virtual agent can autonomously conduct health assessments and thus collect subjective and objective health data, while various wearable and non-wearable sensors can acquire objective health data. Using the Design Science Research Methodology approach, we iteratively developed and evaluated this system, in order to both provide a technology to support health monitoring among older adults, whilst also providing knowledge on effective deployments of such systems in real-world settings. Through four user studies with three cohorts of older adults from long-term care, assisted living, as well as community-dwelling older adults, we gained insight into factors influencing the deployment and adoption of health monitoring systems. To this end, we developed WiSH-TAM (in the wild social agent-integrated health-technology acceptance model), which can be used for integrating such systems into real-world settings. We also determined subjective and objective data sources to gain a comprehensive and holistic overview of the health of older adults. With these data sources we achieved high accuracies in classifying the health status of older adults, with a real-world dataset. We also presented text classification methods suitable to derive contextual health information from human-agent interactions. Combining these data with objective data, we showed that incorporating conversational data contributes to increasing the accuracy of health classifications, and that the conversational data can provide opportunities to support the health literacy and self-care of older adults. Building on the different settings and cohorts involved in our four studies, we gained insights into the potential of our system for these settings and users, and thus can conclude that the system is suitable for deployment in all these settings, provided the older adults are supported in this endeavor. Finally, the system also provides opportunities to support integrated care approaches, like the Integrated Care for Older People (ICOPE) approach proposed by the World Health Organization. With these three outcomes, we have made contributions to the social scientific field of technology acceptance with the WiSH-TAM, to the intersection of machine learning and medicine with our health monitoring system and its health prediction capabilities, and to public health research through the proposed integration of the system into the ICOPE approach.

Das Promotionsprojekt ist interdisziplinär an der Schnittstelle von Mensch-Maschine-Interaktion und Coaching-Wirksamkeitsforschung angesiedelt. Im Rahmen des Projekts wird der Einfluss beziehungsbildender Faktoren auf die Beziehungsgestaltung im Chatbot-Coaching untersucht. Dabei wird die Domäne des Lern- und Studierenden-Coachings ausgewählt. Der Fokus liegt auf der empirischen Untersuchung von Rapport, Arbeitsbeziehung und sozialer Präsenz zwischen Coach und Coachee, also der Frage, ob und wie sich eine Beziehung zwischen Coaching-Chatbot und Coachee entwickelt. Dazu werden potentielle Wirkfaktoren (Self-disclosure, Empathie und User-Persönlichkeit) ausgewählt und aktuelle technologische Entwicklungen (Einsatz von generativer KI und Large-Language-Modellen) als mögliche Wirkfaktoren berücksichtigt. In einem ersten Schritt wird literaturbasiert herausgearbeitet, welche Faktoren zur Gestaltung einer Beziehung aus der Coaching-Forschung und der Forschung zur Mensch-Maschine-Interaktion in einen Coaching-Chatbot übertragbar sind. Im zweiten Schritt werden diese Faktoren zu Handlungsanweisungen und Interventionsstrategien operationalisiert und in experimentellen Studien anhand von Hypothesen untersucht. Daraus lassen sich drittens Empfehlungen für Forschung und Praxis ableiten. Als übergreifender Forschungsansatz wird eine Verbindung von Design-based Research und experimenteller Forschung gewählt. Dies beinhaltet einen Mixed-Methods-Ansatz, der qualitative, quantitative und psychophysiologische Messmethoden berücksichtigt sowie Messmethoden und -instrumente kritisch beleuchtet. Die Ergebnisse zeigen, dass die Faktoren (1) Self-disclosure des Chatbots, (2) empathische Rückmeldungen des Chatbots und (3) Paraphrasieren mit dem Large-Language-Modell ChatGPT-4 einen signifikanten Einfluss auf die Beziehungsbildung in der Domäne des Studierenden-Coachings mit Chatbots haben. Dabei zeigt sich der beziehungsbildende Effekt auf der Ebene der sozialen Präsenz, nicht auf der Ebene der Arbeitsbeziehung. Moderator- und Störvariablen (User-Persönlichkeit und emotionales Befinden der User) haben keinen Einfluss auf die Beziehungsbildung im Chatbot-Coaching. Konkrete Empfehlungen zum Conversation-Design und Prompting von Chatbots können auf struktureller und sprachlicher Ebene aus den Studien abgeleitet werden: Die Variation von klick- und schreibbasierter Interaktion, Text und Bild sowie Information und Selbstreflexion auf struktureller Ebene ist geeignet, um das Chatbot-Coaching abwechslungsreicher zu gestalten. Auf sprachlicher Ebene sollten Self-disclosure- und Information-disclosure-Aussagen variiert werden, ebenso wie empathische Rückmeldungen eines Chatbots eingebunden werden sollten. Darüber hinaus lassen sich Empfehlungen für die verantwortungsvolle Gestaltung von KI-basierten Coaching-Chatbots ableiten. Diese beinhalten die Entwicklung von hybriden Chatbots unter Berücksichtigung verschiedener Design-Ansätze, die Auswahl geeigneter Large-Language-Modelle für den jeweiligen Use Case, sorgfältiges Prompting als Teil des Conversation-Designs, eine intensivere Implementierung von generativer KI in Coaching-Chatbots sowie den Einsatz von sensorbasiertem Coaching. Das Promotionsprojekt leistet damit einen substanziellen Beitrag zum (1) Conversation-Design und Prompting von Coaching-Chatbots, zum (2) verantwortungsvollen Einbezug generativer KI in das Chatbot-Coaching sowie zu (3) Messmethoden und -instrumenten zur Untersuchung der Beziehungsbildung im Chatbot-Coaching. Die Arbeit schließt mit einem Ausblick auf die Zukunft des KI-basierten Coachings und damit einhergehenden Thesen zur Gestaltung von KI-basierten Chatbots sowie zukünftigen Forschungsfeldern. Die Arbeit trägt damit zu Forschung und Entwicklung in den Bereichen gesprächsbasier-te KI, Rapport-Forschung in der Mensch-Maschine-Interaktion sowie angewandte Psychologie, Coaching-Wirksamkeitsforschung und Hochschuldidaktik bei.

AI video generation models such as Open AI’s Sora make it possible to create ultra-realistic animations from scratch and seamlessly merge disparate visual content into new synthetic media. For its developers, however, Sora seems more than just a tool; they rather see it as a “world simulator.” But if Sora, trained on online video content scraped from YouTube or generated by game engines such as Unreal, is supposed to simulate a world, what world is it? As the short essay argues, it is a world of constant modulation and endless flows of patterns—a flat world, the world of platform capitalism.

This study explores the potential of generative AI, specifically Long Short-Term Memory (LSTM) networks, to advance collaborative choreographic composition within the framework of the Body Logic (BL) Method—a choreographic approach grounded in cognitive science designed to challenge inherited habits and practices in contemporary dance. Through five cognitive tasks that emphasize different movement types and their qualities, we investigate how LSTM networks recognize established movement patterns and innovate by combining them in novel ways, mirroring the processes of human creativity. Furthermore, we examine how LSTM-generated sequences, derived from learned data, convey expressive qualities through a variety of movements. The AI-generated movements closely follow the original movement trajectory but exhibit minor deviations attributable to the LSTM model’s inherent prediction uncertainty. These variations illustrate the model’s capability to introduce fresh elements while maintaining learned patterns, akin to human creativity. This research contributes novel perspectives on how technology can enrich artistic expression and challenge habitual decision-making in dance.

Text-to-image generative models offer an innovative method for creating visual content, exploiting the limitless potential of text-based inputs. However, the reliance on text prompts can lead to a labor-intensive process of experimentation for users aiming to achieve high-quality results. This has led to the development of a specialized prompt language with specific, descriptive keywords that users can exploit so as to achieve the best possible visual outcomes. Insight into prompting strategies can be obtained by analyzing the media generated and shared on text-to-image online platforms. Utilizing natural language processing (NLP) and visualization techniques, a detailed analysis of the prompts that led to the creation of images with exceptional popularity (by like count) was performed. The present study is focused on identifying the predominant topics and language patterns that contribute to the creation of images that receive high community ratings. Our analysis reveals a strong focus on surface aesthetics in prompts, which often emphasize conventional beauty and popular visual themes and a gravitation towards erotic and pornographic content. Positive prompts typically involve descriptions of female bodily features, blending elements of fantasy and realism. Negative prompts consistently counteract what seems to be perceived as visual imperfections, often describing body horror, marked by distorted human features as well as technical imperfections related to digital imagery in particular.