Sprecher
Beschreibung
Chances are you've typed something personal into ChatGPT, Replika, or a similar system: a bad day, a late-night thought, a question you weren't ready to ask a person. You're not alone. Conversational AI is increasingly becoming a space for emotional disclosure, comfort, and even companionship. But what actually happens to us when these interactions aren't a one-off, but repeated over weeks or months? The honest answer is: we don't really know. Existing research is fragmented, dominated by single short sessions, and – crucially – often uses AI systems whose "empathic" behaviour is never precisely specified, making findings hard to compare or replicate.
This PhD project tackles that gap through a cumulative research design (two controlled experiments and one longitudinal study) built around one guiding question: How do affective dynamics emerge and evolve in sustained emotional human–AI interaction, and which mechanisms and individual factors shape these processes? Study 1 tests whether more emotionally expressive AI communication causally shapes how we feel. Study 2 asks why, focusing on perceived empathy as the underlying mechanism. Study 3 follows people over several weeks to see how these bonds actually develop over time.
A central contribution is technical: rather than relying on vague or inconsistent "empathic" chatbots, this project builds an AI system whose empathy can be dialed up or down like a control knob (independent of what it actually says!) using a technique called activation steering, grounded in an established empathy framework. This turns "AI empathy" from a fuzzy concept into something measurable, controllable, and reproducible – a tool other researchers can build on.
The result: causal, mechanistic, and longitudinal insight into what it really means to form an ongoing emotional bond with AI, with direct relevance to digital companionship, mental health support, and AI for populations of all ages.
