Large-language model-powered conversational agents, such as AI chatbots and ChatGPT, can emulate complex human-like conversations. An intriguing application area for these chatbots is personal companionship. Through prompt engineering and specially trained LLMs (e.g., InstructGPT), we can "design" chatbot personalities with specific conversational styles, capable of discussing particular topics. For instance, a chatbot could be designed to prompt reflection, enhance skills, or encourage changes in attitudes and behaviors.
In this course, we will design personal companions and explore their impact through an autoethnographic approach. We will investigate the possibilities and challenges of designing specific chatbots, considering questions such as:
How much control can we exert over the chatbot's responses?
How much control is desirable to maintain a desired interaction?
Due to the nature of prompt engineering and training LLM chatbots, it is possible to quickly iterate and experiment with different chatbots, e.g., for different purposes or with completely different personalities. We will integrate these chatbots into our daily lives and document our experiences to critically assess and reflect their impact.
- Dozent/in: Marc Hassenzahl
- Dozent/in: Robin Neuhaus