User Study on the Trustworthiness, Usability and Explainability of Intent-based and Large Language Model-based Career Planning Conversational Agents
Thorsten Zylowski, Nathalia Sautchuk Patrício, Wladimir Hettmann, Katharina Anderer, Karl Fischer, Matthias Wölfel, Peter A. Henning · 2024
Choosing a career and educational path is a challenging decision for young people. Career planning conversational agents (CAs) can assist by identifying suitable occupations and educational paths. Trustworthiness is an important dimension for the acceptance of a career planning CA and is influenced by several factors. We conducted a user study with n=114 participants across three schools in Germany to explore the trustworthiness of different career planning CAs. We examined the correlation between trustworthiness and perceived competence, autonomy, and social relatedness from self-determination theory (SDT), as well as the explainability of interactions and several usability dimensions of the assistants. These dimensions included the ability to guide the conversation, onboarding quality, error tolerance, and information relevance. We tested three different variants of the career planning assistant: a form-based assistant, an intent-based CA, and a large language model (LLM)-based CA. The results showed that the LLM-based CA was on average significantly more trustworthy and was perceived as more explainable than the intent-based CA. Key trust factors included conversation flexibility, chatbot credibility, intent recognition, and maintenance of a secure conversation. Additionally, perceived autonomy was crucial for trust across all types of assistants and perceived relatedness for the two CAs. Our findings highlight key areas essential for developing trustworthy CAs.