Bridging the Uncanny Valley: Improving AI Chatbots for Effective Leadership Mentoring

Abstract

Backg round: A significant challenge in developing AI assistants, such as chatbots as mentors or coaches for leaders, is the Uncanny Valley effect. This phenomenon describes the discomfort leaders feel when AI-chatbots appear almost human-like but not quite perfect, causing a sense of eeriness. To avoid this, a chatbot-design should focus on creating distinctive features rather than overly human-like ones and balance human-likeness with chat-elements to maintain user comfort. Emphasizing functional and straightforward designs that prioritize leaders’ interaction with the chatbot and usability over appearance. Method: This study employed a qualitative research method, utilizing a qualitative content analysis to explore the extent of the Uncanny Valley effect in the interaction between AI chatbots and leaders. Data was collected through an online survey involving 82 leaders, of which 62 completed the entire questionnaire. Participants were shown video excerpts from an existing AI-based chatbot named “KI.m” and participated in semi-structured interviews to provide insights into their perceptions and emotional responses. Results: Key findings indicate that leader’s expectation, voice quality, emotional expression, and social presence significantly influence user discomfort. Participants preferred more natural and human-like voices, suggesting that developers refine pronunciation and intonation. Emotional responses were often perceived as forced and insincere, highlighting the need for contextually relevant and genuine emotional expressions. Social interactions were seen as too mechanical, suggesting a balance between professional, supportive communication and subtle empathy. Personalization and accessibility through customizable settings and multilingual support were emphasized. Users preferred concise, clear answers. Conclusion: This study identified key themes influencing user discomfort with AI chatbots, including voice quality, emotional expression, social presence, and the need for aligning AI’s cognitive abilities with user expectations. Enhancing personalization and accessibility in AI tools was also emphasized as crucial for improving user satisfaction and practical utility. The findings highlight the potential of AI mentoring tools to support leadership development, provided that key discomforts related to the Uncanny Valley effect are addressed. Future research should focus on refining AI attributes and exploring long-term impacts, cultural differences, and ethical considerations to enhance the design and implementation of AI mentoring systems.