Towards Designing Empathetic and Trustworthy AI Chatbots: an Exploratory Study

EasyChair Preprint 11648

Abstract

Intelligent agents like chatbots have recently attracted enormous attention due to their advanced capability to augment humans in information gathering and task execution. While they are designed to be more understanding, the existing literature lacks deep knowledge of how chatbots should track and respond to users' emotions in real-time with empathy. In this exploratory study, we proposed an innovative emotion-detecting system that combines CNN-based facial expression recognition algorithms with text-based sentiment analysis to improve real-time interactions between users and an AI-powered chatbot by recognizing users' emotional expressions and delivering empathetic responses appropriately. We present preliminary results of a human-subject study with distinct versions of chatbots. We confirm that adding facial expression detection improves the predictability of the models of user-perceived trust and empathy.

Keyphrases: Chatbot, Human-AI Interaction, Trust, emotional states, empathy, facial expression

@booklet{EasyChair:11648,
  author    = {Melik Ozolcer and Mohammad Rahul Islam and Abdullah Mohammed and Tongze Zhang and Sang Won Bae and Ting Liao},
  title     = {Towards Designing Empathetic and Trustworthy AI Chatbots: an Exploratory Study},
  howpublished = {EasyChair Preprint 11648},
  year      = {EasyChair, 2024}}