Mental Health Assessment for the Chatbots

Abstract: Previous researches on dialogue system assessment usually focus on the quality evaluation (e.g. fluency, relevance, etc) of responses generated by the chatbots, which are local and technical metrics. For a chatbot which responds to millions of online users including minors, we argue that it should have a healthy mental tendency in order to avoid the negative psychological impact on them. In this paper, we establish several mental health assessment dimensions for chatbots (depression, anxiety, alcohol addiction, empathy) and introduce the questionnaire-based mental health assessment methods. We conduct assessments on some well-known open-domain chatbots and find that there are severe mental health issues for all these chatbots. We consider that it is due to the neglect of the mental health risks during the dataset building and the model training procedures. We expect to attract researchers' attention to the serious mental health problems of chatbots and improve the chatbots' ability in positive emotional interaction.

• The paper introduces a novel method to evaluate chatbot mental health, adapting human psychological assessments for dimensions like depression and anxiety.
• Experiments showed open-domain chatbots have significant mental health issues, including moderate/severe depression/anxiety and low empathy.
• Findings suggest current datasets and training neglect mental health risks, emphasizing the need for developers to prioritize positive emotional interactions.

The paper introduces a novel approach to evaluate the mental health of chatbots, addressing a critical gap in current dialogue system assessments that primarily focus on technical metrics.

Here's a summary of the key aspects:

• The paper establishes mental health assessment dimensions for chatbots, including depression, anxiety, alcohol addiction, and empathy, using questionnaire-based methods adapted from human psychological assessments like PHQ-9, GAD-7, CAGE, and TEQ.
• Experiments on open-domain chatbots like Blender, DialoGPT, Plato, and DialoFlow revealed significant mental health issues, with moderate to severe scores in depression and anxiety, and below-average empathy scores, suggesting a potential negative impact on users.
• The authors suggest that the neglect of mental health risks during dataset building and model training contributes to these issues, advocating for increased attention to mental health in chatbot development to foster positive emotional interactions.