Construction And Validation of Conversational AI Dependence During Stress Scale (CAIDSS)

Abstract

Past few years have seen rise in use of conversational artificial intelligence (CAI) driven systems such as ChatGPT by OpenAI, Grok, Replika. Based on the statistics provided by their parent companies, CAIs have been used for personal matters especially for guidance and decision making. Given this surge in CAI’s popularity, mental health support systems have also seen a shift towards CAI driven applications and chatbots. On one hand CAIs ease accessibility and affordability of support needed but on the other hand they face critical limitations such as lack of emotional intelligence and possible bias against gender, race, caste or ethnicity. Thus, this study was undertaken to develop a short screening tool that can aid people identify the extent to which an individual is uses CAIs to seek help under distress. The 4-item CAIDS scale showed good internal consistency (α = 0.862, ω = 0.865) and has a unidimensional factor structure based on exploratory factor analysis. Additionally, the correlation between CAIDSS and an established measure provided strong evidence for the former’s convergent validity (r=0.82). CAIDS scale is a reliable and valid tool that can be used in clinical and research contexts to better understand causes and outcomes of using CAI agents for guidance under distress.