Primary Investigator (PI) Name
Pamila Dembla & Gunjan Batra
Department
CCOB – Information Systems and Security
Abstract
AI-based mental health chatbots are increasingly being introduced to expand access to psychological support, especially where stigma, cost, scheduling constraints, or limited availability make traditional therapy services difficult to use. Many people are beginning to turn to these tools for immediate, low-threshold assistance when experiencing stress or emotional distress. However, we still know relatively little about the situations in which individuals feel comfortable relying on AI for support, or when they would prefer human help instead, creating uncertainty about how such systems should be designed and deployed responsibly. This study examines how users evaluate AI-based mental health support compared with human counseling across a range of common, non-crisis situations. Using a vignette-driven Design Science approach, six realistic scenarios were developed to reflect everyday experiences such as academic pressure, fear of judgment, late-night distress when services are unavailable, privacy concerns, trade-offs between AI and human support, and doubts about AI’s ability to understand emotions. An initial classroom study (N = 23) was conducted with undergraduate students at a large business school in Southeastern US. Students responded to the scenarios through structured surveys and guided classroom discussion, indicating how likely they would be to use AI, which type of support they would prefer, and what factors influenced their decisions. Early findings suggest three recurring themes: students are more open to AI when distress is mild, they strongly value emotional authenticity and human empathy, and they weigh privacy concerns against the perceived safety of AI interaction. Additional data collection is underway to expand the sample and enhance the robustness of the findings. Overall, this work introduces a practical vignette instrument and a design-oriented framework for understanding when AI mental health tools are acceptable to users. The findings provide guidance for designing and evaluating systems that align with user needs, expectations, and ethical boundaries.
Presented at
2026 - The Thirtieth Annual Symposium of Student Scholars
AI for Mental Health: A Design Science Approach
2026 - The Thirtieth Annual Symposium of Student Scholars
AI-based mental health chatbots are increasingly being introduced to expand access to psychological support, especially where stigma, cost, scheduling constraints, or limited availability make traditional therapy services difficult to use. Many people are beginning to turn to these tools for immediate, low-threshold assistance when experiencing stress or emotional distress. However, we still know relatively little about the situations in which individuals feel comfortable relying on AI for support, or when they would prefer human help instead, creating uncertainty about how such systems should be designed and deployed responsibly. This study examines how users evaluate AI-based mental health support compared with human counseling across a range of common, non-crisis situations. Using a vignette-driven Design Science approach, six realistic scenarios were developed to reflect everyday experiences such as academic pressure, fear of judgment, late-night distress when services are unavailable, privacy concerns, trade-offs between AI and human support, and doubts about AI’s ability to understand emotions. An initial classroom study (N = 23) was conducted with undergraduate students at a large business school in Southeastern US. Students responded to the scenarios through structured surveys and guided classroom discussion, indicating how likely they would be to use AI, which type of support they would prefer, and what factors influenced their decisions. Early findings suggest three recurring themes: students are more open to AI when distress is mild, they strongly value emotional authenticity and human empathy, and they weigh privacy concerns against the perceived safety of AI interaction. Additional data collection is underway to expand the sample and enhance the robustness of the findings. Overall, this work introduces a practical vignette instrument and a design-oriented framework for understanding when AI mental health tools are acceptable to users. The findings provide guidance for designing and evaluating systems that align with user needs, expectations, and ethical boundaries.