Design and Evaluation of a Culturally Adapted Multimodal Virtual Agent for PTSD Screening

Why Read This from an HCI Perspective

This article shows how AI can be experienced differently depending on how it asks sensitive questions and how it responds—rather than treating AI as a mere automation tool. Particularly in situations like PTSD screening, where trust and safety are critical, it is meaningful for HCI and UX practitioners to examine how an avatar’s tone, speaking pace, and guidance style affect users’ openness and discomfort. It’s also worth reading for its practical application of cultural adaptation and multimodal design.

CIT's Commentary

What’s interesting is that the success of this system depends more on how the interaction is managed than on model scores. In PTSD screening, it matters less whether the system provides the “correct answer” and more whether it helps users know when they can stop, what path it takes to connect them to a person in a crisis, and what the system is doing right now. In other words, the key is the interface that enables AI to intervene safely rather than simply looking intelligent. Also, this study chose an observation-based evaluation rather than direct use. That approach is appropriate for early safety validation, but it can easily miss behavioral data such as hesitation or dropout at the moment of actual intervention. In the next step, research that more closely examines direct-use scenarios and failure modes will likely be needed. In the Korean context as well, medical and counseling AI may not be something that can be directly transplanted into products from Naver, Kakao, or startups; it raises the question of designing together for Korean dialects and users’ expectations of “human-like AI.”

Questions to Consider While Reading

• Q.When evaluation is based only on observation without direct interaction, how can we address anxiety, resistance, and dropout that may appear during real use?
• Q.For functions such as crisis detection or safety confirmation, how much minimal transparency is needed to make users trust the system more?
• Q.When adapting a design for Arabic culture to Korean language and domestic medical contexts, which interaction element should be re-validated first?