Causal direction of heavy chatbot use and psychosocial outcomes

Determine whether heavier chatbot use causally drives adverse psychosocial outcomes or instead reflects pre-existing psychosocial outcomes in longitudinal human–AI interaction data.

Background

The paper discusses longitudinal human–AI interaction as a two-way process in which users adapt to systems and systems adapt to users through personalization, memory, and feedback-driven training. It cites a four-week randomized study in which chatbot assignment did not significantly affect psychosocial outcomes, whereas participants who chose to use the chatbot more reported worse outcomes.

Because the heavier-use association may arise either from chatbot use affecting users or from users with worse outcomes choosing greater use, the causal direction remains unresolved. The authors present this as a motivating example for causal modeling of user and assistant behavior over time.