Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

ChatCLIDS: Simulating Persuasive AI Dialogues to Promote Closed-Loop Insulin Adoption in Type 1 Diabetes Care

Created by
  • Haebom

Author

Zonghai Yao, Talha Chafekar, Junda Wang, Shuo Han, Feiyun Ouyang, Junhui Qian, Lingxi Li, Hong Yu

Outline

This paper highlights that the low adoption of closed-loop insulin delivery systems (CLIDS) in patients with type 1 diabetes is due to behavioral, psychosocial, and social barriers, rather than technical issues. We propose ChatCLIDS, a new benchmark to address this issue. ChatCLIDS simulates interactions with a nurse agent using a variety of persuasive strategies, utilizing expert-validated virtual patients. This allows for multidimensional assessments, considering prolonged consultations and social pressure situations. Our findings demonstrate that large-scale language models (LLMs) adapt their strategies over time but struggle to overcome resistance, particularly under realistic social pressures. This highlights the current limitations of LLMs in behavioral change and provides a high-fidelity, scalable testbed for the development of reliable persuasive AI in healthcare and beyond.

Takeaways, Limitations

•
Takeaways:
◦
ChatCLIDS provides an in-depth analysis of the causes of the decline in adoption of closed-loop insulin delivery systems and a new evaluation framework to address them.
◦
We present a high-fidelity testbed that can evaluate the effectiveness and limitations of LLM-based persuasive conversations in real-world situations.
◦
It clearly highlights the limitations of current LLMs' ability to induce behavioral change and suggests directions for developing reliable, persuasive AI.
◦
It provides a general benchmark that can contribute to the development of persuasive AI not only in the medical field but also in other fields.
•
Limitations:
◦
ChatCLIDS's virtual patients and scenarios may not perfectly reflect real-world situations.
◦
Although it clearly shows the current performance limitations of LLM, it does not suggest any specific solutions to overcome them.
◦
Additional assessments that take into account diverse social and cultural backgrounds are needed.
◦
Additional consideration is needed regarding the ethical aspects of the LLM.
👍
Made with Slashpage