Building and Beta-Testing Be Well Buddy Chatbot, a Secure, Credible and Trustworthy AI Chatbot That Will Not Misinform, Hallucinate or Stigmatize Substance Use Disorder: Development and Usability Study

Abstract

Background: Artificially intelligent (AI) chatbots that deploy natural language processing and machine learning are becoming more common in health care to facilitate patient education and outreach; however, generative chatbots such as ChatGPT face challenges, as they can misinform and hallucinate. Health care systems are increasingly interested in using these tools for patient education, access to care, and self-management, but need reassurances that AI systems can be secure and credible. Objective: This study aimed to build a secure system that people can use to send SMS with questions about substance use, and which can be used to screen for substance use disorder (SUD). The system will rely on data transfer via third party vendors and will thus require reliable and trustworthy encryption of protected health information . Methods: We describe the process and specifications for building an AI chatbot that users can access to gain information on and screen for SUD from Be Well Texas, a clinical provider affiliated with the University of Texas Health Sciences Center at San Antonio. Results: The AI chatbot system uses natural language processing and machine learning to classify expert-curated content related to SUD. It illustrates how we can comply with best practices in HIPPA (Health Insurance Portability and Accountability Act) compliance in data encryption for data transfer and data at rest, while still offering a state-of-the-art system that uses dynamic, user-driven conversation to dialogue about SUD, screen for SUD and access SUD treatment services. Conclusions: Recent calls for attention to user-friendly design concerning user rights that honor digital rights and regulations for digital substance use offerings suggest that this study is timely and appropriate while still advancing the field of AI.

Ähnliche Arbeiten