Chatbots in Family Medicine: A Systematic Review and Perspectives

Simon-Konstantin Thiem, Arezoo Bozorgmehr, Christiane Stieber, Birgitta Weltermann

Keywords: Chatbots, Review, Artificial Intelligence, Machine Learning, Family Medicine

Background:
The technology of chatbots has advanced immensely in imitating natural language and the ability to form coherent understandable sentences. In medicine, the most commonly used versions are conversational agents, which infer their knowledge mostly from databases, while the fast-paced options of machine learning (ML) and artificial intelligence (AI) are often not fully explored.

Research questions:
In this project, we will review recent chatbot research and build a vision of an optimal chatbot for the use by primary care patients.

Method:
Building on published work and the PRISMA review approach, we will first review the most recent research on chatbots in the medical field since 2019. We will provide a broad overview of the AI and ML methods used, as well as the medical context for the respective chatbot applications and their evaluation. We will derive which approaches might be of use for family medicine.
Based on the review we build a vision of a System-AI chatbot for family medicine patients that incorporates the natural language skills of modern transformer based architectures, the simulated explainability of rule-based approaches as well as the precision of various neural network approaches.
Lastly, we analyze our envisioned model and compare it with state of the art methods in ML and AI. We will consider if certain AI methods can be combined to form more elaborated models. We will highlight areas with unmet patient needs and outline chatbot models with possible future applications in family medicine.

Results:
The review was started. We will provide first results at the conference.

Conclusions:
We will provide a review and perspectives for chatbots in family medicine.

Points for discussion:
Can camera analysis and body language captures improve chatbots?

How much algorithmic explainability is required?

Can OpenAI’s GPT-3 be helpful to address needs of family medicine patients?