Encoding empathy in Conversational AI

Encoding Empathy: Solving the Healthcare Workforce Crisis Through the Safe Deployment of an AI-Driven, Voice-Based, Clinical Conversational Assistant for Long Term Monitoring

Our team is working with partners in Ufonia, University of York, and the UCL Hospitals NHS Trust to improve automated telephone clinical consultations. The project is set to run 2024-2026, supported with funding from the UK's innovation agency, Innovate UK. Project Lead is Dr Ernest Lim, Director of Science at Ufonia.

We collaborate with the design team of Ufonia to implement performance improvements for their clinical assistant Dora.

The project team

Exploring the possibilities of Large Language Models to drive effective communication in clinical settings.

Encoding Empathy: Solving the Healthcare Workforce Crisis Through the Safe Deployment of an AI-Driven, Voice-Based, Clinical Conversational Assistant for Long Term Monitoring

Co-funded by the UK's innovation agency, Innovate UK. Biomedical Catalyst 2023 Round 2: Industry-led R&D

Analytic approach

Our analytic contribution to this Innovate UK-funded project centres on identifying effective practices used by human clinicians in their consultations with patients. This is then used to train a Large Language Model (LLM) to incorporate these into its automated clinical conversations.

Data for the first phase includes recordings of telephone calls between clinicians and patients, capturing the patients' treatment journeys. This longitudinal approach allows us to observe how rapport develops over time.

We analyze these telephone call recordings using Conversation Analysis (CA), focusing on the micro-details of communication such as word selection, intonation, and pauses. CA research has shown how these subtle shifts can significantly impact the outcome of interactions.

Through this detailed turn-by-turn analysis, we gain valuable insights into effective communication strategies for clinical consultations, which can be used to train a Large Language Model (LLM) to produce more intuitive patterns for patients to interact with.

Implementation