AI Empathy Agent for Mental Health

Stage 1 – Developing an AI Empathy Agent

Current e-therapies lack empathy; they are often experienced as cold and not personally relevant.  This project tested the feasibility of addressing these issues by developing an Empathy Agent (EA) powered by the Peer Support Community and capable of delivering empathic support.   The feasibility study tested how service users responded to and interacted with a mock Empathy Agent.  The first phase involved generating a bank of supportive responses from the service user community via an online survey.  The second phase used workshops and focus groups to simulate how the Empathy Agent functions as an intermediary between the service user community and individual service users using the crowd-sourced response bank.  The third phase focused on the analysis of feedback to gain an understanding of both the feasibility of crowdsourcing the Empathy Agent’s responses, and the Service User experience of the process.  User feedback was vital in evaluating the viability of delivering empathic support through the Artificial Intelligence.  The results were used to inform a report shared with key networks and stakeholders in the areas of peer support and mental health technology.  It is hoped that this will create new possibilities for collaboration for two major areas of development in mental health.

Media / Posters

NewMind Network. (2019). AI Empathy Agent. Link to Page.

The NewMind Network. (2020). The NewMind Network Technology in the Service of Mental Health. Link to Brochure.

Sirois F, Christensen H, Tucker I, Millings A, Easton K, Radin P, Bennion MR, Ganga R. Developing and Testing an AI Empathy Agent. Poster presented at: NewMind Plenary January 2019: Feasibility Study Poster Presentation; 30/01/2019; Manchester Conference Centre. Link to Poster.

Stage 2 – Building and Testing a Demonstrator System for the AI Empathy Agent

This study proposed an always-on, easy-to-use Empathy Agent (EA) to deliver peer-support in a digital arena.  By crowdsourcing empathic responses generated by real users, the EA would grow smarter and more empathic with each use and thus improve over time. Currently, there are no online, automatic systems available like this; the increasingly ubiquitous chatbot technology are mostly based on prewritten scripts.  The primary objective of the project was to extend the NewMind Funded Stage 1 Feasibility work by building and testing a demonstrator system of the Empathy Agent, using several service-user centred design cycles. First, service users rated the existing bank of responses from Stage 1 in an online survey. The study then explored recent advances in natural language processing and machine learning to construct a demonstrator system in order to provide the most appropriate responses. The system was then be trialled with a small group of servicer users to evaluate its ability to deliver empathetic responses in an appropriate manner. The system provided instant responses crowdsourced from peers to deliver the voice of the peer support community in an empathetic and appropriate way in times of need.  Development of the demonstrator was carried out by myself and FeiFei Xiong.  The agent used modern web technologies combined with a Python based machine learning AI backend.

Funding

This research was funded and supported by the New Minds Network.

Media / Posters

NewMind Network. (2019). Building and Testing a Demonstrator System for the AI Empathy Agent. Link to Page.

Xiong F, Sirois F, Easton K, Millings A, Bennion M, Radin P, Tucker I, Ganga R, Christensen H. Natural Language Processing Applied to Empathy Agent for People with Mental Health Problem. Poster presented at: UK Speech Conference 2019; 24/06/2019; University of Birmingham.

Project team:

Principal Investigator: Fuschia M. Sirois
University of Sheffield, UK

Co-Investigators:
Ian Tucker
University of East London, UK

Abigail Millings
University of Sheffield, UK

Paul Radin
Nottingham Healthcare NHS Foundation Trust, UK

Matthew Bennion
University of Sheffield, UK

Rafaela Ganga
Liverpool John Moores University, UK

Katherine Easton
University of Sheffield, UK

Heidi Christensen
University of Sheffield, UK

FeiFei Xiong
University of Sheffield, UK