AI for Good: Training 4000+ mental health volunteers at MHI | Mark Ungless, Director of Data Science, AI and Research

In the wider world this week:

• Are Microsoft looking to split from OpenAI? Reports suggest they’re developing their own AI models

• Mistral releases a document converter API for converting multi-language, complex documents

A minor caveat for this week’s story: Temrel worked on the SCS platform project in 2024 and 2025 helping extend the AI feature set and infrastructure robustness, and as an example of using the power of artificial intelligence for a genuine social good, it’s a project still close to our hearts.

Shout is a free, confidential, and 24/7 text messaging application service for people in the UK needing mental health support, managed by Mental Health Innovations (MHI). Shout messaging is staffed by volunteers who train on the Shout Conversation Simulator (SCS).

In a recent interview with Mark Ungless, the Director of Data Science, AI and Research at MHI, I learned about the remarkable evolution of the AI chatbot used by MHI for volunteer training. What began as a simple keyword-search bot in 2018 has transformed into a sophisticated conversation simulator powered by Llama 3, thanks to funding from Jensen Huang's family foundation.

"It started as a very basic keyword search," Mark explained, "but we knew we needed something more to really prepare our volunteers."

This significant upgrade, implemented in 2021, has greatly improved volunteer training and engagement. The new model was trained on MHI's extensive dataset of 100 million text messages, allowing it to simulate highly realistic conversations.

Over 4,000 individuals have used the simulator, and feedback has been overwhelmingly positive, with an 85% positive response rate regarding skill improvement. Additionally, volunteer activity and retention have increased by approximately 10% since the simulator's implementation. "We've seen a real difference in how prepared our volunteers feel," Mark noted, "and that translates into better support for our users."

The simulator provides a safe and effective space for volunteers to practice their skills and gain confidence before interacting with real users. As Mark put it, "It's a risk-free environment where they can make mistakes and learn from them."

Looking ahead, MHI is focused on further enhancing the platform. An exciting area of development is improving the model's ability to manage the full arc of a conversation, including greetings and closings, to make the simulations even more comprehensive.

"We're still working on the model's ability to manage the full arc of a conversation," Mark said, "including those greetings and closings." The goal is to provide an ever more immersive, realistic and effective training experience.

MHI is considering ways to integrate elements of its simulator directly into the Shout platform, providing real-time suggestions for volunteers to help them deliver effective support to users but ‘it’s early days’, says Mark. Ethical considerations, such as transparency regarding AI assistance, are a prime consideration.

The potential for AI to serve as a "stepping stone" to human interaction is being explored, with the goal of enhancing, not replacing, human support. "We see AI as a tool to augment our volunteers, not replace them," Mark emphasized.

The AI chatbot has proven to be a valuable tool for MHI, significantly enhancing volunteer training and ultimately improving the support provided to those in need. The future looks incredibly promising, with ongoing improvements and integrations set to further revolutionize the volunteer experience.

Temrel is an AI Automation and Workflow partner. We help companies implement AI to improve outcomes, lower costs, and increase efficiency. Interested in starting a conversation?
- book a call
- send us an email