1 of 27

Global Stakeholder Meeting

​

14th August 2025

2 of 27

Agenda

  • Introduction to SOLACE-AI
    • SOLACE-AI team
    • Background and rationale
    • Understanding the system
    • Approach
    • Live demonstration
    • Tentative case study examples

​

  • Participatory approach and role of stakeholder

​

  • Discussion
    • Development and data sources
    • Implementation
    • Broader considerations
    • Engagement

​

3 of 27

4 of 27

SOLACE-AI Background

  • 5-year grant under the Wellcome Trust call for “Advancing Evidence Synthesis for Climate and Health”

​

  • Started in April 2025, ending in 2030

​

  • Extreme events and climate change related emergencies will increase in frequency and intensity with far reaching health impacts

​

  • Preparing for, responding to and developing adaptation actions requires accurate and rapid evidence to inform decisions and policies

​

  • Traditional evidence synthesis methods are too slow, require too many resources, do not always take issues of health equity or local contexts into account, usually do not capture information from so-called grey literature

​

5 of 27

What is SOLACE-AI

  • Synthesizing Online Literature on Adaptation to Climate Emergencies

​

​

  • Scientifically sound AI system that will generate rapid, real-time evidence to support policy and humanitarian response to climate related health emergencies

​

​

  • Will draw on scientific literature, case studies, grey literature to synthesize accurate and context-specific summaries for policy makers, health and humanitarian workers

​

​

6 of 27

What is SOLACE-AI

  • SOLACE-AI synthesis can drastically reduce time, effort and cost of evidence synthesis

​

  • Evidence will be on-demand and up-to-date

​

  • Has the potential to address climate-health issues, countries, and health-equity issues that are often ignored

​

  • Contribute to better, evidence informed decision making and improve the lives of affected communities

​

​

7 of 27

How do Language Models work, anyway?

8 of 27

How do Language Models work, anyway?

9 of 27

How do Language Models work, anyway?

10 of 27

How do Language Models work, anyway?

Scale this idea massively (100s of billions of weights)

11 of 27

How do Language Models work, anyway?

Next word prediction

12 of 27

How do Language Models work, anyway?

13 of 27

What is our approach?

14 of 27

Approach

    • Engagement and co-creation of system with global partners and stakeholders living or working in climate change affected areas
    • Transdisciplinary community-based participatory approach, policy labs- both online and in-person

Workstream 1- Stakeholder engagement and co-creation

    • Methods to automatically identify evidence needed to address climate-health problems
    • Development of new LLM based methods to identify comprehensive collection of documents which address lateral language questions on climate-health problems

Workstream 2- Automating evidence identification

    • Automated critical summaries of individual evidence reports
    • Semi-structured extraction templates for evidence reports will be created
    • Evaluation of zero-shot and few-shot approaches for very large models, and fine-tuning methods for smaller language models

Workstream 3- Automated evidence summaries of reports

    • Design, implement, and evaluate novel strategies that will allow generation of rigorous and transparent overviews of article collections from both published and indexed sources, and unpublished sources
    • State-of-the-art machine learning translation approaches to provide evidence in users’ local or preferred language

Workstream 4- Automated, scientifically robust, synthesis

    • Combine processes in workstream 1-4 to provide real-time response to queries
    • Stress-testing and evaluating the system using tailored ‘red-teaming’ methods
    • Deploying the system in 5 unique and global case-studies to evaluate accuracy, usability, usefulness of outputs to inform and improve decision making

Workstream 5- Evaluation of methods and synthesis of final system

15 of 27

Live demonstration

16 of 27

Tentative case study examples

Ethiopia: Addressing poor mental health and substance misuse related to climate change

Somalia/Somaliland: Climate change related anthrax risk

India: Cardiovascular impacts of extreme heat in vulnerable populations

South Africa: Managing cholera risk after flooding

Scotland: Managing impacts of climate change on respiratory disease

Global/MSF: Responding to arbovirus outbreaks

17 of 27

Participatory approach and role of stakeholders

Participation in co-creation

    • Advising and providing critical oversight of project
        • Informing methods
        • Informing evaluation and system design (including user interface, etc)
        • Providing expertise, technical, and local knowledge

​

Contribute to defining case studies

      • Based on urgent problems faced by regions or communities
      • Case studies will provide focus for the design and real-world evaluation of SOLACE-AI

​

Regular engagement

      • Through emails, periodic online meetings, workshops (where feasible)
      • We will use participatory approaches eg storyboarding, and low-fidelity prototypes to co-design the system

​

​

18 of 27

Discussion

19 of 27

Discussion

  1. Developing the SOLACE-AI tool

​

  • Using the SOLACE-AI tool to respond to climate health emergencies

​

  • Broader considerations for SOLACE-AI

​

  • Stakeholder engagement with the project

​

​

​

20 of 27

Developing the SOLACE-AI tool

What other climate health case studies could the project consider?

​

What evidence and data sources should the tool use?

​

Any other considerations for developing the SOLACE-AI tool?

21 of 27

Tentative case study examples

Ethiopia: Addressing poor mental health and substance misuse related to climate change

Somalia/Somaliland: Climate change related anthrax risk

India: Cardiovascular impacts of extreme heat in vulnerable populations

South Africa: Managing cholera risk after flooding

Scotland: Managing impacts of climate change on respiratory disease

Global/MSF: Responding to arbovirus outbreaks

22 of 27

Using SOLACE-AI to respond to climate health emergencies

Who would use the SOLACE-AI tool?

​

How could it be used in practice?

​

How should the information be presented to users?

​

Should it give recommendations, as well as summarise evidence?

​

Any other considerations for using the SOLACE-AI tool?

​

23 of 27

Broader considerations for SOLACE-AI

How can it be adapted across different contexts?

​

What are the implications for the environment, inequalities, etc?

​

How can trust be built in AI, governance, evidence, etc?

​

What is needed for the next 5-10 years, and beyond?

​

Any other considerations?

​

24 of 27

Stakeholder engagement throughout the project

How do you want to be engaged throughout the project?

​

Who else needs to be involved in these conversations?

​

25 of 27

Participatory approach and role of stakeholders

Participation in co-creation

    • Advising and providing critical oversight of project
        • Informing methods
        • Informing evaluation and system design (including user interface, etc)
        • Providing expertise, technical, and local knowledge

​

Contribute to defining case studies

      • Based on urgent problems faced by regions or communities
      • Case studies will provide focus for the design and real-world evaluation of SOLACE-AI

​

Regular engagement

      • Through emails, periodic online meetings, workshops (where feasible)
      • We will use participatory approaches eg storyboarding, and low-fidelity prototypes to co-design the system

​

​

26 of 27

Next Steps

27 of 27

THANK YOU!

​

Global Stakeholder Meeting

​

14th August 2025

​

Website: www.solaceai.org

PI Contact: iain.marshall@kcl.ac.uk