The Project

Towards a Population Dynamics Foundation Model for the UK

Understanding how populations change over time and space is essential for local authorities, institutions and small and medium enterprises (SMEs). Still, they seldom have the time and resources necessary to use smart data for that purpose. To address that need, I will advance my current work on geographic Artificial Intelligence (AI) methods to pioneer the development of the UK’s first open population dynamics foundation model. The model will be a new asset of the Smart Data Research UK, supporting researchers, policymakers, businesses and the public in understanding and planning for the future of our communities.

Digital fingerprints

A population dynamics foundation model can generate general-purpose, rich and concise digital fingerprints at fine geographic detail by integrating and compressing a broad spectrum of data on population, health, housing, mobility, the environment and the economy. Local authorities, institutions and SMEs will be able to use the digital fingerprints as input for their analyses, saving the time and resources necessary to process and model the data and their complex relationships.

Building on a strong tradition

The UK has a long tradition of creating open socio-demographic resources that inform policy and public debate. A well-known example is the Output Area Classification (OAC), which distils census data into accessible geodemographic profiles. These tools have proven the value of open data and code for delivering effective, policy-relevant insights about population dynamics. Over the last decade, advanced AI approaches have enabled us to move beyond traditional geodemographic methods, with novel frameworks capable of creating profiles that encapsulate more complex patterns.

The foundation model paradigm

More recently, the unprecedented advances made in modelling human language that have led to the “ChatGPT moment” and the current wave of generative AI have also sparked interest in a novel paradigm. Rather than focusing on task-specific models developed using a specific dataset for a specific purpose, the attention is shifted to creating general-purpose models trained on a wide range of datasets and applicable to many different tasks within a particular domain, commonly referred to as foundation models.

Fellowship components

The fellowship will allow me to develop the theoretical foundations and a prototype of the UK’s first open population dynamics foundation model at census output area level (OAFM). To establish national capability in this domain, lay the groundwork for the creation of a future, full-scale OAFM and ensure the creation of a trustworthy resource, the fellowship will include three main components.

• Engage: Establish and engage a stakeholder network, including local authorities and businesses, to ensure an OAFM addresses real-world tasks.
• Formulate: Advance the theoretical foundations of geographic AI, to deliver an effective framework for an OAFM.
• Develop: Build, validate, and release a prototype OAFM.

Impact

The stakeholder network will benefit from tailored case studies based on the prototype OAFM, which will be disseminated through workshops, meetings and an open-access report. Researchers in geography and the broader Smart Data Research UK community will benefit from the open-source prototype, which will be made available through a public repository, including code, output and resources. The roadmap outlined by the outputs of this fellowship will have the long-term benefit of the development of a full-scale OAFM, which will allow policymakers to assess national and local government initiatives through a more nuanced understanding of regional dynamics, and SMEs to make smarter, data-driven decisions on locations for new services and estimate local demand.