Skip to content
Project

Foundation models in the public sector

A rapid review of the development and deployment of foundation models in the UK public sector

pile of white cube building blocks stacked on top of each other
Project lead
Elliot Jones

At the Ada Lovelace Institute, we use the term ‘foundation models’ – which are also known as ‘general-purpose AI’ or ‘GPAI’. Definitions of GPAI and foundation models are similar and sometimes overlapping. We have chosen to use ‘foundation models’ as the core term to describe these technologies. We use the term ‘GPAI’ in quoted material, and where it is necessary for a particular explanation.

The rapid evolution of artificial intelligence has brought us to a point where foundation models are being widely adopted in a variety of sectors. From the launch of widely accessible large language model (LLM) interfaces such as ChatGPT, to image generators like Dall-E 2 and Stable Diffusion, our access to foundation models is easier than even.

We are already seeing search engines such as Google and Microsoft’s Bing embed foundation models into everyday search, tools like Photoshop integrate image generation models, and companies like Morgan Stanley use LLMs for internal knowledge search and retrieval.

Public- and private-sector organisations are looking at these foundation models with enthusiasm; not least as a potential way of tackling some of the big problems in society. These include the cost of living crisis, the desire for growth through innovation, solving the complexity of data sharing across government and the need for more efficient local government services.

Central government, local authorities and other public sector organisations are already considering how these can use these systems to assist with a broad spectrum of tasks. The tasks being considered include decision making, information and research sharing, enabling wider access to data, delivery of services and service provision monitoring.

In April 2023, the Department for Science, Innovation and Technology (DSIT) announced the launch of a £100m Foundation Model Taskforce to accelerate UK’s capabilities in AI and ‘generate better outcomes for people across the country through better public services’.

In this announcement, the Government promised that ‘the funding will be invested by the Foundation Model Taskforce in foundation model infrastructure and public service procurement, to create opportunities for domestic innovation. The first pilots targeting public services are expected to launch in the next six months.’[1]

While these uses may seem routine, there is a great deal of discussion and hype around the opportunities foundation models could bring, from generating emails to high level decision-making.

We are very much at the start of the foundation-model journey and at the start of our understanding of what opportunities and risks it could bring. It is still uncertain whether these systems will be accurate enough, reliable enough, and a good enough value-for-money to provide worthwhile solutions to existing problems.

Project aims

The Ada Lovelace Institute is undertaking a rapid review of the development and deployment of foundation models in the public sector, with a particular focus on the UK. The purpose of the review is to help identify current and potential use of the technology within local and national government to carefully understand, consider and specify the potential uses of foundation models in a transparent way, and what appropriate governance and oversight should look like.

We want to explore the opportunities, risks, and governance required by the use of foundation models in the public sector, through the following questions:

  1. What are foundation models, and how do we define and categorise the different kinds of terms around this technology?
  2. What foundation models are already being deployed by government, both formally and informally and identify what uses of GPI systems are in development or are being considered.
  3. What problems and opportunities do local and national government see the development and use of foundation models solving in the delivery of public services? Can these problems can be tackled using foundation models or whether there are better, less unstable and unregulated pre-existing tools already available?
  4. What are the risks, unintended consequences and limitations of foundation models? What kinds of risks would the use of these systems raise for governments and their autonomy?
  5. What are the regulatory challenges posed by the use of foundation models? How should government use of these technologies be governed? What hard rules or guidance should governments follow?

We are interested in speaking to people involved in the public sector use of foundation models across government, industry, and civil society. This engagement will be complemented by our own desk research.

The first output of this work will be an explainer on foundation model AI systems (and distinguishing them from large language models, generative AI, general-purpose AI etc.)

This will be followed by a rapid review report and supplementary analysis of how foundation models and related technologies are currently governed in the UK public sector.


[1] Department for Science, Innovation and Technology, Prime Minister’s Office, 10 Downing Street, Michelle Donelan MP and Rishi Sunak MP, ‘Initial £100 million for expert taskforce to help UK build and adopt next generation of safe AI’, (GOV.UK 24 April 2023), <www.gov.uk/government/news/initial-100-million-for-expert-taskforce-to-help-uk-build-and-adopt-next-generation-of-safe-ai> Accessed 23 May 2023.


Image credit: teekid