The UK Government should make open-source AI models ‘part of the menu’ for public sector tech adoption, or risk missing out on transparency, security and public trust benefits, according to a new report from the Social Market Foundation (SMF).

The cross-party think tank’s latest report, Keeping the Door Open: A roadmap for integrating open-source AI in public services, argues that because open-source AI models, unlike proprietary ‘closed’ models, can be more easily studied, modified, and distributed, these systems lend themselves better to the particular requirements of public services.

For example, despite security often being the biggest reservation around adopting open-source solutions, the report claims that these designs could enhance security for some of the UK’s most sensitive data, hosted by institutions such as the NHS, DWP, and HMRC.

Since open-source models could be run locally on government-owned data centres, the SMF said that citizens’ sensitive financial and health records would not have to leave trusted environments, reducing the risk of data leaks.    

According to the report, open-source AI models are also open to more scrutiny, so vulnerabilities can be more quickly identified and fixed, as well as allowing for a better understanding of model biases.

With open-source AI models able to be run directly by public bodies, rather than through renting existing software, adoption would also mitigate vendor lock-in. That could eliminate costly service charges and allow for low-cost tech reuse, offering more flexibility and value for money than using proprietary tools.

However, the SMF said that while the government has acknowledged the role open-source AI models could play, it has not taken any detailed positions on the differences between ‘open’ and ‘closed’ models, or how these will be deployed in the public sector.

Freedom of Information requests by the SMF revealed that although there are areas of government already developing or making use of open-source AI, notably DSIT and the Home Office, there are still major central departments which do not have any current or planned use cases, including the Cabinet Office, the Foreign Office, and the Treasury.

The SMF acknowledged that the so far slow adoption of open-source AI is in part due to significant challenges in rolling out a workable system.

Recommended reading

• Security and Legacy Tech Hinder Financial Services Open-source Ambitions
• Open-Source AI: The New Frontier for Developers
• New Report Reveals Rising Threat of Open Source Malware

Current funding, for example, emphasises one-off, project-based ventures, ill-suited to open-source development, while poor coordination between public bodies would undermine the reusability benefits of open-source AI.

Perhaps more of an issue is the skills gap, with the success of open-source relying on significant technical expertise, something the public sector has struggled to attract and retain.

“We need to use AI in our public sector to improve efficiency and unlock savings. However, concerns around the security, transparency and trustworthiness of the technology have to be addressed,” said Sam Robinson, head of AI at the SMF.

“The government has only vaguely acknowledged the role that open-source AI models can play in addressing these concerns, resulting in uneven and slow adoption. 

“Open-source AI offers a clear pathway forward for many key use cases in the public sector – but the government needs to give departments the clarity, resources and funding they need for its benefits to be tangibly realised.”