How is Al used to assess welfare benefits?

These AI technologies are trained on large amounts of data about previous welfare benefit applicants. This can include data about their employment history and disability status.

The systems use data about previous decisions about who is and isn’t eligible to learn patterns about the features of the decisions. The data about past decisions informs decisions about new applications. Many applicants will only be considered by staff once a system has marked them as eligible. 

The use of AI to automate administrative tasks can have positive effects. For example, it can make processing applications quicker. Research by Deloitte and the Committee on Fuel Poverty found that AI technologies could help improve identification, and provide financial help to ​‘fuel-poor’ households. In 2019, the Government Digital Service and the Office for Artificial Intelligence partnered with The Alan Turing Institute to produce joint guidance on how to build and use AI in the public sector.

However, several studies raise concerns about the harms that AI systems can cause, particularly when people, especially those from marginalised backgrounds, are not able to get support when they need it. An investigation from 2020 found that some councils have stopped using AI systems to make decisions about welfare and benefits, because of the risks of the decisions harming applicants.

Understanding how decisions are made is also important. Academics and civil society organisations are concerned that people affected by AI-led decisions often do not know how decisions are made, or how to challenge a decision to seek redress, in case of mistakes — although initiatives such as the joint guidance created by the Turing and the Information Commissioner’s Office aims to improve this.

In 2023, the legal charity Public Law Project started to list automated tools used by the UK Government.