'Rushed implementation' of AI in DWP benefits system could 'harm' people, experts fear

Keir Starmer has announced a plan to “transform our public services” through artificial intelligence (AI) – but experts have warned against “rushed implementation” which could cause “serious harm to people relying on support”. This includes within the Department for Work and Pensions (DWP).

Labour’s 50-point plan for “mainlining AI into the veins” of the UK does not explicitly mention the DWP, speaking more broadly about how it will “make our public services better” and “boost growth”.

But the government has previously announced that the DWP will be using “the latest technologies and AI” to reform Jobcentres, providing information on jobs, skills and other support to save work coaches time.

• AI was meant to make the DWP benefits system more efficient. Instead it’s brought bias and hunger
• Tony Blair thinks the DWP could be automated by AI. Here’s why that’s so dangerous

In fact, the DWP is a government department which is already using AI and machine learning to “read and automatically identify the most vulnerable people”, intended to connect them to support more quickly. It also uses automated systems to detect fraud and error within the welfare system.

The Tony Blair Institute previously found that the DWP could free up 40% of its time using AI tools, equivalent to a productivity gain of close to £1bn a year.

Shelley Hopkinson, head of policy and influencing at Turn2us, said: “AI has the potential to improve speed and consistency in the social security system, but errors from rushed implementation could cause serious harm to people relying on support. We are concerned that the historical data used to train AI may perpetuate bias and discrimination against marginalised groups.”

An investigation by the Guardian found that the machine-learning programme used by the DWP to detect universal credit fraud incorrectly selected people from some groups more than others. It showed that there was bias according to people’s age, disability, marital status and nationality.

Further research found that 200,000 people have been wrongly investigated for housing benefit fraud and error because of poor algorithmic judgment.

The Big Issue has previously reported on the impact errors within the benefits system, with people being mistakenly accused of owing the DWP thousands of pounds. One single mother who was mistakenly accused of owing £12,000 described being left “devastated” and “fearful of ever claiming again”.

Hopkinson added: “AI integration should be guided by consultation and transparency so that the system meets the needs of everyone and builds trust that decisions are fair and accurate. This must come with clear accountability and safeguards to allow for challenge. AI must work for people, not against them, prioritising people’s lives and wellbeing in decision-making.”

Do you have a story to tell or opinions to share about this? Get in touch and tell us more. Big Issue exists to give homeless and marginalised people the opportunity to earn an income. To support our work buy a copy of the magazine or get the app from the App Store or Google Play