The Department for Work & Pensions (DWP) has confirmed it is using AI to help find overpayments and fraudulent claims across a range of benefits that include Universal Credit and Personal Independence Payment (PIP). An AI mechanism dubbed ‘intelligent claims analysis’ has been deployed to spotlight cases warranting additional scrutiny.
Millions claim Universal Credit and PIP, benefits aimed at assisting those facing economic challenges or living with long-term disabilities. The DWP said this AI tool aids in uncovering discrepancies by comparing current claim information against prior submissions and other databases like tax and employment records.
Human agents assess cases flagged by the tool before any further steps are executed. The approach adopted includes parsing vital signs within an application, setting them against archived data patterns, and allocating a “risk score”, and the DWP boasts an 89% success rate in identifying claims that necessitate extra inspection.
Following the AI’s “red flag”, the matter is entered into the fraud and compliance infrastructure of the department, leading to a rigorous manual review by dedicated experts. This inclusion of human judgement ensures that resolutions concerning payment suspension or denial are made with human decision-making at the core.
Though operational since mid-2021, the department has only recently made public the specifics of this technology. Reportedly, the system has overseen in excess of 1.2 million case evaluations, heralding a significant step in automation within the DWP.
The system has notably reduced roughly 60,000 operational hours by minimising needless manual reviews. Nonetheless, early versions of the system from 2021 to 2023 did experience hiccups with a greater number of false positives than anticipated, where valid claims were mistakenly flagged.
Through refinements in the algorithm that now considers a more diverse array of contextual factors, its accuracy in prediction has been enhanced.
Concerns about the impartiality and transparency of using automation in benefit decisions remain pressing issues. Advocacy groups caution that AI systems tasked with detecting fraud might unduly affect those in precarious situations.
James Walker, director at Welfare Rights UK, voiced his concerns: “While ensuring taxpayers’ money is used appropriately is important, we must also ensure that AI-driven fraud detection does not unfairly target those who genuinely need support. The DWP must provide clear explanations and appeal routes for those affected by AI-flagged claims.”
In an incident reported in December, another AI tool deployed by the DWP for identifying welfare fraud was revealed to have biases associated with age, disability, and nationality. An in-house review conceded that there were “statistically significant outcome disparities” in its outcomes, as noted by The Guardian.
Despite this, the DWP asserts that their latest claims analysis tool bears minimal bias risk because it doesn’t pull upon personal identifiers such as names or ethnic backgrounds, but rather zeroes in on patterns in claims and monetary information. The department’s own assessment of the tool also acknowledges.
Addressing concerns that automation might reduce human control, the report highlights: “To mitigate this, staff have received training to critically assess flagged cases and ensure AI recommendations are not accepted without scrutiny.”
This comes as news emerges about the use of an AI tool by the Department for Work and Pensions (DWP), amidst Keir Starmer’s drive with Labour to accelerate the implementation of AI in public services. The Prime Minister is banking on AI to enhance the efficiency of systems within welfare, healthcare, and education.
However, despite the optimism, this fraud detection system remains one of the scarce few AI technologies the DWP has revealed on the algorithm transparency register—a requisite for every government department for over a year. The DWP acknowledges the existence of an internal catalogue of AI tools, yet it withholds publishing the full roster, despite repeated requests.
In response to a September query under the Freedom of Information Act requesting disclosure of the complete array of AI systems, the DWP commented: “Public authorities like the DWP must retain control over how and when they release information. The ability to manage disclosures is essential to the effective operation of government functions.”
With the focus likely set on Universal Credit, given its substantial claimant base of 6.4 million individuals as of January 2024, as reported by the Express, officials may also zoom in on those receiving Jobseeker’s Allowance, Employment and Support Allowance, and Housing Benefit due to historically higher discrepancies associated with these benefits.