No government in 2023 can afford to put its head in the sand when it comes to AI. Whether we like it or not, the genie is out of the bottle and artificial intelligence is only going to become more prominent in all our lives.
Amid concerns that UK AI regulation is being outpaced by the EU, the Labour Party has set out its stall for how it would approach AI if it’s successful in forming a government at the next election. While this is to be welcomed, it must carefully avoid ethical and legal pitfalls.
One such proposal is to use AI systems at the Department of Work and Pensions (DWP), to identify benefits fraud and error, help match unemployed people with suitable jobs, and process claims more efficiently.
It’s an admirable goal, but one riddled with potential problems. Any government looking to use AI to tackle fraud to a greater degree, particularly based on lessons derived from historical data, would do well to remember the Dutch childcare benefits scandal.
In short, the affair centered around a machine learning (ML) algorithm used by the Dutch government to identify fraudulent childcare benefits claimed.
The system wrongly flagged applicants as risky and was later found by an investigatory committee to contain institutional biases, which were reflected in the algorithm’s output. The result was tens of thousands of innocent parents accused of benefits fraud.
Although there’s a clear political lesson here - the scandal led to the downfall of the Dutch government - there’s also a clear technological one too. AI is best applied to areas in need of improved productivity, but must always play second fiddle to human workers.
The DWP’s current use of machine learning is actually a perfect example of this in action. It uses an algorithm trained on bundles of historical claims data to spit out potentially fraudulent Universal Credit claims.
Cases flagged by the algorithm are then reviewed by humans - a shining example of how tech can be used as an accelerant rather than an agent of total reform.
Shadow secretary of state for work and pensions Jonathan Ashworth has indicated that human workers would still make final decisions under Labour’s AI plans, which raises the question of how the party thinks its approach would improve upon DWP’s existing algorithm.
At the end of the day, if Labour is serious about using AI to improve the economy and recoup lost government spending, it could do a far better job applying it to HMRC than the DWP, as the former needs far greater scrutiny.
Benefits overpayments totaled £8.3 billion in FY2023, while underpayments totaled £3.3 billion. In comparison, a parliamentary report into tax compliance found that HMRC has failed to chase £42 billion in unpaid taxes and only around £1.1 billion of an estimated £4.5 billion lost during the pandemic to erroneous and fraudulent support claims.
Applied sensitively, AI would play a far more significant role in buttoning up the UK’s tax landscape than it could in working on the margins of unemployment.
A more effective HMRC, bolstered by thoughtful AI implementation, would bring an immense boost to the UK economy. This would come as a benefit to all operating in the UK, with businesses able to operate in a more thriving market and the government in a more secure place to provide corporate subsidies.
But while Labour rightfully highlights the threat of AI to jobs - supported by the recent cases of AI-driven mass layoffs and Big Tech hiring overhauls - the boosts to productivity and system efficiency for organizations that are struggling for staff are too tantalizing to pass up.
With HMRC having been identified as one such area in need of improvement, it would be foolish to not pursue this further.
