The UK's Department for Work and Pensions has discontinued at least six artificial intelligence prototypes, according to The Guardian, offering insights into the challenges of scaling AI from proof-of-concept to production in large organisations.
Among the discontinued projects were two that had been highlighted in DWP's annual report as successful proofs of concept: A-cubed, designed to help staff guide jobseekers into work, and Aigent, intended to accelerate the processing of personal independence payments for people with disabilities.
According to meeting notes released under FoI, DWP officials identified key challenges in moving AI systems from proofs of concept to full implementation, specifically noting the need to ensure "products are scalable, reliable [and] thoroughly tested." The scope of AI testing has been substantial - out of 57 ideas tested through i.AI, the government's AI incubator, only 11 progressed to various stages of testing and scaling, according to the incubator's director Laura Gilbert.
The Guardian notes that the implementation process has involved collaboration with major technology providers, with Gilbert noting that i.AI has been working with OpenAI, Anthropic, Google, and Microsoft. The Guardian also raises the poiunt that no information about AI used by the DWP in the welfare system has been disclosed on the government's algorithm transparency register, despite this being a Whitehall requirement for almost a year.
The DWP's experience illustrates how even large organisations with significant resources must navigate complex challenges in AI implementation. Their experience with multiple parallel pilots provides insights into realistic success rates for AI initiatives and highlights the importance of thorough testing and evaluation protocols before full-scale deployment.