DWP runs ‘generative AI Lighthouse programme – to test and learn in a safe environment’

Shortly after department updated its personal data policies, minister discusses initiative to use automated tech to support digital services ‘with a human touch in a safe, ethical and considered way’

The Department for Work and Pensions is running a dedicated “lighthouse programme” to explore the use of sophisticated artificial intelligence tools “in a safe and governed environment”.

The programme – which includes examination of generative AI, a technology exemplified by ChatGPT and Google Bard – is intended to help inform the department’s use of automation by allowing testing in controlled environment, according to social mobility minister Mims Davies.

“As part of our approach, and in line with the prime minister’s Foundation Model Taskforce, DWP has created a Generative Artificial Intelligence Lighthouse Programme which will safely guide our innovation in emerging Artificial Intelligence technology,” she said. “The role of this programme is to ‘test and learn’ in a safe and governed environment where all types of AI can be used to assist us in the delivery of our customer outcomes and department efficiencies.”

The department’s uses of the AI to date have included its implementation in helping to detect fraud in claims for Universal Credit advances – a deployment which civil society groups warned lacked “transparency and accountability”, and had the potential to embed bias. Such usage of automated tools is set to ramp up significantly in the coming months after the department recently revealed plans to invest £70m between now and 2025 in expanding the use of algorithms and analytics to help tackle benefit fraud.


Related content


Davies, answering a written parliamentary question from fellow Conservative MP Damien Moore, said that, even where AI supports decision making, a human always makes the final call if a citizen is to be denied a payment.

 “Where artificial intelligence is used to assist its activities in prevention and detection of fraud within UC applications, DWP always ensures appropriate safeguards are in place for the proportionate, ethical, and legal use of data with internal monitoring protocols adhered to,” she added. “Through the work of departmental governance, we can always explain how the AI reaches the conclusions through the use of data that it does. DWP will not use AI to replace human judgement to determine or deny a payment to a claimant; a human agent always makes final decisions, safeguarding the protection of individuals. Where appropriate, Equality and Data Protection Impact Assessments have been carried out.”

The minister pointed towards the rules and conditions laid out in the department’s personal information charter – which PublicTechnology recently revealed was amended to make more explicit reference to “automated processing in some decision making”. The linguistic changes – including the removal of references to “meaningful input from staff” – appear to provide for decisions to be made entirely via automation, where “the law allows this”.

Although the department declined to comment on the changes, it did refer PublicTechnology to an unchanged segment of the guidelines for personal data use which reads “DWP does not use AI to replace human judgement to determine or deny a payment to a claimant; a final decision in these circumstances always involves a human agent”.

Having reinforced this point, Davies added that the department intends that its wider use of automation will go hand in hand with a “human touch”.

“DWP is continually exploring the use of all types of artificial intelligence and its potential to support providing more digital services with a human touch in a safe, ethical and considered way,” she said. “Artificial intelligence will never replace the role of our colleagues in supporting customers throughout their journey. We are using artificial intelligence to undertake administrative or repeatable tasks freeing up our staff to spend more time with their claimants.”

Sam Trendall

Learn More →

Leave a Reply

Your email address will not be published. Required fields are marked *

Processing...
Thank you! Your subscription has been confirmed. You'll hear from us soon.
Subscribe to our newsletter
ErrorHere