A new investigation has revealed that UK civil servants are allegedly using AI algorithms to make important life decisions regarding benefits and marriage licence approval.

The Guardian reported this after looking into the use of AI across the UK government, and also revealed that AI systems are continually being used by the Metropolitan Police for facial recognition.

While government applications of AI are often thought of as commonplace, such as for biometric passport checks, other applications are less transparent, and potentially disastrous for citizens.

For instance, the Guardian found that an AI algorithm employed by the Department for Work and Pensions (DWP) potentially lead to some families having their benefits removed, as it was accused of doing so by a minister.

Civil servants alerted Parliament to the AI tool’s bias, citing their concerns the DWP was increasing investment into this technology, with a planned £70m over the next two years.

The DWP says the AI is still learning, and insisted it did not use nationality as a factor when analysing applications, though they also did admit to not fully understanding the underlying algorithm in a hearing this September.

Further, an AI model used by the Home Office to approve marriage license was disproportionately flagging marriage applications from certain countries and nationalities as fraudulent.

In the Met Police, a facial recognition algorithm was less able to recognise different black faces than white faces in certain instances.

These startling revelations follow the disbandment of an independent advisory board which held the government to account for AI applications.

Concerns over AI bias are ramping up as their applications diversify and make more decisions over people’s lives.

The insights shine a spotlight on the UK government’s use of AI comes just a before its AI Safety Summit, set to take place on 1 and 2 November at Bletchley Park.

This potentially reveals bias within these government systems – AI algorithms do not create their own bias, and only rely on the bias inputted in their system via data.

Recommended reading

• Can Generative AI Improve Work/Life Balance?
• Collaboration is Key to UK Gov’s AI Safety Summit
• UK Gov Confirms AI Safety Summit to Take Place on 1st & 2nd of November

If AI algorithms, for instance, see more applications being denied for marriage licences from people from certain nationalities without other causes, then it may do the same depending on the algorithm employed by the system.

Generative AI can also be unpredictable, so can potentially make decisions somewhat indiscriminately.

The UK government’s use of AI is also antithetical to the EU’s current AI legislation, which restricts the use of AI in major decision making applications.

The use of algorithms with bias can put the UK government in jeopardy of being taken seriously as a proponent of AI safety and responsible AI ahead of the summit, which is set to discuss ethical applications of AI and the potential for regulation.