Home Office AI tools reduce asylum processing times

The Home Office has been piloting the use of artificial intelligence (AI) in the asylum decision making process, as it seeks to deliver a more efficient service without limiting human oversight or autonomy. 

The department is currently trialling two tools in this area of its work. The Asylum Case Summarisation (ACS) tool employs Large Language Model (LLM) technology to extract and condense information from asylum interview transcripts, helping reduce the time it takes users to read and analyse case documents and interview transcripts. The Asylum Policy Search (APS) tool functions as an AI search assistant that speeds up the process of finding and summarising relevant Country Policy Information Notes (CPINs) and Country of Origin Information Requests (COIRs). 

Using a combination of case logging, a survey of users and semi-structured interviews, the evaluation discovered that, despite some limitations, there are potential time savings associated with both the ACS and APS tools. On average, decision-makers utilising ACS processed transcripts 23 minutes faster per case and APS implementation yielded an average time saving of 37 minutes per case. In both cases, quality assurance reviews indicated no adverse impact on decision quality. 

The majority of users (65% for ACS and 54% for APS) want to continue to use the tools, but there were concerns about whether they were delivering the right amount of information, including a lack of source references. Although some users were not fully confident in the information provided by the tools, most thought they could be improved by adding additional functionality. This highlights the difficult balance with ‘human in the loop’ AI implementations, where additional functionality can improve efficiency but also has the potential to reduce human autonomy in decision making and risks an over-reliance on the technology.

At the end of December 2024, there were 90,686 asylum cases (relating to 124,802 people) awaiting an initial decision. Streamlining asylum processing will deliver clear benefits: alleviating the uncertainty for applicants caught in administrative limbo, reducing the impact on the public purse, and strengthening public confidence in the government’s competence regarding immigration. Although the ACS and APS tools are part of relatively small-scale pilots, they clearly demonstrate the potential AI has in delivering strategic benefits and productivity improvements. 

As we discussed earlier this year (see Plans and progress of UK government AI adoption), the government’s AI adoption progress to date has been underwhelming, struggling to generate the momentum and targeted funding to move beyond underdeveloped pilots. It needs to move quickly to address the barriers to using AI at scale.

Posted by: Dale Peters at 10:04

Tags: research   government   AI   trial   LLM