The UK's Home Office is potentially under investigation for its use of AI tools in processing asylum applications, with a government watchdog considering a probe following public concerns about accuracy and the department's secrecy.
The Independent Chief Inspector of Borders and Immigration (ICIBI), which oversees the Home Office, is reportedly gathering information from the digital rights group Open Rights Group (ORG) for its future inspection plans. ORG's research, which involved parliamentary questions and information requests, has highlighted concerns about the accuracy and transparency of these AI tools.
While the Home Office maintains that AI tools are only used to assist caseworkers with interview records and policy information, not to make decisions independently, a spokesperson stated that their performance is continuously monitored and evaluations will be published. The department uses tools like Asylum Case Summarisation (ACS) and Asylum Policy Search (APS), powered by OpenAI's GPT-4 and Microsoft Foundry, to speed up the review of asylum claims and related country information.
Researchers and lawmakers are worried about the potential for factual errors and bias in AI-generated summaries, especially given the high stakes involved in asylum decisions. ORG found that the actual error rate for ACS summaries might be higher than the 9 percent reported in pilot evaluations, and there's a lack of quality control for the APS tool. Concerns also exist about cultural biases within GPT models potentially leading to discriminatory outcomes.
Adding to the controversy, a UK judge recently accused the Home Office of using an "AI hallucinated" document in an asylum refusal. MP Bell Ribeiro-Addy voiced strong reservations, stating, "I just don’t think it’s appropriate that machines are doing that" when dealing with people's lives and sensitive circumstances.
Further complicating matters, the Home Office's AI tools are not listed on the government's public repository of algorithmic decision-making tools, as mandated by the UK's "AI Playbook." Asylum seekers are also reportedly not informed when these AI systems are used in their cases, which critics argue could breach UK data rights laws. Both ORG and the Public Law Project have shared their findings with the Information Commissioner, the UK's data protection regulator, which is already investigating the Home Office's handling of AI tool information.
