Can we trust AI to detect healthy multilingual English speakers among the cognitively impaired cohort in the UK? An investigation using real-world conversational speech

Researchers at a UK institution have released a study examining the reliability of artificial intelligence models in detecting healthy multilingual English speakers among cognitively impaired individuals, published on February 26, 2026, amid concerns about rising dementia rates among ethnic minority populations in the country. The study, which appears on the arXiv preprint server, addresses a critical healthcare challenge as conversational speech is increasingly recognized as an indicator of early cognitive decline conditions such as dementia and Mild Cognitive Impairment (MCI). With one in four people in the UK belonging to ethnic minority groups, and projections showing dementia prevalence will grow most rapidly among Black and Asian communities, the researchers sought to evaluate whether AI diagnostic tools might exhibit bias when applied to these populations. The investigation specifically focuses on potential biases within AI models that are designed to analyze conversational speech patterns for signs of cognitive impairment, aiming to provide evidence that can help developers create more inclusive AI tools that can reliably distinguish between healthy cognitive function and early signs of decline in multilingual populations.