What happened
Artificial intelligence models trained to detect sleep problems from brain waves work fine in healthy volunteers but break down completely when applied to stroke patients whose sleep patterns are disrupted. This means hospitals can't yet use automated AI systems to diagnose sleep disorders in the people who need the diagnosis most — those with serious neurological conditions.
Why it matters
This is honest failure documentation showing that AI succeeds in controlled research settings but fails on real patients in ways that matter clinically — the model looks at the wrong parts of the brain data when actual disease is present, and the researchers have proven it can't be deployed safely without being rebuilt specifically for sick populations.