AI second opinion falters when problems are rare, study finds

This study aimed to understand how often a problem occurs (its 'prevalence') affects how well people make decisions when they get help from an artificial intelligence (AI) system. Twelve participants completed a computer task where they had to spot 'bad' items, similar to a doctor looking for a disease. Researchers tested if AI advice was more or less helpful when 'bad' items were very common (90% of the time) or very rare (10% of the time). The goal was to gather knowledge about human-AI teamwork, not to treat any medical condition.