Boston University study: Calling AI a 'coworker' makes managers 18% worse at error detection

Boston University professor Emma Wiles conducted a study where managers were asked to review work attributed to either an AI 'employee' (named Alex with a title and responsibilities) or a regular chatbot. When the AI was framed as a coworker, managers caught 18% fewer errors. The finding suggests that treating AI agents as human-like teammates incorrectly biases managers to trust the AI's output more and reduce their own critical oversight.

This research arrives as tech giants like Microsoft, OpenAI, Anthropic, and Google accelerate the release of tools for managing teams of AI agents—often marketed as digital colleagues. The study indicates that such framing is counterproductive, degrading human performance rather than augmenting it. James O'Donnell of MIT Technology Review warns this is 'a losing proposition for workers' and an alarming glimpse into a future where corporate terminology around AI could quietly erode human accountability and quality control.