Study on Trust and Delegation in Human-AI Question Answering

A recent study published on arXiv examines how individuals choose to assign tasks to AI or follow its recommendations during cooperative question answering. The research featured 24 matches that paired 23 expert humans with 16 AI agents, leading to the collection of 387 delegation decisions and 1,440 adoption choices. The findings indicate that collaboration between humans and AI yields better results than either working independently; however, humans frequently err in determining when to place their trust in AI. This study differentiates between delegation (allowing AI to operate independently without reviewing its output) and adoption (assessing AI recommendations), a dual focus that has been infrequently addressed in previous research. The results aim to enhance the understanding of human dependency on AI in teamwork contexts.