Our hunch was that many of the null results fell into a valley where the cognitive cost of verifying the AI's explanation were roughly the same as the cognitive costs of just doing the task yourself. When doing it yourself and verifying the AI are both costly, people ignore the explanations and just rely on the AI.