In the canonical learning model, the multi-armed bandit with independent arms, a decision maker learns about the different alternatives only through his private experience. It is well known that any optimal experimentation strategy for this problem is ex-post inefficient: it sometimes leads the...