I think that was the implication -- that experts are often wrong. One could go the No True Scotsman route and say that a true expert would occasionally select randomly following a Bayesian Bandit approach for optimal decision-making.