A bandit problem is classic in sequential decision making. The explicit goal in the context of medicine is to treat patients as effectively as possible, those in a clinical trial and those who come after the trial. The latter patients benefit from the results of the trial but so do the former patients. The optimal decision (clinical trial design) balances the conflicting desiderata of treating the current patient effectively (earning) while simultaneously getting information for treating subsequent patients (learning). I will describe the introduction of bandit strategies and their modifications into medicine, via what I call platform clinical trials.