So give your human validator a lineup - not 1 match but say the top 5 matches plus a few chosen randomly from lower down the confidence list, presented in a ran

So give your human validator a lineup - not 1 match but say the top 5 matches plus a few chosen randomly from lower down the confidence list, presented in a random order with no information on which was the algorithm's top choice.