Spotting When Algorithms Are Wrong

Abstract Users of sociotechnical systems often have no way to independently verify whether the system output which they use make decisions is correct; are epistemically dependent on system. We argue that this leads problems when wrong, namely bad and violations norm practical reasoning. To prevent from occurring we suggest implementation defeaters: information a unreliable in specific case (undercutting defeat) or independent wrong (rebutting defeat). Practically, design defeaters based different ways might produce erroneous outputs, analyse suggestion with study risk classification algorithm used by Dutch tax agency.