Not the machine’s fault: taxonomising AI failure as computational (mis)use

Abstract This paper proposes a re-examination of connectionist AI failures (controversial incidents) from the perspective of technological use. It advances four categories of failure: technically sound outputs inherent to connectionist programming; machine-world mis-configuration; motivational failure that deploys technology for illegitimate ends; and finally epistemic failure of misapplication where computing and AI are being used to solve for the wrong sets of social problems. Drawing on the history of computing, the paper argues that computational machines and its software (classical or connectionist) are numerical, procedural, and electronic in nature, and therefore, are geared to treat problems through the functions of numerical calculation, tabulation, approximation, and extrapolation. On account of these limitations, failure ensues when computers meet the many problems that cannot be solved on these grounds. The paper proposes and calls for an ontological comparison between computers and the problems they are pressed to serve prior to any pragmatic deployment.