근기(Near-term) AI의 형사절차적 활용방안 - 가정폭력 재범위험성 평가를 예로 하여 -

In recent years, artificial intelligence is in the spotlight in the field of ‘prediction’. In the United States, an artificial intelligence algorithm called COMPAS is widely used to judge sentencing, but various problems have been raised due to the technical nature of the algorithm called black box science and technology and the opacity of judicial procedures. It is difficult to expect artificial intelligence algorithms to be technically perfect at this stage. However, it is expected that a large amount of information processing through big data is possible, and rapid and neutral judgment is possible compared to human judgment. Artificial intelligence algorithms are not appropriate for predicting the risk of recidivism of all crimes. Domestic violence appears in a pattern similar to the perpetrator’s behavioral situation and the victim’s psychological state to some extent due to its nature. Accordingly, as part of the criminal legal use of artificial intelligence in recent years, we sought to introduce artificial intelligence algorithms to predict the risk of recidivism of domestic violence. The risk of recidivism of domestic violence is quantitatively judged, and this is used as judgment data in the domestic violence processing procedure. However, artificial intelligence algorithms fail to take into account the multidimensional aspects of recidivism risk, and the problem of overfitting can arise. Therefore, continuous improvement of artificial intelligence algorithms is needed. This paper proposes to construct and evaluate an “ethics matrix” for the ethical issues that incomplete AI algorithms will cause.