The use of artificial intelligence in interrogations: voluntary confession

Abstract

Interrogation is a crucial step in the investigation of criminal acts. Artificial intelligence has been used to increase the efficiency of interrogation. In this study, we developed a confession probability identification system to help investigators analyze the emotions of their interrogees while they are answering questions and determine the probability of them confessing. Based on these analysis results along with their own experience, investigators may adjust the content and direction of their interrogations to penetrate the interrogees’ defenses. The proposed system uses OpenFace and FaceReader to capture data and incorporates the multi-grained cascade forest (gcForest) and long short-term memory (LSTM) algorithms for deep learning. Our results indicated that the recognition accuracy of the gcForest algorithm exceeded that of the LSTM algorithm, which is consistent with the fact that the gcForest algorithm is more suitable for smaller sample sizes. In addition, heart-rate-based assessment may lead to erroneous determination of whether an interrogatee is telling the truth or lies because their heart rate may increase as a result of emotional responses.