Using Artificial Intelligence to Detect Lies During Court Hearings

AI for detecting lies during court hearings

Abstract

Globally, the judicial system is overworked and under-resourced, with fewer judges, longer trial durations of trials, an ever-increasing number of filed cases, coupled with defendants challenging verdicts of lower courts in upper courts. As false testimonies unnecessarily jeopardize administering justice, this paper focuses on developing an AI-based framework for detecting lies. Our proposed ‘AI-based lie detection framework’ utilizes text, audio, and visual footage of the courtroom to differentiate between truth and falsehood. In particular, we employ a multilayer perceptron for textual signals, a long-short-term memory neural network model for audio feed, and a two-stream convolution neural network for video classification. Experiments on different neural architectures conclude that our proposed hybrid model outperformed existing techniques with an AUC score of 0.93 for text, 0.88 for audio, and 0.98 for video-based deception detection.