Unmasking deceptive decisions in a two-player competitive game: Integrating ERPs, spectral-perturbation signatures, and interpretable neural-network decoding

초록

This study explores the feasibility of decoding deceptive or lying behavior from electroencephalography (EEG) neural signatures in a novel, interactive two-player game. The game was specifically designed to overcome previous methodological limitations in lie detection research, including insufficient incentives to lie, confounding lying with memory recollection, and a lack of control over participants' risk-taking tendencies. The approach utilized a multi-modal EEG analysis framework that incorporated event-related potentials (ERP), event-related spectral perturbations (ERSP), as well as neural network-based single-trial decoding to differentiate between instructed and spontaneous lying or truth-telling. Key findings revealed that early ERP components (P200 and N200) successfully differentiated instructed truth-telling from other conditions, while a late ERP component (N300) and late positive potentials were indicative of deceit under varying conditions, correlating with participants' risk-taking propensities. Alpha and low-beta spectral perturbations from the posterior regions further discriminated between the lying and truth-telling conditions. Cross-validated single-trial accuracy up to 57%-modest but reliable-demonstrates detectable deception signatures under strategic, incentive-calibrated interaction. This work advances our understanding of the neural dynamics of deception and lays a foundation for future real-world applications.

키워드