Conditioning to true content and artificial intelligence in psychophysiological intention recognition
Vladimir Randjelovic
DOI: https://doi.org/10.1016/j.ijpsycho.2023.112296
IF: 2.903
2024-01-06
International Journal of Psychophysiology
Abstract:Objective The objective is to introduce a novel method for classical conditioning to true content (CtTC), and for the first time, apply this approach in the concealed information test (CIT) to effectively discern intentions. During CtTC, participants are trained to exhibit electrodermal responses whenever they recognize true content on a screen. Additionally, the objective is to evaluate a novel CIT-dataset preprocessing algorithm, employed to enhance machine learning (ML) classification performance. Methods A total of 84 participants were evenly divided into four groups. Two groups of participants devised plans for stealing money from a supermarket, while the other two groups did not engage in any planning. One planning group and one non-planning group underwent CIT examination, while the remaining groups were subjected to CtTC. Results The CIT accuracy initially stood at 52 % and increased to 71 % after Z-score and ML classification (McNemar test, p < 0.05). Conversely, the CtTC accuracy was 76 % and significantly improved to 93 % following Z-score and 95 % following ML classification (McNemar test, p < 0.05). In the best-performing classifiers, CtTC exhibited significantly superior metrics for guilty/innocent classification compared to CIT (Fisher's exact test, p 0.90). In the CtTC group, reactivity and sensitivity significantly increased, indicated by higher EDR amplitudes (p < 0.05, two-tailed t -test, power 1 − β = 0.89) and the number of EDRs (p < 0.05, Fisher's exact test, power 1 − β = 0.90). There was no statistically significant difference between the Z-score and ML classification. Conclusions In the assessment of intentions, CtTC enhances both the sensitivity and accuracy of the CIT.
psychology, experimental,physiology,neurosciences, biological