This paper examines how error-driven learning came to serve as a general model for explaining cognition across psychology and artificial intelligence. Tracing a trajectory from early twentieth-century notions of regulation through cybernetics, the perceptron, and gradient-based training, it argues that contemporary AI stabilizes learning as optimization relative to predefined criteria. Within this dominant configuration, error functions primarily as a regulative signal guiding convergence toward performance targets, while the organization that produces successful outputs can remain epistemically opaque. Developmental constructivist approaches, by contrast, treat systematic error as privileged evidence of an agent's current organization and as a driver of qualitative reorganization. Using Papert's constructionism as a bridge between computational practice and genetic epistemology, and drawing on Piaget's model of constructive equilibration and on the discussions of cybernetics that took place within the Geneva school, we articulate two ideal-typical regimes of error and describe common hybridizations. The distinction clarifies why large language models simultaneously satisfy behavioral markers historically associated with reasoning and yet provoke recurring doubts about understanding: dominant evaluation practices largely inherit regulatory assumptions that treat improvement as deviation reduction. We conclude that current disputes about machine cognition stem less from empirical uncertainty than from the historical naturalization of optimization as the operational definition of learning, and we outline minimal requirements for evaluation practices that aim at cognition in a strong organizational sense.
Industrial virtual reality (VR) training often involves related sequential sessions, creating opportunities for transfer of learning across scenarios. This study examined whether an additional VR training scenario influenced learning, situational awareness, as well as physiological and behavioral performance and their interrelations in a subsequent related scenario. A VR experiment compared a control-group that received only the target electricity-maintenance VR training with a transfer-group that first completed an additional gas-maintenance VR training before the same target scenario. The two scenarios differed in technical content but shared a comparable visual, procedural and safety structure. Results showed that additional VR training led to a higher baseline knowledge and overall knowledge levels, although knowledge gain did not differ significantly between the groups. Behavioral performance improved with additional VR training, evidenced by reduced time-on-task and fewer errors. However, while electrodermal activity (EDA) responses remained consistent across groups, significant differences in heart rate measures indicated additional VR training reduced stress in participants. Situational Awareness Rating Technique scores suggested that additional VR training does not significantly alter awareness perception. Multi-group structural equation modeling revealed distinct differences in how EDA explains post-training knowledge and the varying links between situational awareness components, knowledge, and behavioral measures. These findings advance the understanding of VR efficacy in training and transfer of learning within industrial and more ecologically valid scenarios. They also highlight the complexity of learning processes and the interplay between cognitive, behavioral, and physiological measures, providing a solid foundation for future exploration of how VR can benefit training.
This paper proposes the concept of an extended psychological profile and substantiates the importance of including empathy as a core factor in establishing trust between a human and an artificial empathetic partner. The global objective of this study is to investigate the psychological mechanisms of trust formation in the digital environment and to evaluate human vulnerability to empathetic impacts. To achieve this, we utilize two complementary experimental platforms modeling different communication channels: the social-emotional game “Teleport” (establishing trust through collaborative non-verbal actions) and the “Virtual Hotel Receptionist” system (establishing trust through verbal dialogue based on the eBICA cognitive architecture and language models). The paper systematically addresses three research tasks aligned with specific measurement approaches: the connection between empathy metrics and stable personality traits is determined via correlation analysis of the Boyko, Mehrabian, and Cattell 16PF scales; the effectiveness of structured dialogue with an empathetic partner in facilitating trust formation and fostering social contact is evaluated via a pre-post testing design using the Artificial Social Agent (ASA) questionnaire; the influence of empathy channels on subjective perception and partner evaluation under severe information deficits is verified through a blind indistinguishable setup in the gaming environment. The obtained results allow for modeling both pro-social interaction and human vulnerability to digital manipulations, including potential cyber-fraud scenarios.
