Algorithmic Deviance and Radicalisation in Digital Platform Societies: Neurocognitive Reinforcement and AI Recommendation Systems

Abstract

Artificial intelligence–driven recommender systems increasingly shape how information circulates within digital platforms and how users encounter political and social narratives. As a result, processes of radicalization, extremist mobilization, and digitally mediated deviance can no longer be explained solely by social strain or ideological indoctrination, but must also be understood within algorithmically curated environments designed to maximize user engagement. This research develops an interdisciplinary framework explaining how recommendation algorithms interact with neurocognitive reward mechanisms to reinforce and amplify radicalization pathways. Bringing together criminological theory, digital sociology, and cognitive neuroscience, the study draws on General Strain Theory, Social Learning Theory, and Actor–Network Theory, alongside research on dopaminergic reward systems, emotional salience processing, predictive coding, and neuroplasticity. Within this framework, the article introduces the concept of Algorithmic Strain Environments (ASEs), defined as digitally mediated ecosystems in which engagement-optimized recommendation systems repeatedly amplify grievance narratives, emotional arousal, and identity polarization through recursive feedback loops. To translate these dynamics into measurable signals, the study proposes four analytical indicators: the Extremity Drift Index (EDI), the Engagement Volatility Score (EVS), the Homophily Density Metric (HDM), and the Narrative Convergence Rate (NCR). These indicators are designed not only for retrospective analysis but also for early detection of radicalization trajectories, thereby positioning the model as a predictive rather than purely descriptive framework. A simulation based on a hypothetical dataset illustrates how such indicators can be integrated into a quantitative approach for analyzing algorithmically mediated radicalization dynamics. Finally, the article examines the governance implications of these processes within emerging regulatory frameworks, including the European Union Artificial Intelligence Act, the Digital Services Act, the United Kingdom Online Safety Act, and ongoing regulatory debates in the United States. It proposes a neuro-algorithmic governance framework that integrates algorithmic auditing, cognitive risk modeling, and systemic platform accountability. Overall, the findings suggest that radicalization in platform societies is increasingly shaped through the interaction between human cognitive vulnerabilities and engagement-driven algorithmic infrastructures, highlighting the need for governance approaches capable of addressing both technological design and neurocognitive reinforcement mechanisms.