Dirk Scheele

Artificial intelligence (AI) companions can provide emotional support, but the availability, responsiveness, personalization, and affirmation that make them effective regulators may concentrate regulation within a single synthetic relationship. We propose a neurocognitive framework that situates AI companionship along a continuum from regulatory scaffolding to regulatory substitution, linking this transition to distributed systems supporting social-regulatory relief, motivational valuation, self-AI representation, and behavioral control. At the core of the framework is synthetic companion dependence, defined by reduced regulatory substitutability: functions served by a particular companion become increasingly difficult to redistribute across alternative people, strategies, or contexts. We distinguish regulatory scaffolding, which preserves or expands a broader regulatory repertoire, from regulatory substitution, in which regulation becomes concentrated within the synthetic relationship. Regulatory centrality, diminished control, and disruption sensitivity are candidate markers of dependence, whereas persistent displacement or functional impairment marks maladaptiveness. The framework links social-regulatory relief, motivational valuation, self-AI representation, and behavioral control to distributed neurocognitive systems and predicts a progressive shift in the relative neural prioritization of the preferred companion over viable regulatory alternatives. It thereby shifts attention from use intensity to whether regulatory functions remain flexibly replaceable.

Human health is profoundly shaped by the quality of our social environments, yet the mechanisms through which social experiences become biologically embedded remain poorly understood. Social Safety Theory (SST) proposes that perceived social safety and threat influence health through neuroendocrine and immune pathways, but it remains unclear which features of social interaction engage these pathways. Here, we argue that conversational artificial intelligence (AI) provides a powerful experimental tool to address this mechanistic gap. Synthetic agents allow researchers to systematically vary core social signals, including responsiveness, reciprocity, empathic cues, agency, and multimodal information, while holding other features of the interaction relatively constant. This controllability creates a unique opportunity to identify which social signals are necessary or sufficient to generate perceived safety or threat and to engage downstream biological responses. Building on SST, we propose a psychoneuroimmunological framework linking experimentally controlled features of synthetic social interaction to neuroendocrine and immune processes and derive testable hypotheses concerning perceived humanness, social-signal strength, and multimodal embodiment. By enabling the causal decomposition of social interaction, conversational AI could help reveal what makes an interaction biologically social and clarify the mechanisms linking social disconnection to health.

Patients with borderline personality disorder (BPD) suffer from severe emotional dysregulation and disturbances in body image and self-perception. Interoception, the processing and perception of internal body signals, is closely linked to emotional processing, but it remains unclear whether BPD impairs specific interoceptive facets and how these deficits respond to treatment. We investigated two key interoceptive facets, accuracy and attention, in BPD patients and healthy controls (HC) using self-report and objective measures before (n _BPD =55,n _HC = 31) and after (n _BPD =37,n _HC = 31) four-week residential Dialectical Behavior Therapy (DBT)/a waiting interval. Interoceptive accuracy and its metacognitive awareness were evaluated using a heartbeat discrimination task, while interoceptive attention was measured through questionnaires, intensity ratings, and uni- and multivariate neural responses during a functional magnetic resonance imaging (fMRI) interoceptive attention task. Before DBT, BPD patients showed reduced self-reported interoceptive attention, which was associated with more interpersonal problems. Patients further exhibited higher similarity in activity patterns evoked by cardiac interoceptive and exteroceptive attention in the insular and dorsal anterior cingulate cortex. There were no significant group differences in behavioral or self-reported interoceptive accuracy or metacognitive awareness. However, behavioral interoceptive accuracy was impaired in BPD patients with more severe symptoms. After four-week residential DBT, self-reported interoceptive attention significantly improved, while other interoceptive facets showed no significant changes. Our findings indicate that BPD involves selective interoceptive attention disturbances, with only self-reported measures responding to treatment. These results support multifaceted assessments of interoception and the potential benefit of interoceptive attention training for all BPD patients, with additional accuracy training in more severe cases.

Episodic memories can be shaped by various contextual factors. While social and sensory cues such as odors and music have been shown to influence encoding and retrieval, the role of tactile information remains unclear. In this preregistered study, we investigated the effects of affective touch on face memory. 57 healthy adults (40 women) completed the Cambridge Face Memory Tests and the Social Touch Questionnaire to assess general face recognition ability and attitudes toward social touch. During encoding, participants viewed neutral faces while receiving either static, dynamic, or no touch from a hidden experimenter and rated each face’s attractiveness and trustworthiness. Recognition was tested two days later. Outcome measures included recognition accuracy, metacognitive sensitivity (i.e., the ability of confidence ratings to distinguish between correct and incorrect responses), and judgments of attractiveness and trustworthiness. No significant differences emerged between touch conditions, and neither face recognition ability nor attitudes toward touch moderated these effects. Bayesian analyses provided moderate to strong evidence for the absence of touch effects on recognition accuracy and confidence ratings, whereas evidence regarding metacognitive sensitivity and trustworthiness evaluations was inconclusive. Overall, the findings suggest that brief social touch in a controlled laboratory context has limited measurable effects on face memory, and that richer social context may be required for touch to influence memory processes.

Background: Social disconnection has emerged as a major public health concern, with health risks comparable to established biomedical factors. At the same time, pain remains the leading cause of years lived with disability worldwide, imposing profound individual and societal costs. Although social factors are increasingly implicated in pain perception and chronification, existing evidence is fragmented across heterogeneous and often conflated constructs of social connectedness. It remains unclear whether subjective experiences such as loneliness or structural factors such as social isolation differentially relate to pain outcomes. A comprehensive synthesis directly comparing these dimensions has been lacking. Methods: We conducted a preregistered multivariate meta-analysis (PROSPERO: CRD420250643896) including 239 studies, 520 effect sizes, and 1,407,803 participants from clinical and non-clinical populations. Pain outcomes encompassed sensory, affective, cognitive, and functional domains. Social connectedness was operationalized as loneliness, social isolation, social support, and social exclusion. Multilevel random-effects models accounted for within-study dependency, with extensive sensitivity analyses and correction for small-study bias. Results: Across populations and social outcomes, greater social connectedness was associated with lower pain (z = −0.09, 95% CI −0.11 to −0.07). Notably, loneliness emerged as the strongest correlate (z = 0.14, 95% CI 0.11 to 0.17). Associations with social isolation were smaller compared to loneliness but were also significant (z = 0.09, 95% CI 0.05 to 0.13). Social support showed modest, significant inverse associations (z = −0.05, 95% CI −0.08 to −0.03), primarily confined to affective and somatic pain components. No reliable association was observed for social exclusion. Associations were consistent across age, sex, and clinical status, and longitudinal evidence supported temporal links between changes in social connectedness and subsequent pain outcomes. Conclusions: This large-scale synthesis identifies subjective social disconnection, particularly loneliness, as a robust correlate of pain across populations and pain dimensions, exceeding the relevance of objective social isolation. Given evidence linking loneliness to increased analgesic and psychotropic medication use, social disconnection may contribute to pharmacological burden and polypharmacy risk in vulnerable populations. Social connectedness emerges as a clinically meaningful, non-pharmacological determinant of pain and a potential target for integrative pain prevention and management strategies.

Episodic memories can be shaped by various contextual factors. While social and sensory cues such as odors and music have been shown to influence encoding and retrieval, the role of tactile information remains unclear. In this preregistered study, we investigated the effects of affective touch on face memory. 57 healthy adults (40 women) completed the Cambridge Face Memory Tests and the Social Touch Questionnaire to assess general face recognition ability and attitudes toward social touch. During encoding, participants viewed neutral faces while receiving either static, dynamic, or no touch from a hidden experimenter and rated each face’s attractiveness and trustworthiness. Recognition was tested two days later. Outcome measures included recognition accuracy, metacognitive sensitivity (i.e., the ability of confidence ratings to distinguish between correct and incorrect responses), and judgments of attractiveness and trustworthiness. No significant differences emerged between touch conditions, and neither face recognition ability nor attitudes toward touch moderated these effects. Bayesian analyses provided moderate to strong evidence that brief social touch during encoding does not influence face memory or social evaluations. These findings suggest that brief touch in a laboratory context exerts minimal effects and that meaningful social context may be required for touch to impact memory processes.

Background: Patients with borderline personality disorder (BPD) suffer from severe emotional dysregulation and disturbances in body image and self-perception. Interoception, the processing and perception of internal body signals, is closely linked to emotional processing, but it remains unclear whether BPD impairs specific interoceptive facets and how these deficits respond to treatment. Methods: We investigated the two key interoceptive facets, accuracy and attention, in 55 BPD patients and 31 healthy controls (HC) using self-report and objective measures before and after four-week residential Dialectical Behavior Therapy (DBT). Interoceptive accuracy and its metacognitive awareness were evaluated using a heartbeat discrimination task, while interoceptive attention was measured through questionnaires, intensity ratings, and both uni- and multivariate neural responses during a functional magnetic resonance imaging (fMRI) interoceptive attention task. Results: Before DBT, BPD patients showed reduced self-reported interoceptive attention, which was associated with more interpersonal problems. Patients further exhibited higher similarity in activity patterns evoked by cardiac interoceptive and exteroceptive attention in the insular and dorsal anterior cingulate cortex. There were no significant group differences in behavioral or self-reported interoceptive accuracy or metacognitive awareness. However, behavioral interoceptive accuracy was impaired in BPD patients with more severe symptoms. After four-week residential DBT, self-reported interoceptive attention significantly improved, while other interoceptive facets showed no significant changes. Conclusions: BPD involves disturbances in specific interoceptive facets that respond differently to treatment. Our findings support multifaceted assessments of interoception and the potential benefit of interoceptive attention training for all BPD patients, with additional accuracy training in more severe cases.