Michael Gaebler

: AI-generated images, videos, and news have become an inseparable part of daily life, leading to increasing skepticism toward online information. Understanding how expectations about the presence of AI-generated content influence perception is crucial for elucidating the underlying neural mechanisms and facilitating the generation of naturalistic avatars. In this study, we analyzed an existing EEG dataset (N = 29) in which participants viewed emotional facial photographs of only real individuals while being informed that upcoming faces were either real (“REAL”) or AI-generated (“FAKE”). We examined pre-stimulus oscillatory activities in the one-second EEG interval before stimulus onset and found significantly lower alpha power when participants expected “FAKE” compared to “REAL” faces. This effect was region-specific, particularly in right occipito-parietal and temporo-parietal regions, as identified by both sensor- and source- level analyses. In addition, the modulation effect between pre-stimulus alpha activity and post-stimulus event-related potentials (ERPs) was measured. A significant correlation between changes in late positive potential (LPP) amplitudes and pre-stimulus alpha power was observed exclusively for “FAKE” smiling faces, consistent with our previous findings. These results suggest that AI-related expectations modulate neural preparatory states, with lower alpha activity presumably reflecting increased attentional demands for stimuli believed to be artificially generated. This study demonstrates that top-down beliefs systematically shape both pre- and post-stimulus neural dynamics, providing new insights into how cognitive expectations bias perceptual processing, with implications for understanding human-AI interaction and improving the design and evaluation of AI- generated content in real-world contexts.

Stress involves coordinated central and peripheral processes that unfold dynamically and can be assessed through brain, autonomic, endocrine, and subjective measures. Centrally, acute stress has been linked to altered functional connectivity, particularly in the salience (SN), frontoparietal networks (FPN), and default mode networks (DMN). Here, we used cortical gradients to characterize stress-related reconfiguration in macroscale functional space and assessed their relation to peripheral stress measures. We performed secondary analyses on data from 67 young males completing the Trier Social Stress Test or a control task with resting-state fMRI before and after, concurrent peripheral (autonomic, endocrine) and subjective measures. To assess region- and network-specific changes in functional organization, we derived eccentricity and within- and between-network dispersion for the first three cortical gradients. Acute stress was associated with selective gradient reconfigurations in the right ventral prefrontal cortex and left insula and with increased SN–DMN and SN–FPN dispersion, indicating DMN and FPN decoupling from the SN. Although no associations with peripheral or subjective stress measures survived multiple-comparison correction, nominal effects suggested partly distinct links of saliva cortisol with local gradient changes and HRV with network-level reconfiguration. Together, these findings show that acute stress selectively reconfigures macroscale cortical organization.

Heart rate variability biofeedback has emerged as a validated intervention for stress management and emotional regulation through coherence breathing techniques. Existing biofeedback systems typically present cardiac and respiratory signals as separate feedback channels or rely on discrete interface elements that fragment attention and disrupt immersion. We present a design framework that extends Viscereality, a bioresponsive VR framework originally developed for breath mapping, to integrate multi-channel biofeedback including heart rate within immersive VR, centered on a particle-based environment that functions as a spatial body-mapping interface. By leveraging bioresponsive breath interactions for embodiment and weak representational aesthetics for cardiac coherence, the framework aims to encode both streams within a shared visual substrate, reducing the attentional task-switching and experiential fragmentation that multi-channel displays otherwise risk. The system translates lung volume into spatial expansion and contraction of the environment itself, maps cardiac coherence onto affective visual dimensions (angularity, brightness, saturation, symmetry), and embeds breathing guidance within the same geometric substrate. We present the design rationale and implementation of this framework, with empirical validation of its effectiveness to be conducted in future studies.

When the brain and body misalign, emotional experience and sense of reality can be disrupted. Although such atypical experiences are central to schizophrenia spectrum disorders (SSD), interoception, processing of internal bodily signals, remains poorly understood in individuals with SSD, particularly across subjective, behavioural, and neural domains. We tested whether SSD is associated with convergent alterations across interoceptive domains and whether these relate to clinical symptoms in a cross-sectional observational design. Patients with SSD ( n = 53) and matched healthy controls (HC; n = 60) completed an EEG experiment comprising eyes-closed and eyes-open resting-state recordings and a heartbeat counting task (HCT), followed by self-report measures. Interoceptive accuracy was derived from the HCT, while cortical responses to afferent cardiac signals were indexed by heartbeat evoked potentials (HEP). In individuals with SSD, subjective interoceptive awareness was altered, characterised by impaired regulation and negative bodily appraisal, alongside elevated depersonalization. At the behavioral level, interoceptive accuracy was marginally lower. HEP positivity was attenuated, most clearly during HCT and, to a lesser extent, in eyes-open rest, over centro-parietal regions, consistent with context-dependent alterations in cortical processing of afferent bodily signals. Symptom–interoception links were largely modest, with depersonalization emerging as the most consistent correlate of clinical severity, suggesting that interoceptive disturbances in SSD may reflect a trait-like, disorder-central alteration. Together, these findings indicate that SSD involves pervasive disturbances in bodily self-experience, marked by impaired interoceptive regulation and by context-dependent neural attenuation during interoceptive engagement. This profile highlights interoceptive dysfunction as a disorder-central feature with potential prognostic and therapeutic relevance.

Face perception typically occurs in three-dimensional space, where stereoscopic depth cues enrich the perception of facial features. Yet, most neurophysiological research on face processing relies on two-dimensional displays, potentially overlooking the role of stereoscopic depth information. Here, we combine immersive virtual reality (VR), electroencephalography (EEG), and eye tracking to examine the neural representation of faces under controlled manipulations of stereoscopic depth. Thirty-four participants viewed computer-generated faces with neutral, happy, angry, and surprised expressions in frontal view under monoscopic and stereoscopic viewing conditions. Using time-resolved multivariate decoding, we show that EEG signals in immersive VR conditions can reliably differentiate facial expressions. Stereoscopic depth cues elicited a distinct and decodable neural signature, confirming the sensitivity of our approach to depth-related processing. Yet, expression decoding remained robust across depth conditions, indicating that under controlled frontal viewing, the neural representation of behaviorally distinguishable facial expressions is invariant to binocular depth cues. Eye tracking showed that expression-related gaze patterns contained comparable information but did not account for neural representations, while depth information was absent in gaze patterns—consistent with dissociable representational processes. Our findings demonstrate the feasibility of EEG-based neural decoding in fully immersive VR as a tool for investigating face perception in naturalistic settings and provide new evidence for the stability of expression representations across depth variations in three-dimensional viewing conditions.

Face perception typically occurs in three-dimensional space, where stereoscopic depth cues enrich the perception of facial features. Yet, most neurophysiological research on face processing relies on two-dimensional displays, potentially overlooking the role of stereoscopic depth information. Here, we combine immersive virtual reality (VR), electroencephalography (EEG), and eye tracking to examine the neural representation of faces under controlled manipulations of stereoscopic depth. Thirty-four participants viewed computer-generated faces with neutral, happy, angry, and surprised expressions in frontal view under monoscopic and stereoscopic viewing conditions. Using time-resolved multivariate decoding, we show that EEG signals in immersive VR conditions can reliably differentiate facial expressions. Stereoscopic depth cues elicited a distinct and decodable neural signature, confirming the sensitivity of our approach to depth-related processing. Yet, expression decoding remained robust across depth conditions, indicating that under controlled frontal viewing, the neural representation of behaviorally distinguishable facial expressions is invariant to binocular depth cues. Eye tracking showed that expression-related gaze patterns contained comparable information but did not account for neural representations, while depth information was absent in gaze patterns—consistent with dissociable representational processes. Our findings demonstrate the feasibility of EEG-based neural decoding in fully immersive VR as a tool for investigating face perception in naturalistic settings and provide new evidence for the stability of expression representations across depth variations in three-dimensional viewing conditions.

Deep learning models highly accurately predict brain age from MRI but their explanatory capacity is limited. Explainable artificial intelligence (XAI) methods can identify relevant voxels contributing to model estimates, yet they do not reveal which biological features these voxels represent. In this study, we closed this gap by relating voxel-based contributions to brain-age estimates, extracted with XAI, to human-interpretable structural features of the aging brain. To this end, we associated participant-level XAI-based relevance maps extracted from two ensembles of 3D-convolutional neural networks (3D-CNNs) that were trained on T1-weighted and fluid-attenuated inversion recovery images of 1855 participants (age range 18–82 years), with regional cortical and subcortical gray matter volume and thickness, perivascular spaces (PVS), and water diffusion-based fractional anisotropy of major white matter tracts. We found that all neuroimaging markers of brain aging, except for PVS, were highly correlated with the XAI-based relevance maps. Overall, the strongest correlation was found between ventricular volume and relevance (r = 0.69), and by feature, temporal-parietal cortical thickness and volume, cerebellar gray matter volume, and frontal-occipital white matter tracts showed the strongest correlations with XAI-based relevance. Our ensembles of 3D-CNNs took into account a plethora of known aging processes in the brain to perform age prediction. Some age-associated features like PVS were not consistently considered by the models, and the cerebellum was more important than expected. Taken together, we highlight the ability of end-to-end deep learning models combined with XAI to reveal biologically relevant, multi-feature relationships in the brain.

Scientific discoveries often hinge on synthesizing decades of research, a task that potentially outstrips human information processing capacities. Large language models (LLMs) offer a solution. LLMs trained on the vast scientific literature could potentially integrate noisy yet interrelated findings to forecast novel results better than human experts. Here, to evaluate this possibility, we created BrainBench, a forward-looking benchmark for predicting neuroscience results. We find that LLMs surpass experts in predicting experimental outcomes. BrainGPT, an LLM we tuned on the neuroscience literature, performed better yet. Like human experts, when LLMs indicated high confidence in their predictions, their responses were more likely to be correct, which presages a future where LLMs assist humans in making discoveries. Our approach is not neuroscience specific and is transferable to other knowledge-intensive endeavours.

Previous studies have found that threatening stimuli are more readily perceived and more intensely experienced when presented during cardiac systole compared with diastole. Also, threatening stimuli are judged as physically closer than neutral ones. In a pre-registered study, we tested these effects and their interaction using a naturalistic (interactive and three-dimensional) experimental design in immersive virtual reality: we briefly displayed threatening and non-threatening animals (four each) at varying distances (1.5–5.5 m) to a group of young, healthy participants ( n = 41) while recording their electrocardiograms (ECGs). Participants then pointed to the location where they had seen the animal (approx. 29 000 trials in total). Our pre-registered analyses indicated that perceived distances to both threatening and non-threatening animals did not differ significantly between cardiac phases—with Bayesian analysis supporting the null hypothesis. There was also no evidence for an association between subjective fear and perceived proximity to threatening animals. These results contrast with previous findings that used verbal or declarative distance measures in less naturalistic experimental conditions. Furthermore, our findings suggest that the cardiac phase-related variation in threat processing may not generalize across different paradigms and may be less relevant in naturalistic scenarios than under more abstract experimental conditions.