@inproceedings{OPUS4-7215, title = {Neuroadaptive Technology Conference 2025 Proceedings (NAT'25)}, editor = {Klug, Marius and Zander, Thorsten}, doi = {10.26127/BTUOpen-7215}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:co1-opus4-72151}, year = {2025}, abstract = {Neuroadaptive Technology (NAT) continues to advance the integration of real-time neurophysiological measures into adaptive intelligent systems. The Neuroadaptive Technology Conference 2025 (NAT'25), held in Berlin, Germany, brought together scientists and practitioners from neuroscience, computer science, engineering, and the social sciences, alongside representatives from politics, industry, and funding agencies. The conference emphasized both scientific and societal dimensions of neuroadaptive systems, particularly their implications for ethics, policy, and innovation ecosystems. Current developments showcased at NAT'25 highlight how neuroadaptive methods are being extended beyond the laboratory to real-world applications through Brain-Artificial Intelligence Interfaces (BAIs), explainable and neuroadaptive AI (XAI/NAI), and multimodal physiological computing. A recurrent theme across sessions was the societal responsibility accompanying the increasing agency of neuroadaptive and AI-driven systems — from workplace and healthcare applications to autonomous technologies. The programme included six keynote lectures, invited panels, and over forty peer-reviewed contributions covering Brain-Computer and Brain-AI Interfaces, Human-AI Co-Adaptation, Neuroadaptive Systems in Society, Ethics and Regulation, and Machine Learning for Applied Neuroscience. NAT'25 thus marked a further consolidation of the field toward human-aligned, context-sensitive, and socially responsible artificial intelligence.}, subject = {Machine learning; Artificial intelligence; Physiological computing; Brain-computer interface; Neuroadaptive technology; K{\"u}nstliche Intelligenz; Maschinelles Lernen; Neuroadaptive Technologie; K{\"u}nstliche Intelligenz; Maschinelles Lernen; Gehirn-Computer-Schnittstelle; Neurowissenschaften}, language = {en} } @article{GhermannZander2025, author = {Ghermann, Diana E. and Zander, Thorsten O.}, title = {Towards neuroadaptive chatbots : a feasibility study}, series = {Front. Neuroergonomics}, volume = {6}, journal = {Front. Neuroergonomics}, publisher = {Frontiers}, address = {Lausanne}, issn = {2673-6195}, doi = {10.3389/fnrgo.2025.1589734}, year = {2025}, abstract = {Large-language models (LLMs) are transforming most industries today and are set to become a cornerstone of the human digital experience. While integrating explicit human feedback into the training and development of LLM-based chatbots has been integral to the progress we see nowadays, more work is needed to understand how to best align them with human values. Implicit human feedback enabled by passive brain-computer interfaces (pBCIs) could potentially help unlock the hidden nuance of users' cognitive and affective states during interaction with chatbots. This study proposes an investigation on the feasibility of using pBCIs to decode mental states in reaction to text stimuli, to lay the groundwork for neuroadaptive chatbots. Two paradigms were created to elicit moral judgment and error-processing with text stimuli. Electroencephalography (EEG) data was recorded with 64 gel electrodes while participants completed reading tasks. Mental state classifiers were obtained in an offline manner with a windowed-means approach and linear discriminant analysis (LDA) for full-component and brain-component data. The corresponding event-related potentials (ERPs) were visually inspected. Moral salience was successfully decoded at a single-trial level, with an average calibration accuracy of 78\% on the basis of a data window of 600 ms. Subsequent classifiers were not able to distinguish moral judgment congruence (i.e., moral agreement) and incongruence (i.e., moral disagreement). Error processing in reaction to factual inaccuracy was decoded with an average calibration accuracy of 66\%. The identified ERPs for the investigated mental states partly aligned with other findings. With this study, we demonstrate the feasibility of using pBCIs to distinguish mental states from readers' brain data at a single-trial level. More work is needed to transition from offline to online investigations and to understand if reliable pBCI classifiers can also be obtained in less controlled language tasks and more realistic chatbot interactions. Our work marks preliminary steps for understanding and making use of neural-based implicit human feedback for LLM alignment.}, subject = {Passive brain-computer interfaces (pBCI); Large-language models (LLM); Error-processing; Moral judgment; AI alignment}, language = {en} } @phdthesis{Andreessen2023, author = {Andreeßen, Lena M.}, title = {Towards real-world applicability of neuroadaptive technologies : investigating subject-independence, task-independence and versatility of passive brain-computer interfaces}, doi = {10.26127/BTUOpen-6652}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:co1-opus4-66521}, school = {BTU Cottbus - Senftenberg}, year = {2023}, abstract = {Brain-computer interfacing (BCI) is a relatively new field of study that involves engineering, neuroscience, psychology, and physiology. A BCI is a system that allows direct communication between the brain and the environment, using signals from electroencephalography (EEG) to identify a user's intentions and mental states. This can be used for human-machine interaction (HMI) to adapt software or control a device. Particularly passive BCIs are the most promising for the general population, as they can detect and interpret a user's mental state without requiring their attention. This can enable neuroadaptive technology (NAT), which allows a machine to learn over time how the user perceives and interprets the world. However, there are still obstacles to be overcome before BCI technology can be applied in real-world situations for NAT. In this work I firstly address the issue of subject dependence in training data collection for classifier calibration, which can be time-consuming and impractical. I investigated the training of a subject-independent predictive model that is trained on a group of other users' data and applied to online-testing data of a new person. This was done in the context of training an automatic classifier for error detection and correction. Results showed that a classifier model can be trained without user-specific calibration and with high accuracy. The number of electrodes used in training the model was also reduced. Further it was validated that the trained classifier models were based on cortical sources and not other modalities. In a second study I address the issue of task-dependence. To that end I tested the application of a potentially task-independent calibration paradigm for mental workload assessment in a new task. This new task was a speed reading context, where subjects read texts of varying difficulty and speed. The study found that the mental workload prediction model was accurate in classifying mental workload in different reading tasks, indicating that it can be used as a task-independent classifier for mental workload. In a third study I examine if it is possible to measure neural correlates of human moral assessment using a passive BCI on a single-trial basis. A calibration paradigm was developed using pictures that were ranked as morally unacceptable and morally neutral. However, the results showed low classification accuracies and it was not possible to reliably distinguish between a user's subjective moral evaluations on a single-trial basis using current classification approaches. The results presented in this thesis provide solutions towards real-world applicability of NAT enabled by passive BCIs, as examples of a subject-independent and a task-independent classifier are demonstrated and discussed. Further, approaches for increased versatility of passive BCI technology are presented, that could make passive BCIs more feasible for use in future real-world human-machine interaction settings.}, subject = {Passive brain-computer interfaces; Mental state assessment; Neuroadaptive technology; Human-machine interaction; Electroencephalography; Passive Gehirn-Computer-Schnittstellen; Neuroadaptive Technologie; Mensch-Maschine-Interaktion; Elektroenzephalographie; Erfassung mentaler Zust{\"a}nde; Elektroencephalographie; Hirnfunktion; Mensch-Maschine-Interaktion; Gehirn-Computer-Schnittstelle}, language = {en} } @inproceedings{OPUS4-6271, title = {Programme and Proceedings of the Third Neuroadaptive Technology Conference 2022 (NAT'22)}, editor = {Zander, Thorsten and Fairclough, Stephen}, doi = {10.26127/BTUOpen-6271}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:co1-opus4-62719}, year = {2022}, abstract = {Neuroadaptive technology (NAT) utilizes real-time measures of neurophysiological activity within a closed control loop to create intelligent software adaptation. Measures of electrocortical and neurovascular brain activity are quantified to provide a dynamic representation of the psychological state of the user, with respect to cognitions, emotions and motivation. As such, NAT can access unique aspects of human information processing, and human intelligence, which can subsequently be used to enable more versatile and more human-like forms of machine intelligence. Current trends in different scientific fields indicate an increased interest in integrating context-sensitive information from the human brain into Artificial Intelligence. NAT'22, the Neuroadaptive Technology Conference 2022, was intended to bring scientists interested in Physiological Computing, Applied Neurosciences and Passive Brain-Computer Interfaces together with experts from the fields of Artificial Intelligence, Machine Learning and Intelligent Systems. The main goals of the conference were an exchange of research questions and findings from these fields and the identification of common goals and joint ventures in the domain of Neuroadaptive Technology, including: real-time signal processing, unsupervised vs. supervised ML, designing neuroadaptive interaction, explainable AI (XAI), neuroadaptive applications, hybrid AI systems (DL + symbolic AI) for applied neurosciences, ethics of neurotechnology in real world (responsibility for action, security), cloud-based solutions for data management and more. NAT'22 was held in L{\"u}bbenau, near Berlin, and organised by the Society for Neuroadaptive Technology. These Proceedings contain the abstracts of six keynote lectures and a total of 39 submissions in the categories of Brain-Computer Interface \& Applications, Ethics \& Perspectives, Artificial Intelligence \& Machine Learning, and a poster session.}, subject = {Neuroadaptive technology; Brain-computer interface; Physiological computing; Artificial intelligence; Machine learning; Neuroadaptive Technologie; K{\"u}nstliche Intelligenz; Maschinelles Lernen; K{\"u}nstliche Intelligenz; Maschinelles Lernen; Gehirn-Computer-Schnittstelle; Neurowissenschaften}, language = {en} } @article{KlugBergGramann2023, author = {Klug, Marius and Berg, Timo and Gramann, Klaus}, title = {Optimizing EEG ICA decomposition with data cleaning in stationary and mobile experiments}, series = {Scientific Reports}, volume = {14}, journal = {Scientific Reports}, publisher = {Springer}, address = {Heidelberg}, issn = {2045-2322}, doi = {10.1038/s41598-024-64919-3}, year = {2023}, abstract = {Electroencephalography (EEG) studies increasingly utilize more mobile experimental protocols, leading to more and stronger artifacts in the recorded data. Independent Component Analysis (ICA) is commonly used to remove these artifacts. It is standard practice to remove artifactual samples before ICA to improve the decomposition, for example using automatic tools such as the sample rejection option of the AMICA algorithm. However, the effects of movement intensity and the strength of automatic sample rejection on ICA decomposition have not been systematically evaluated. We conducted AMICA decompositions on eight open-access datasets with varying degrees of motion intensity using varying sample rejection criteria. We evaluated decomposition quality using mutual information of the components, the proportion of brain, muscle, and 'other' components, residual variance, and an exemplary signal-to-noise ratio. Within individual studies, increased movement significantly decreased decomposition quality, though this effect was not found across different studies. Cleaning strength significantly improved the decomposition, but the effect was smaller than expected. Our results suggest that the AMICA algorithm is robust even with limited data cleaning. Moderate cleaning, such as 5 to 10 iterations of the AMICA sample rejection, is likely to improve the decomposition of most datasets, regardless of motion intensity.}, subject = {Electroencephalography; Independent component analysis; Signal-to-Noise Ratio; Elektroencephalographie; Signal-Rausch Abstand; Unabh{\"a}ngige Komponentenanalyse}, language = {en} }