Accurate labels (word sequences that represent mental activity) require a solid conceptual understanding of the mind. Yet the subjective, experiential mind is often overlooked in BCI design. A typical BCI task might focus on just one or two aspects of mind (e.g., “visualize the cursor moving left” or adding “with high cognitive workload”). In reality, device control activates many mental components simultaneously. This presents a major opportunity: by incorporating more of these components into decoder labels, we can create richer, more effective classifiers.
The more comprehensively your labels represent the mind, the greater the percentage of brain activity you can decode. For instance, the command “imagine raising my left arm” can expand to include associations like grasping a phone, sending an important text, focused attention, excitement, confidence, sensory predictions, and inner speech. Each unique combination creates distinct brain signal features available for classification.
Classifiers and the Mind
A narrow view of the user’s mind leads to a narrow set of classifiers. If a classifier captures only a single mental state or process (“paradigm”), it can only decode the neural correlates of that one feature. In real-world tasks, however, dozens of mental states and processes operate—often at the same time. An imagined movement, for example, might involve imagination plus perception (visual, auditory, somatosensory), emotion (excitement, confidence, frustration, anxiety), motivation, goals, attention, cognitive workload, intention, prediction, and even factors like fatigue or pain.
Broadening mind categories and creating multiple classifiers for the same task provides more “shots on goal” for accurate decoding. This approach assumes a close (if not one-to-one) correspondence between subjective mental experience and brain activity.
The Mind/Brain Connection
Mind and brain activity are tightly linked, if not mirroring each other. Consider human movement: efferent signals can respond to virtually any mental state or process—not just direct motor intentions, but also perception, emotion, prediction, goals, and attention. Everyday observation confirms this—your head and eye movements, facial expressions, and body language shift instantly with thoughts, emotions, and intentions while reading this text.
Cognitive neuroscience is built on this connection. Specific stimuli, thoughts, emotions, and intentions reliably produce coordinated patterns of neural activity and distinct brain oscillations (Cox et al., 2018). The mind, therefore, offers a powerful way to reverse-engineer brain signals: acknowledge the mind as a real phenomenon, model it systematically, map its components across space and time, and link them to neural networks and signals. The resulting “signatures” become the foundation for robust classifiers.
A User-Centered Approach
Researchers have called for a more “user-centered” BCI design that accounts for users’ mental states and skills to boost efficiency, effectiveness, and usability (Lotte et al., 2018). I agree—and suggest extending this further to encompass the full range of mental states and processes. Rather than ad-hoc selections, we can systematically define the user’s mind using a comprehensive model, then focus on the components most active during a specific task and context (environment, situation, performance history, etc.).
Mind-Based Brain Signal Classifiers
As a mind component (or combination) expresses itself across trials, a consistent signature of brain activity emerges. This signature—representing a range of similar expressions rather than a sparse, one-off pattern—can form the basis of a classifier. Developing mind-based classifiers involves:
- Acknowledging the mind as a subjective phenomenon distinct from (yet linked to) the brain.
- Defining it clearly with a mind model.
- Identifying the most active mind components during a mental command, within the full mind/brain/body/environment context.
- Extracting corresponding brain signal characteristics to build signatures and classifiers.
- Prioritizing those that are most motivating, relevant and easy-to-repeat.
A rich set of classifiers (potentially dozens) dramatically increases decoding options. These can be selectively applied or weighted based on predicted mind states for a given task and context.
The main barrier has been the lack of a robust mind model. The human mind remains poorly understood, with no systematic way to define mental states in context (Poldrack & Yarkoni, 2016). A viable mind/brain model now exists to address this gap.
User Empowerment
Emphasizing the subjective mind empowers BCI users. By consciously manipulating their mind, users directly influence their brain signals. This clarity fosters more natural yet intentional control. Mind-based classifiers also support personalization—users can identify and refine the mental states that are easiest and most consistent for them, aligned with their interests, goals, and lifestyle. These become clear “mind targets” that improve both user performance and classification accuracy.
The Core Problem: The Mind Is Ignored
Despite cognitive neuroscience’s focus on linking mind and brain, the subjective mind is often minimized or overlooked in BCI work due to materialist tendencies. While tasks are described in detail, the broader mental experience behind them is largely ignored. This leads to suboptimal decoding and encoding, leaving out positive states (calm, confidence, motivation) and negative ones (frustration, anxiety, distraction) that strongly influence signals.
The solution? Treat any mental phenomenon as a potential classifier component. Strong, consistent states become primary classifiers; sporadic or noisy ones can be labeled as noise and filtered.
Summary
By accurately defining the user’s mind, linking it to brain activity, and building detailed signal signatures, we can develop higher-performing classifiers and clearer “mind targets.” A mind-first approach benefits neurotech professionals and users alike.
References
Chaudhary, U., Chander, B. S., Ohry, A., Jaramillo-Gonzalez, A., Lule, D., & Birbaumer, N. (2021). Brain Computer Interfaces for Assisted Communication in Paralysis and Quality of Life. International Journal of Neural Systems, 31(11). https://doi.org/10.1142/S0129065721300035
Chavarriaga, R., Fried-Oken, M., Kleih, S., Lotte, F., & Scherer, R. (2016). Heading for new shores! Overcoming pitfalls in BCI design. Brain-Computer Interfaces, 4(1-2), 60-73. https://doi.org/10.1080/2326263X.2016.1263916C
Cox, R., Schapiro, A., & Stickgold, R. (2018). Variability and stability of large-scale cortical oscillation patterns. Network Neuroscience, 2(4), 481-496. https://doi.org/10.1162/netn_a_00046
Lotte, F., Jeunet, C., Mladenovic, J., N’Kaoua, B., & Pillette, L. (2018). A BCI challenge for the signal processing community: considering the user in the loop. In Signal Processing and Machine Learning for Brain-Machine Interfaces. IET. Full text (HAL)
Poldrack, R. A., & Yarkoni, T. (2016). From brain maps to cognitive ontologies: Informatics and the search for mental structure. Annual Review of Psychology, 67, 587-612. https://doi.org/10.1146/annurev-psych-122414-033729 | PDF
