Explainable artificial intelligence approaches for brain-computer interfaces: a review and design space

Explainability in brain-computer interfaces remains developer-focused: Current XAI research for BCIs primarily serves researchers and developers seeking to justify model outcomes and improve performance, rather than addressing the needs of patients, clinicians, or other end-users who interact with these high-stakes medical devices.

There's a critical gap between explainability and real-world usability: The review identifies a fundamental tension in BCI systems—achieving explainability often compromises accuracy in the predictive models that interpret brain signals, yet the field lacks integrated frameworks to guide practical design decisions that balance these competing demands.

Standardized definitions are needed across disciplines: The paper emphasizes the importance of differentiating key concepts like "explainability," "interpretability," and "understanding" in BCI contexts, suggesting that confusion around these terms may be hindering responsible development and deployment of these technologies.

Broader stakeholder perspectives are underrepresented: By drawing insights from philosophy, psychology, and social sciences alongside technical approaches, the authors highlight that responsible XAI for BCIs requires input from multiple disciplines and stakeholder groups—not just computer scientists and engineers.

The field needs a comprehensive design framework: The proposed design space for XAI4BCI addresses the need for systematic approaches to visualization and investigation of brain-computer interface explanations, suggesting current ad-hoc methods are insufficient for ensuring trustworthy, safe deployment in clinical and assistive applications.