Prediction errors from distinct perspectives induce separable EEG features for brain-computer interface

The ability to efficiently detect error is fundamental for human adaptive behaviors, and plays an increasingly crucial role in developing more intelligent brain-computer interface (BCI). Error-related potential (ErrP), which can reflect prediction error, has been widely used by the BCI to read whether outcomes accord with users’ expectation or not. However, current ErrP-BCI cannot distinguish the prediction error is induced by user’s own (first-person perspective, 1PP) or other’s (third-person perspective, 3PP) wrong action, hindering it from being applied in social interactions. This study used virtual reality (VR) to make subjects aware of prediction errors from the first- or third-person perspective, and recorded electroencephalogram (EEG) data of 22 healthy subjects. Event-related potential (ERP), event-related spectral perturbation (ERSP), inter-trial coherence (ITC) and shrinkage discriminant canonical pattern matching (SKDCPM) algorithm were used to investigate EEG features and the separability of prediction errors induced by distinct perspectives. Consequently, ErrP induced by the 1PP emerged significantly earlier than that of 3PP, and caused greater ERSP and ITC in the prefrontal region in the theta and alpha bands. Decoding result achieved 76.4%± 9.13% accuracy for the two types of errors (1PP-incorrect vs 3PP-incorrect). This study fills in the fine-grained classification of different error types and provides a finer metric for the systematic error correction efficiency of two-person collaborative brain control, which is the basis for future human-machine hybrid intelligence.