A preprint yet:
Aberrant sense of agency (SoA, a feeling of control over one’s own actions and their subsequent events) has been considered as key to understanding the pathology of schizophrenia. Behavioral studies demonstrated that both excessive and diminished SoAs were observed in schizophrenia. Several neurophysiological studies have suggested that such aberrant SoA may be due to temporal delays (TDs) in sensory-motor prediction signals. In the current study, we examined this hypothesis via a computational modeling approach using a recurrent neural network (RNN) model.
The sensory-motor prediction process in the behavioral task of SoA was modeled using an RNN model that can learn to reproduce the SoA task performance of human participants. The RNN model receives inputs of the states of visual, auditory, and proprioceptive senses, and generates forward predictions for those at the next time step and SoA judgement. In the simulation, the RNN models were trained using the actual behavioral data of 17 healthy controls completing an SoA task; then, our hypothesis was tested by adding TDs in the signals between context units of the RNN.
The RNN model successfully reproduced the behavioral features of the healthy controls. Moreover, bidirectional (i.e., excessive and diminished) schizophrenia-pattern SoA was reproduced with the TDs in context units. Three control experiments (random noise addition, TDs in outputs, and TDs in inputs) with quantitative analysis demonstrated no schizophrenia-pattern aberrant SoA.
The results theoretically support the proposed hypothesis that the aberrant SoA in schizophrenia may be due to TDs in sensory-motor prediction signals.