We explore the use of brain activity in scenarios of Human-Computer Interaction. Specifically, we aim at the detection of EEG correlates of error awareness to dynamically adapt a Human activity recognition system. We design a Human Computer Interaction experiment which consists in: - a memory game controlled by a Human Activity Recognition System - an EEG - Error Potential (ErrP) detection System We use EEG signal processing to recognize error related potentials (ErrP) on single trial basis. ErrP are emitted when a human observes an unexpected behaviour in a system: we propose and evaluate performance improvements provided by the ErrP detection system as a "teacher" for the on-line adaptation of a user centered activity recognition system. The gesture recognition system becomes self-aware of its performance, and can self-improve through re-occurring detection of ErrP signals.