Learning from rivals: the memory-inconsistent strategy