Recently, in the field of artificial intelligence (AI), special attention has been paid to deep neural networks (DNNs) because of remarkable performance in various pattern-recognition tasks, especially in visual classification problems. However, interesting differences as well have been pointed out between human vision and current DNNs arousing questions about the generality of DNN computer vision, with the claim that AI should be more cognitive. This paper describes how human active perception is modeled for robotic cognition and simulates robotic awareness control through natural language communication with people, suggesting how machine learning (ML) should acquire knowledge to be available for higher level cognition such as abstract reasoning.