Interpretability of neural networks is crucial in developing trust between humans and machines and human's ability to take decisions based on the outputs of neural networks. In this paper, the role observer/interpreter in the interpretability is discussed, where interpretability refers to the confidence an arbitrary observer has in the decisions based on the outputs from output generating systems and observer can be a machine or a human. The first step in the interpretability is to ascribe meaning to the output which observer receives from the system. To ascribe meaning, binary system representation scheme is employed where an output is a unique configuration of n binary systems. Furthermore, it has been shown that outputs of neural networks and dreams are fundamentally equivalent, given that observer employs same representation scheme. Lastly, two governing factors in interpretability; observer's understanding of the phenomena and prior performance data of output generating system pertaining to phenomena are qualitatively discussed, taking inspiration from chess.