Abstract
Machine learning algorithms boast remarkable predictive capabilities and deep learning, a branch of machine learning, has already provided the much required breakthroughs for recognition and authentication. This has enabled the deployment of a face recognition based biometric identification system by the Department of Homeland Security at U.S. airports. Deep learning algorithms require huge amounts of data for training. However, this shall not be an issue in smart cities. Instead, the ability to use the same deep learning technology for ulterior motives raise some issues. Machine learning algorithms have already been developed which can generate fake images and videos and rendering humans incapable of differentiating between real and generated content. Such content can be used for spreading disinformation regarding individuals and may lead to legal issues. Any system, relying on face recognition may be mislead using such technology. Similarly, researchers were able to create master finger prints, i.e., a set of finger prints that may be used instead of an original finger print to defeat authentication biometric systems. Alongside data pollution, these image and video processing issues may present significant challenges to governance of smart cities that shall rely on automated processing of such data. This paper presents an introduction to interpretability, attempts made to improve understanding of ML algorithms and their results. Area where we believe the definition of interpretability is lacking are also highlighted and a few example scenarios are presented. Finally, some possible directions for addressing the raised issues are introduced.