Detecting DeepFakes for future robotic systems

Abstract

Future robotic systems are likely to include AI-based professional to provide expert advice. For example, AI-based legal experts or judges try to explain the laws, and AI-based medical experts provide medical diagnosis and corresponding prescriptions. DeepFake is a collective of algorithms that aim to replace the face in one source video with the face of another person. The advancements in deep learning has made video manipulation and synthesis much more accessible than before. Thus,what if an AI-based DeepFake synthetic judge tries to explain the law differently and makes offensive decisions, or if a DeepFake synthetic doctor tries to give a wrong diagnosis and prescribes harmful medicines? Effective detection methods are important to combat malicious spread and use of damaging fake media as well as harmful fake robots in future robotic expert systems. This chapter first provides the general background on future robot-based expert systems,and on DeepFake,such as facial manipulation,forged video datasets,and detection techniques. These include 3D modeling and deep-learning-based manipulation techniques,forged video datasets from different sources,and current detection methods. This chapter covers a proposed detection method that utilizes image segmentation as part of the solution.