Cheated by deepfakes? deepfake detection ability, people’s reactions, and ethical implications

Date

2020-06-29

Authors

Jin, Lu (M.S. in information studies)

Journal Title

Journal ISSN

Volume Title

Publisher

Abstract

Recent dramatic developments in the fields of computer vision and deep learning technology have opened up a range of possibilities not previously imagined. The applications of computer vision technology include manipulating any face in any video and changing the environment of photos, just to name a couple of the new applications. However, these applications are already having impacts on our everyday lives. Given these recent advances in computer vision technology, people may not be able to trust images and videos we see on any media channel. These videos and images have the potential to deceive us. Throughout the history of technology development, the pros and cons of new technology are often in dispute. New technology is often sensationalized in terms of the benefits for people, which may go beyond anyone’s control and imagination. For example, the internet was started with a goal of developing a decentralized network. However, due to how it was commercialized in use, the Internet actually became more centralized than had been intended. Since a centralized platform has the advantage of controlling all users’ data and information, these can be sold to companies to help them engage in targeted marketing. Thus, the Internet fell short of its expectations and hype. Now, the focus and hype has largely shifted to artificial intelligence. In which direction will these new technologies go? What are humans’ relationships with these emerging technologies? How can we use this technology safely and ensure that it leads to a future that we want? This goal is the starting point for this report. In this report, I will use the latest FaceForensics++ dataset as a base for an experiment to answer three research questions: First, how well do people detect deepfakes, and what factors affect their ability to detect deepfakes? Second, what are their reactions when deepfakes are revealed ? Third, what do they see as the ethical implications of deepfakes, and how deepfakes could be used or abused? For RQ1, I explore the elements that can help people detect deepfakes. For RQ2, I evaluate their reactions. For RQ3, I explore how they perceive the ethical implications of deepfakes. More generally, my findings offer guidance for thinking about how to rebuild trust in video data in an era of deepfakes?

Department

Description

LCSH Subject Headings

Citation