Competing Responses to the Growing Threat of Deepfakes

How can the use of emerging technologies be managed to balance innovation with ethics? Artificial intelligence, machine learning and other developments in digital technology can support valuable innovations. Yet the rapid development of AI has outpaced regulation, threatening data privacy, security, user trust, and authenticity.

How do human and machine evaluations of cloned voices differ in assessing realness, and what key vulnerabilities exist in current deepfake detection methods? The capstone project, advised by Dr. Gregory Gerling in the Systems Engineering department, explores the factors that influence the perceived realness of AI-generated cloned voices. Using commercial and open-source voice cloning tools, a voice library of cloned voices was created, varying in demographic factors (age, gender, ethnicity) and technical factors (cloning tool, training time, background noise). A smaller subset of both cloned and real voices were selected for an online survey for human listeners as well NISQA for machine evaluation of these cloned voices, with a focus on identifying key vulnerabilities in current deepfake detection methods.

In the US, how are social groups responding to the threats deepfakes pose to public safety, national security, and political stability? Social groups disagree about how best to regulate deepfakes. The rapid advancement of deepfake technology has sparked ethical concerns about its regulation. Most favor means of deterring criminal abuses, protecting vulnerable populations and preventing the spread of dangerous misinformation. Some groups also seek means of safeguarding democracy. Some groups, however, contend that excessive regulation may infringe freedom of expression.