Calendar of events

Host: Professor Kim Wade

Title: The Gen-AI Generation: how worried should we be about the democratised ability to generate synthetic content?

Abstract: The advent of generative AI has taken the ability to generate fake content to a new level, and is changing the way we live. In 2018 Generative Adversarial Networks (GANs) started to become popular for image synthesis, but the more recent emergence of increasingly powerful diffusion models has since democratised access to content synthesis—allowing almost anyone to create diverse images, audio, and video via simple text prompts. Generative AI can have positive uses but inevitably invites nefarious exploitation, with harms including political disinformation, financial fraud, catfishing, and the creation of sexual digital forgeries (SDF). In this talk I will discuss my research examining: 1) the realism and trustworthiness of faces generated using GAN and diffusion models; and 2) human use of guidance information from (purportedly) either an AI algorithm or a group of human experts when completing simple decision-making tasks. Findings indicate that AI-generated faces are highly realistic and more trustworthy than real ones, and that AI-derived guidance may be uniquely placed to engender biases in humans. I will also discuss plans to develop a system to detect SDF that will be codesigned with the police, charities, and survivors of SDF-related crimes.