AI-generated Anonymization in VR Journalism

AI-generated Anonymization in VR Journalism

AI GENERATED ANONYMITY IN VR JOURNALISM from Making Culture Lab on Vimeo.

Taylor Owen (Graduate School of Journalism, University of British Columbia), Kate Hennessy and Steve DiPaola (School of Interactive Arts and Technology, Simon Fraser University)

This project was awarded a 2017-2018 Knight Foundation/ Google News Lab / Online News Association Journalism 360 Challenge Grant.

Abstraction from our AI neural network system. © Steve DiPaola

This project explores the use of computational abstraction in video-based virtual reality journalism for the purpose of anonymizing subjects. Using an AI video filter tool developed by project collaborator Steve DiPaola, we will test whether emotional resonance, or empathy, can be maintained when a character is algorithmically distorted to the point at which their identity is hidden. Our goal is to address a central challenge in VR journalism – that standard tools of anonymization such as blurring or blacking out faces and silhouettes breaks the emotional connection that is the central premise of the form. Using animated abstraction could allow for a viewer to remain emotionally connected to a character or subject, while still granting the confidentiality needed for investigative journalism.

This project has three components. First, we are adapting an AI filter for use in 360 3D video environments. Second, we are running a series of experiments to determine whether the empathetic connection created in immersive video translates to abstracted animations. This involves three tests: (1) a study to determine which magnitude and types of abstraction best convey an emotional connection of an interview; (2) determine at what point of abstraction emotional connection is lost; and (3) develop a low barrier interactive tool to allow interview subjects to fine tune their own level of abstraction. Third, we will produce a short piece of journalism portraying various characters with different levels of anonymization.