Analyzing Affect and Identity Through the Creation of Deepfakes

open_access
Open Access
Author:
Franco Prasanna, Gracy
Area of Honors:
Data Sciences
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Betsy Campbell, Thesis Supervisor
Suhang Wang, Thesis Honors Advisor
Keywords:
Deepfakes
Autoethnography
Social Justice
Identity
Affect
Abstract:
Deepfakes are a generative artificial intelligence technology that replicate individuals’ likenesses through facial, movement, and voice cloning. This research aims to address a gap in existing literature regarding such technology by providing an ontological perspective to existing deepfake literature. Specifically, this research delves into my process and insights during the creation of my own deepfake using the deepfake technology Colossyan, and my observations while leading a deepfake creation workshop for a participant with a speech impediment resulting from aphasia. The data collected through this process comes from my observations regarding deepfakes, generative AI, society, media, and the political climate during the time frame of this research, that I routinely logged in a physical journal, as well as my insights drawn from the workshops. From these insights, then, I reflected on my own feelings and emotional reactions related to my identity, and how this was reflected by the deepfake that I created, as well as all my recorded observations. How did I feel about the way the deepfake represented my “look” and “sound”, and how did I feel about my perceptions of the representation of the workshop participant with the speech-related disability? This research explored how I observe my own and others’ distinguishing features as interpreted by a technology, and how this speaks to larger contexts of how much control individuals have over their own distinguishing features – especially those considered “unconventional” – and perceptions of such. From this research, then, I gleaned two insights. The first is that, as the deepfake created might be considered indistinguishable from myself, that deepfake represents my identity as interpreted through the lens of existing data. The second is that deepfakes, as well as generative AI, draw from data that might disproportionately represent the dominant identities of white, cisgendered, heterosexual, able-bodied individuals, which might be generalized to all populations—including those of individuals with non-dominant identities—and that it is important to ensure that the data that our technologies run on accurately represent the populations that they describe.