The genealogy website, MyHeritage.com, has introduced a service allowing users to upload photos and have them animated using artificial intelligence. The service intends to offer glimpses of the natural facial movements of deceased relatives using “deepfake” technology.
Deepfake – a portmanteau of “deep learning” and “fake” – technology is a form of synthetic media whereby an existing image or video of an individual is substituted with another person’s likeness. The technology relies heavily on powerful algorithms that extrapolate novel but expected information from known data. The data are often called “training data” and prime the algorithms on likely consequences to make predictions or decisions and then “learn” from those predictions. The algorithm’s processes and optimization are termed Machine Learning (ML).
ML and artificial intelligence combine in the deepfake technology to create realistic-looking videos or images, making it exceptionally difficult to tease out real images from fake ones. Recently, a TikTok account called @deeptomcruise produced a set of viral videos showing a deepfake Tom Cruise performing simple tricks and being silly.
@deeptomcruise I love magic!
Deepfake technology has been used in the film industry for the past few years and rumors abound regarding the use of deepfake technology for future films. Disney, for example, is reported as experimenting with the technology to bring back deceased actors for new films. Disney Research Studios in Zürich “focuses on exploring the scientific frontiers in a variety of domains in service to the technical and creative filmmaking process.”
In a recent research paper, Disney Research proposed an algorithm for high-resolution lifelike face-swapping that Isis the “first method capable of rendering photo-realistic and temporally coherent results at megapixel resolution.”
The result would produce state-of-the-science images that are indistinguishable from images captured from actual people. Disney shared a video describing the research.
The use of deepfake technology has also been sinister. Deepfakes have been used in financial fraud, hoaxes, fake news, and most disturbingly in “revenge porn,” abusive non-consensual pornography that is a form of image-based sexual abuse. Women are often the usual targets.
Fortunately, this form of abuse has been outlawed in various jurisdictions including Australia, Canada, Germany, Italy, Israel, the United Kingdom. Most states within the U.S. and the District of Columbia have legislation either in place or pending.
The abusive, unethical and even illegal applications of deepfake raises a set of similar questions with its application to the photos of ancestors animated with similar technology.
First, there is the fundamental question of consent. Did an ancestor who posed for a photo consent to the use of that photo in any future manner besides a photo? It seems unlikely they did. They likely posed for a photo as a keepsake for themselves or their immediate family. An ancestor may also have regarded that photo as a memento to the future. Some may have posed for news media or collections, but still as photos.
It is also likely that the older the photograph the more serious the act of taking a photo was. That is, in the earliest days of photography, the subject of a photo dedicated significantly more time and attention to the act of being a photographic subject than later dates when photography became increasingly commonplace.
But it is unlikely that ancestors envisioned a photo being the basis of an animated film or a gif. There would be no basis for an ancestor to understand these types of technological possibilities let alone deepfakes, especially if the image was gathered before the development of animation. Without that basis, it becomes difficult to navigate how consent would be possible.
Second, there is a question of accuracy. Honoring an ancestor and their memory presumes an intent to be accurate about the memory. When we discuss ancestors with family, we convey history that may be colorfully enhanced at times, is still intended to bear a sense of truth.
The Machine Language that builds the deepfake reanimation of photos makes no such promise. The data are extrapolations of expected movements based on the general population and the photo’s characteristics. If Aunt Agatha always winked as she smiled, the deepfake reanimation will not produce those idiosyncrasies without specific information about them. The current state of the accessible version programs does not allow for a user to provide that information, so the animation is, at best, an echo of what a machine wants Aunt Agatha to act like- generic, statistically likely movements based on the general population.
Deepfake technology still portends even more. The simple acts of blinking and smirking are only the beginning. Deepfake and deep-learning technologies promise that audio estimations of voices will also be possible, followed by the full animation of the body, not just facial expressions.
The ultimate question posed by reanimating ancestors using deepfake technology is, “who owns the face?” We may physically own a photo, but do we have a right to manipulate the face of an ancestor as we would choose?
While images can enter the public domain and, even before, individuals may not have absolute ownership of their faces under the law, one’s right to privacy is generally violated when one’s voice, photographs, or likeness is reproduced without obvious consent. Ethics and law do not always align, and the deceased are often not acknowledged to have equivalent rights as the living, under many ethical and legal approaches to the rights of Ancestors.
Pagan perspectives are different, however. Especially for those Pagan traditions that include Ancestor veneration, the deepfake reanimation of ancestors seems misaligned, if not deeply incongruent with the act of reverence. Deepfakes seem to violate their privacy and agency.
Moreover, there are examples of how Ancestors might convey consent to future generations. In some traditions, this might include divination before using deepfake technology, but ancestors may also leave explicit consent.
Majel Barrett, who played Nurse Christine Chapel and Lwaxana Troi on Star Trek and was the wife of the series creator, Gene Roddenberry, also voiced the computer on the series after it was given. Before she died in 2008, Majel Barrett recorded a complete voice library that included phonetic sounds for future usage. Barrett intentionally offered her voice to future generations to voice the Star Trek computer.
While deepfake technology may have artistic uses, its history and commonplace applications have been mostly trivial or sinister. Adding to that the Pagan respect for consent and the reverence of Ancestors in many traditions, the five-second best-guess glimpse of an ancestor’s blink seems ethically questionable and spiritually clumsy.
The Wild Hunt is not responsible for links to external content.
To join a conversation on this post:
Visit our The Wild Hunt subreddit! Point your favorite browser to https://www.reddit.com/r/The_Wild_Hunt_News/, then click “JOIN”. Make sure to click the bell, too, to be notified of new articles posted to our subreddit.