Wearing a knitted cap and facing a laptop screen, Ri Akin types into a chat box: “Hello, how is your day?”
“Hello, my dear friend,” responds the voice of a woman speaking with a familiar British accent. “My day’s been filled with meetings and events focused on the various causes that are close to my heart.”
But the UC Davis design major and the person they’re chatting with are not friends. The woman isn’t real.
On the screen, Akin sees a realistic caricature of a woman with feathery blonde hair framing her face, blue eyes and blushed cheeks. The image shares a stark resemblance to the late Princess Diana trapped in time, free of age marks and laugh lines.
The Princess of Wales died in 1997 at age 36.
This artificial presence is one of 77 famous figures whose digital doppelganger is available to chat with on Eternal AI's website.
Eternal AI is one of many companies offering ways to feel like you're interacting with the dead. Some of these services are billed as ways to help those mourning the loss of loved ones by providing a simulation of the person via avatars, chatbots and voice simulations. While some data is uploaded — and permission given — before death, other data is uploaded by those left behind. Eternal AI offers consumers the option to create a "digital legacy," which includes a lifelike avatar, voice cloning and the "preservation of persona."
Students in Professor of Art Darrin Martin’s Experimental Documentary course last year researched this type of technology and experimented with different tools available today. Their work culminated in the short film Ghost AI, premiered during the Film Festival at UC Davis in May. The film offers a chilling exploration of issues of consent and ownership when it comes to artificial intelligence (AI), raising many ethical questions around the use of AI in art and culture as well as in our personal lives — even after we’re gone.
“Ghost AI is next level regarding trying to have a conversation with an entity that tries to emulate that person, and the complexity of permissions,” Martin said. “They were really clever with their editing and thinking about celebrity and questions around those permissions.”
“I was pretty intrigued by the idea of AI and immortalization of a person,” said Raul Huerta, a recent cinema and digital media graduate. “You can see that now in Hollywood with the deep fakes, using the likeness of deceased actors, but now we’re seeing it in regular peoples’ lives.”
Thinking critically and creatively on a deadline
For a class project assigned by Martin, Huerta and Akin decided to delve more into the topic. With just 10 weeks in a quarter, however, researching, scripting, filming and editing a five-minute experimental documentary with original footage wasn’t part of the assignment. Instead, Martin encouraged students to combine their own shot materials with found footage and materials generated by AI tools in order to think critically about the technology.
“There were a lot of really different responses to the prompt,” Martin said.
For example, some students produced their films using ChatGPT to write a script, which they then edited, many keeping some of the most nonsensical aspects provided by the AI.
“They were recontextualizing what AI wrote for them to make it make more sense but still embracing some of the more hallucinatory artificial intelligence components,” Martin said. “They definitely embraced some sci-fi tropes around AI — Ghost AI not being an exception — although this stuff is not science fiction anymore."
Huerta and Akin decided to use screen recordings of their interactions with Eternal AI’s avatars in order to explore the potential and limits of the technology as well as the transparency of the technology. Inspired by Ai Weiwei’s work Ai vs. AI, in which the artist poses 81 questions to an AI chatbot, Raul and Akin decided to pose questions to the celebrity chatbots to see what their programming would have them say about their lives, the use of their likenesses and the ethics of this type of AI technology.
They interviewed avatars of Princess Diana, Heath Ledger, Kobe Bryant, Juice Wrld and John Lennon. Posing the same questions to each deceased icon, the duo compared the answers and, during the editing process, spliced them together seamlessly.
“It was kind of strange when we put all the responses together and realized how close they were,” Huerta said.
For example, Akin asks: “How am I able to hear your voice?”
Voices sounding similar to Bryant, Juice Wrld and Lennon continue one another’s sentences: “The voice you are hearing is generated by a sophisticated language AI model. The model has been trained on a vast amount of text data that is designed to mimic my voice based on available recordings and data, allowing you to hear it even though I am no longer alive.”
The simulations couldn’t provide answers to some questions, including those about what data models were used in their creation.
“Often we found that they will dodge the question,” Huerta said.
Who own’s this face? Asking the AI about consent.
In the short film, Akin eventually asks the AI, “How does consent apply in AI replication?”
“Consent is an important aspect to consider when it comes to AI replicating or using someone’s identity,” we hear the Diana voice say, followed by imitations of JuiceWrld and Ledger: “When it comes to creating a digital representation of a person, whether it’s through their voice, image or persona, obtaining explicit consent from the individual themselves or, if not possible, from their legal representatives or designated individuals is of utmost importance.”
“In the case of deceased individuals, consent may come from their estate, family members or individuals with legal authority to make decisions on their behalf,” the Lennon-ish voice continues.
Despite these acknowledgements, the only avatar featured in Ghost AI who said their family gave consent was Ledger’s.
Eternal AI states its mission is to “breathe digital life into the legends of the past, empowering people to explore, learn, and engage with iconic figures.” The company also warns, “While our digital heroes may appear incredibly lifelike, always remember that they're not real.”
In an email with the College of Letters and Science, representatives with the company said their offerings are in legal compliance "given these figures are in the public domain, and we explicitly state these are AI representations."
These simulations could be seen as transformative interpretations of publicly available information. In a way, a similar idea would be if a cartoonist sees one of us on a public street, there's nothing we can do to stop them from drawing an unflattering portrait of us and selling it later that day.
Are they copies or interpretations? Does it make a difference if AI is creating the art instead of a human hand? Is this ethical? That is still up for interpretation.
“I believe that consent is a very important thing when using or replicating one’s likeness,” Huerta added. “I think it’s something we should all be thinking about before any AI company just jumps right in to try to experiment with these AI replications."