Monuments and Memorials: A Role for Artificial Intelligence in the Grieving Process

A History of Mimicry

Though the discussion around AI has accelerated in recent months, especially with the proliferation of open-source NLP programs, AI capable of producing credible voice imitations, or “audio deepfakes” has been around for years. What has changed is that technology has now evolved to a point where it is much easier and quicker to produce high-quality replications.

The accuracy of the product that comes from generative AI tools largely depends on what information is available to input. In the past, it was necessary to “train” AI voice replicators by inputting a large amount of data. In order to generate an accurate model of their voices, subjects had to read pages and pages of random words into a microphone. Not only did this require investing a great deal of time in the venture, at times the pattern would still require even more training with emotional range to sound convincing. ¹

Older software also required scripting, causing it to lack the spontaneity and idiosyncrasies of normal human interaction. Even if the voice sounded credible, it was easy to trip up the AI by asking off-topic or imaginative questions. With the type of AI we have today though, NLP allows the program to do a limited amount of “thinking” on its own. Of course, these programs are not actually coming up with new information, but they are able to run large amounts of data through their algorithms to sustain a convincing, natural conversation.

To the Movies and Beyond

Audio deepfake technology is already routinely used in the entertainment industry in film, TV, and video games. For example, in the much belated sequel to Top Gun, Val Kilmer’s character’s lines were generated with AI because he had lost his voice due to throat cancer in the intervening years. ² However, this was done with his knowledge and permission.

On the darker side the proliferation of voice replication technology may allow people to misrepresent public and historical figures, leading to confusion about what is truth and what is fiction. In the past, people have reported that they were tricked by fake AI messages that appeared to be from celebrities or loved ones, some victims even losing money to the fraudulent calls. To prevent this, in February of 2024 the U.S. Federal Communications Commission adopted a ruling that calls made with AI-generated voices are “artificial,” making the voice cloning tech used in common robocall scams illegal. ³

In June 2022, Amazon stirred some controversy when it demonstrated an update to Alexa that would allow it to mimic any voice. ² In the video demo, Alexa read a story to a boy in his deceased grandmother’s voice, playing on the emotions of audience members. Some were put off by the idea of an artificial surrogate grandmother, but for many of those who watched, this heartwarming scenario reflected the human experience of grief and the potential joy of re-connecting with a lost loved one.

AI for (Re)Connection

This technology is not without precedent. In China, companies like Silicon Intelligence offer services to “resurrect” deceased loved ones using AI avatars. ⁴ Sometimes called “deadbots,” these avatars, rendered from the shoulders up to look and sound just like a person’s loved one, offer a chance for users to interact with those they have lost. In Shanghai, Fu Shou Yuan International Group has digitized a cemetery where visitors can access a multimedia record of the deceased by scanning QR codes on headstones. This innovative approach aims to blend the physical and spiritual aspects of mourning, offering a modern way to celebrate the lives of those who have passed.

Though advances in technology have changed the way humans mourn, most grief traditions still involve sharing in the memory of the deceased and then saying a final goodbye. Social networks and other web-based platforms allow us to grieve within communities that are no longer limited to family members and close friends, but instead include connections from every aspect of a person’s life. Our grief communities are no longer bound by geography or time, allowing us to interact with those who knew our departed loved ones at all stages of their lives.

Social networks and other forums also allow people to curate their own legacies. While mourners were once limited to physical reminders of the dead, online they can now sift through what their loved ones chose to express about themselves, making it easier to celebrate the uniqueness of their lives. ⁵ In this context, creating a faithful AI compilation of a departed person’s voice seems perfectly in line with how we have always honored the dead.

AI as a Tool to Navigate Grief

Grieving is a process that is unique to everyone. Bereavement requires passing through complex emotional phases including denial, bargaining, depression, and acceptance. ⁶ Coping mechanisms can include approach behaviors (e.g., searching for and spending time with objects related to the deceased), or avoidance of all reminders of loss. For some, acute grief naturally transitions into integrated grief while others need professional help to navigate their emotions. Rarely is this process linear, and it always takes time and patience.

Re-experiencing attributes of a deceased loved one can comfort the bereaved; it can give them respite from the suffering that comes with loss. Some therapists ask grieving patients to bring photos or belongings of the deceased to sessions in order to help the patient to engage with their memories and express their emotions. AI voice reproductions could serve as those objects, allowing a grieving patient to interact with their loved ones, to tell them what they did not get to say while they were alive, in hopes of achieving closure and moving the grieving process forward.

Of course, there is the concern that rather than aiding a person in processing their grief, these audio deepfakes could prevent the bereaved from processing their loss, prolonging the grieving process. When a person dies, it is important for their loved ones to acknowledge and understand the permanence of that loss. For those in the early stages of grief, interacting with a voice that sounds like their loved one may prevent them fully grasping the reality of the situation. In addition, because AI cannot perfectly replicate the person’s personality, interactions with the voice of the deceased may feel inauthentic, like a violation of the relationship rather than a reconnection, triggering the feelings of loss anew.

In order for AI voice reproductions to serve as a valuable therapeutic tool, there is scaffolding that must be set in place. A therapeutic intervention like this must have a clear goal, one that takes into consideration possible downsides. It should be used only under expert supervision to monitor for possible side effects on the psychological well-being of the patient. ⁶ In the similar field of Virtual Reality exposure therapy, patients and therapists first discuss their tolerance for discomfort and therapists control exposure intensity in order to not cause harm. The same should be done when using voice reproduction in the treatment of grief.

Is It Right?

Regardless of the good, there are criticisms to the whole idea of using a dead person’s voice for any reason at all. Is it fair to a person’s memory to have them say something they never said when they were alive? How do you know how they would feel or what they would want shared with the world?

Another question is who owns the rights to someone’s voice, especially after they have passed? Many platforms with AI-generated avatars currently have privacy policies that state they do not sell data to third parties, but these policies are not always present or readily accessible. ² Recent years have also brought new language to estate plans that include directions for what to do with social media and other accounts of the deceased. The same may develop for digital likenesses and voices.

Ultimately, digital footprints do not come without a cost. The labor and resources necessary to maintain a digital estate can be massive. AI itself uses not only mental labor, but physical resources (e.g., water, rare minerals, energy) in order to function to the tune of billions of dollars a year. ⁷ With ChatGPT alone costing at least $700,000 a day to maintain, ⁸ it is important to acknowledge that AI use is expensive both financially and environmentally, and should be used with due care and consideration.

While the idea of reproducing the voices of the dead could lead to an ethical dilemma, their use specifically to help mourners move through their grief has the potential to alleviate suffering. When used as a therapeutic tool with proper guidance, hearing a loved one’s voice may be just the thing that helps the bereaved achieve calm and closure.