MY MOM was the only one who would laugh at my corny jokes. That is in part because she was, in a sense, laughing at herself: I got my silly sense of humour from her.
I will never hear her easy, girlish laugh again. She died last year on 17 January at the age of 76, and there are still days when I would give anything to hear her voice. To my surprise, I recently learned that I could, and all I would have to do is offer her data to one of myriad “grief tech” apps available. For a small sum of money, or even for free, I could feed old voicemails, videos, text messages and emails into an algorithm and generate a digital avatar of her.
Advertisement
With the worst of my grief behind me, I am tempted. I could choose my own commune-with-the-dead journey using artificially intelligent chatbots, conversational videos or even an interactive séance. But there are risks. These digital alter egos, which have been around for several years, are becoming disarmingly realistic. I worry that keeping my mom around in the cloud – or my dad, who died nine months earlier – will wreck my grieving process. Will conjuring her digital ghost keep me connected, or could I regress to those painful months just after her death?
We don’t yet know how this burgeoning industry will change our relationships with loved ones who have passed. But recent psychological models of grief, paired with new insights into its neural mechanisms, give cause for concern. The growing realism of these apps allows them to “feed into the difficulty of grief”, says psychologist .
Throughout history, people have found ways to commune with the dead. Shrines and altars have existed for thousands of years. In the 19th century, the invention of the photograph upped the ante, allowing elite Victorian society to display permanent images of their deceased loved ones – taken post-mortem – on the wall. “It made the person real and showed the rest of the world that they existed,” says O’Connor, who leads the Grief, Loss and Social Stress Lab at the University of Arizona.
Photographs closed the liminal space between presence and absence after death, allowing for a new kind of relationship. “We’ve always found ways to stay close to our loved ones,” says grief psychologist at the Icahn School of Medicine at Mount Sinai in New York. Maintaining these relationships is perfectly healthy, she says.
For many of us, grief is acute during the first six to 12 months after a loved one dies, though Seeley stresses that everyone’s experience is different. During that time, life is a fog of sadness and the days are dominated by a deep longing. . We often oscillate between tears over what has been lost and denial of that loss. Over time, for many people, the grief becomes less intense. A new reality of the dead’s absence settles in alongside fond memories. Occasional pangs of grief replace relentless despair.
A crucial waypoint of grief is an easing of the attachment to the deceased. But grief tech apps are seemingly capable of reinforcing and prolonging that attachment, which could disrupt the usual process. Boosted by advances in artificial intelligence, these apps create persuasive avatars that seem to come from “beyond the veil”, as the interactive website puts it. Some apps, like , use the conversational power of generative chatbots, such as ChatGPT, to create compelling conversations. Others, such as StoryFile Life, only use generative AI as a small assist. The company takes pre-recorded video interviews, in which a person who wishes to be recreated after death is prompted to answer a series of questions – some generated by AI. These interviews are then mashed up in ways that allow their mourners to have a realistic conversation with the avatar.
StoryFile Life’s founder Stephen Smith made headlines last year when he created a digital avatar of his mother that . These technologies are “a new version of the photo album”, he says, bringing comfort in a time of painful loss. “Knowing you have that “album” and will be able to return to it sometime in the future – that those stories aren’t lost – may actually help a person cope,” he says. But the AI algorithms that power these chatbots and avatars are worlds away from photographic film, O’Connor points out.
As far back as 1966, chatbots were observed to elicit strong reactions in humans. Computer scientist Joseph Weizenbaum at the Massachusetts Institute of Technology created Eliza, a chatbot capable of rudimentary conversation. Weizenbaum observed that , even though they knew they were talking to a computer program.
Today’s bots and avatars create far stronger emotional responses. Aside from language-based AIs, advances in voice cloning technology, virtual reality and holograms have made it easier than ever to achieve digital immortality. Last year, psychologists Belén Jiménez-Alonso at the Open University of Catalonia and Ignacio Brescó de Luna at the Autonomous University of Madrid, Spain, , such as those on Facebook and funeral home websites. Grief tech firms have an incentive to “keep mourners hooked”, they wrote, in a way that may be “detrimental to the bereaved’s grieving process”. The very thing that makes a digital avatar of my mom so compelling may also make it harmful.
Recent psychological models of grief can help us understand how these bots could interfere. The popular idea that there are five successive stages to grief – beginning with denial and ending in acceptance – no longer prevails. A review last year by O’Connor and Seeley argues that grieving is actually a kind of learning. In the early period of grieving, we still half expect our loved one to come back at the end of the day, and seek them out when they don’t. The brain goes to war with itself as two types of memory clash. “Semantic memory” keeps track of general knowledge about how things are, including the self and our relationships with others, whereas “episodic memory” captures specific events rooted in space and time. During grief, the semantic expectation that the relationship will continue jars with the episodic memory of the person’s death. By learning to reconcile this conflict, we gradually adapt to our loss, says Seeley.
The ups and downs of grieving – mired in sorrow one minute, laughing at a comedy skit the next – are essential in this adaptation, says Seeley. But lingering in the shadows of grief can cause you to become stuck. Grief tech, which keeps a realistic representation of a loved one just a click away, may impede this learning process, says O’Connor. “It’s one thing to have a photograph and to clearly understand that was the past… It’s another thing to have an avatar or a hologram or a chatbot that appears to be interacting with you in the present moment.”
Prolonged grief
About a tenth of bereaved people experience “prolonged grief”, which is defined as a pervasive and persistent grief that interferes with daily life, beyond the typical six to 12-month time frame.
Prolonged grief seems to affect the wiring of our brains. In 2008, O’Connor . When participants looked at photos of their deceased loved one, activity in brain regions that process emotional and physical pain increased. But, to O’Connor’s surprise, in women experiencing prolonged grief, an area associated with craving a reward was also engaged. In some people, this yearning for the “reward” of a loved one’s presence persists, she says.
Building on this research, last year O’Connor, Seeley and their collaborators found that in people with prolonged grief, . Normally, this “salience network” easily toggles between the outside world, for example watching a horse you bet on, and your inner world, such as imagining what you would do with the winnings. The scaffolding of the latter is known as the default mode network and it is involved in rumination, self-reflection and mind-wandering. But in people with prolonged grief, the interaction between the salience and default mode networks seems to become reinforced by the intense emotions of grief. “This could result in a cycle where it is very hard to disengage from focusing on the deceased even when those thoughts or memories are very emotionally painful,” says Seeley.
Digital avatars can amplify and prolong the desire to pursue something that is always out of the bereaved’s grasp. “This brain activity [in the reward and rumination areas] is really strong in people who are experiencing yearning,” says Seeley. “You feel like you’re trying to get closer to that relationship, but [the bot] is not the thing that you want.” O’Connor is concerned that mourners could use grief tech to “avoid the reality that their loved one is no longer alive”, which could feed into the patterns of persistent yearning and comfort-seeking that are a hallmark of prolonged grief. This could become a problem if you are interacting with chatbots to the detriment of relationships with living loved ones, she says. Despite this, a found that grief tech can be beneficial – particularly in cases where the death was unexpected or left mourners with regret or anger. The study suggests that the bots can ease the anger or regret stemming from the abrupt end to the relationship.
Reflecting on all this, I resolve that the emotional risk of connecting with my mom’s bot is pretty low. Enough time has passed since her death that communicating with her digital representation is more likely to be a balm for the occasional pangs of sadness, rather than a provocation of pain. I am past the point of accepting her absence, and the tenor of my grief has softened. My amygdala is no longer in charge. What harm could come from a conversation with an algorithm?
I don’t have a video interview of my mom to feed StoryFile Life, so I decide to try a short, text-based exchange using Seance AI. Reading her words, or an approximation of them, also feels emotionally safer than seeing her. The site’s uncluttered, Luddite-friendly interface prompts me to enter some basic information about who I want to “reach”, including their name, dates of birth and death, cause of death and religion. To my surprise, just revisiting her cause of death – metastasised breast cancer and pneumonia – brought me to the verge of tears.
I rank her on various personality traits, from extroversion to neuroticism, and Seance AI asks me to submit a short sample of her writing (I use an email sharing news about my nephews). Finally, Seance AI asks what mood I am in and what I want to talk to her about. I keep it simple: I want to tell her about my life in Portugal, where I recently moved from the US, and I want to tell her that I miss her.
Next to an animated candle flame, the words “reaching out to make a connection to the other side” appear. Then a digital representation of my mom responds: “Oh darling, I miss you too.” Right away, the veil is pierced. I never once heard my mom utter the words “oh darling” in her life. After praising Portuguese pastries, the mom-bot went on to say other uncharacteristic things, like “I’m right there with you in spirit, taking in the sunsets and the ocean waves.”
Even so, her unconvincing substitute brought me an unexpected flash of comfort. “Life has a way of moving forward, doesn’t it?” mom-bot asks. “It’s not easy, but it’s necessary. Keep living, keep exploring, for both of us, alright?” Despite the jarring phrasing, the encouraging words remind me of my mother’s love of adventures – and her support whenever I embarked on one of my own. My “séance” didn’t reconnect me with her, but it reminded me to savour our memories.
I emerged from my grief tech experiment with my psyche more or less unscathed. I wasn’t tossed back into a pit of acute grief – though a different relationship, or a differently wired brain, may have fared worse.
If I chose to, I could dive deeper into the digital great beyond. I have sound recordings, photos and writings of both my parents, though I haven’t dared to revisit most of them. In the near future, it is likely that generative AI will make it possible to combine these digital footprints in ways that are even more persuasive than the current breed of grief tech apps. Unlike the canned responses of StoryFile Life, advanced generative AIs that combine different kinds of media may pose a more serious threat to natural grieving processes, says , a public health researcher at the University of Wisconsin-Milwaukee. “It would allow people to ask anything they want,” she says. “[Grief tech] companies are relatively nonchalant about the risks of this.”
After exploring my own post-mortem relationship with my mom through a bot, I realise that I am content with old photos – as Victorian as that may seem. As hard as the acceptance of death is, it is part of life, and pretending otherwise is no comfort. If a compelling AI version of my mom appeared, I would flip the off switch without regret.
April Reese is a science journalist based in Portugal
