This article, written by Professor Michael Mair, Dipanjan Saha, Dr Phillip David Brooker and Dr Terence Heng from the School of Law and Social Justice, was originally published by The Conversation.
Generative AI – which encompasses large language models (LLMs) like ChatGPT but also image and video generators like DALL·E 2 – supercharges what has come to be known as “digital necromancy”, the conjuring of the dead from the digital traces they leave behind.
Debates around digital necromancy were first sparked in the 2010s by advances in video projection (“deep fake” technology) leading to the reanimation of Bruce Lee, Michael Jackson and Tupac Shakur. It also led to posthumous film appearances by Carrie Fisher and Peter Cushing, among others.
Initially the preserve of heavily-resourced film and music production companies, the emergence of generative AI has widened access to the technologies that were used to re-animate these and other stars to everyone.
Even before ChatGPT burst into public consciousness in late 2022, one user had already used OpenAI’s LLM to talk with his dead fiancée based on her texts and emails. Seeing the potential, a series of startups like Here After and Replika have launched drawing on generative AI in order to reanimate loved ones for the bereaved.
This technology, for some, seems to cross a cultural and perhaps even ethical line with many experiencing a deep unease with the idea that we might routinely interact with digital simulations of the dead. The dark magic of AI-assisted necromancy is viewed, as a result, with suspicion.
This may have some people worried.
But as sociologists working on cultural practices of remembrance and commemoration, who have also been experimenting with raising the dead using generative AI, we think there is no cause for concern.
A new dark art or more of the everyday?
Continuing bonds with the dead through text, images and artefacts is commonplace – part of our lives with others both living and dead.
People have long placed emotional value on likenesses and relics as means of keeping the dead with them. While having a portrait painted was no longer a widely adopted way of memorialising images of loved ones at the time, the diffusion of photography in the 19th century quickly became an alternative means for preserving the deceased.
Many of us today have photos and videos of loved ones past, which we return to as memories and consolations. And, of course, famous people’s likeness, works or remains have been circulated to preserve them – often at their behest – for as long as we have recorded history. Religious relics across cultures offer just one case-in-point.
When it comes to generative AI, then, there isn’t anything particularly world defining going on. The speed with which AI’s necromantic possibilities have been exploited tells us a lot about how well the technology works with our existing practices of grieving, remembrance and commemoration – rather than “disrupting” or “changing” them.
But isn’t AI different?
The AI startups in this domain build on earlier do-it-yourself ventures in bringing back loved ones using generative AI. Using writing (for example, on social media and in emails), audio recordings of speech, photographs and videos of loved ones submitted by clients, they train AI models that make it possible to interact posthumously with “them” through images, voice and text.
As noted by Debra Bassett, who has studied digital afterlives extensively, some dissenters to this use of AI have stated that they are worried that the reanimated may be made to say things they wouldn’t when alive and are instead acting out someone else’s script. For Bassett, the concern is the dead are being “zombified” in a violation of their integrity.
This is, of course, a possibility but we should always look at these things on a case-by-case basis. Generally, however, we should remember that we imagine and initiate conversations with the dead all the time.
In moments of crisis or joy, we reflect upon what those we have lost might have said to us, the attitudes they might have had and the encouragement they may have offered in relation to challenges and accomplishments in the here-and-now.
Images, text and artefacts like past possessions or prized heirlooms have long been usable media for that kind of communion and new technologies, most recently cameras and recording devices, have always made such media more easily and widely accessible.
Others, in reflecting on the strangeness of encounters with dead people brought back into digital interaction with us, argue that those communicating are not in fact the dead at all but frauds. Where done exploitatively and in a concealed way, as with the charlatans of the Victorian spiritual revival movement armed with their ouija boards, this is, of course, highly problematic.
However, again, we should remember that we do not ordinarily treat our personal messages, photographs or videos of the dead as if those records themselves were our loved ones. Instead, we use them as conduits to their memory, standing in for them as proxies for us to think of or communicate through. To suggest we routinely get confused or delude ourselves about such media is a misconception.
That’s why general worries about digital necromancy are wildly overblown: when we overly concentrate on their strange and sinister aspects, to adapt the philosopher Ludwig Wittgenstein, we lose sight of the ways in which these new technologies speak to and resonate with what we are and do as human beings already.