Aaron M. Spitler, a researcher in technology and society, discusses the emerging phenomenon of AI griefbots and how policymakers should respond.
For some, grief can feel like a wave, crashing over them as they struggle to catch their breath. What if there was a way to easily stem the tide? This promise has been made by the creators of “griefbots,” referring to products powered by artificial intelligence (“AI”) that can be used to generate digital avatars of the deceased. Large language models (LLMs) analyze troves of data—from social media posts to text messages—to “resurrect” a loved one by mimicking their speech patterns or distinct idiosyncrasies. Some developers believe their solutions offer the bereaved a port in the storm, enabling them to come to terms with their loss. Others have gone further, marketing their griefbots as a way for users to “never say goodbye” to those they cherish. Regardless of their mission, companies hope that emotionally adrift consumers find their products to be the liferaft they never knew they needed.
Critics of griefbots are less convinced they could help those lost at sea. In their eyes, these AI companions pose significant drawbacks, prompting many to question whether they do more harm than good. For instance, the deceased have no say in how they are portrayed by griefbots. Moreover, the fact that these highly personal avatars are often products controlled by companies gives many people pause. Users overwhelmed by loss are also likely to be adversely impacted by the rise of griefbots. These individuals may become attached to digital doppelgangers of the deceased in ways that leave them primed for exploitation. While some developers minimize this concern, many are motivated by factors beyond users’ mental health. Guardrails are needed not only to uphold the rights of those no longer capable of advocating for themselves, but also to protect the wellbeing of others who may be swept away by the allure of griefbots.
Disregarding Individual Consent
In theory, the therapeutic benefits of griefbots are apparent. Users can process their feelings with an avatar of the deceased, providing them with a degree of closure. However, in practice, this application of AI has less obvious downsides. The Guardian noted that the dead themselves are stripped of the autonomy they enjoyed during their lives, as many are transformed into AI-powered caricatures without their consent. While the bereaved may find comfort in seeing a familiar face through griefbots, their loved ones have no means to effectively influence how their likeness is generated. For the dead, the issues with griefbots go beyond a lack of control over their physical appearance. The training data griefbots depend on to create a convincing persona of the deceased may be singularly sensitive. As a result, details about the dead that were never meant to be disclosed may be inadvertently shared in interactions with users. Consumers should bear these concerns in mind, knowing that the “personalized experiences” offered by griefbots can come at the expense of those they have lost.
Questions of ownership define the debate around griefbots in more ways than one. Specifically, the role of companies in bringing individuals back from the dead has been a source of controversy. As AP News reported, numerous AI-focused firms in this growing space have faced financial headwinds, with some going as far as seeking bankruptcy protection. It is common for tech startups in this situation to take steps such as selling off their assets to address their outstanding debts while winding down. Yet floundering companies whose properties primarily include digital recreations of the dead find themselves in a unique predicament. Standards on how to manage these deeply personal resources have yet to be established in the industry. Furthermore, the friends and families of the deceased have no obvious path for recourse when companies close their doors. Half-measures, such as transferring the data used to train and personalize a griefbot, may provide temporary relief to those in mourning. Even so, what may happen to these digital personas when the firms that own them go under remains unclear.
Creating Emotional Dependency
Among the living, griefbots pose a different, but equally alarming, set of challenges. Individuals grappling with their mental health following the loss of a loved one, for example, may be particularly vulnerable to manipulation through griefbots. Nature explored this issue, unpacking how a business model where access to the deceased is mediated through a company has the potential to be problematic. This dynamic can create an uncomfortable scenario in which the bereaved, hoping to restore their connection with the dead, will go to extreme lengths to achieve this goal. This may include taking financial risks to ensure that they can still converse with griefbots well into the future. As a result, users looking for comfort and consolation may mortgage their futures for a chance to relive the past. Firms, for their part, are not necessarily incentivized to prevent these situations from unfolding. Performance indicators, such as continued engagement from the user base, may hold greater weight. While the creators of griefbots may not intend to harm their clientele, they nevertheless profit off their customers’ pain.
When facing this criticism, companies putting griefbots on the market emphasize that the needs of their users come first and foremost. Yet, economic realities tell a different story. The Conversation charged that developers of griefbots, in spite of their high-minded rhetoric, share the same priorities as other entrepreneurs in the tech industry. Like their counterparts elsewhere in the sector, griefbot-focused companies view metrics like user engagement and clientele growth as instrumental to their long-term success. In a competitive field, ensuring that these indicators trend in a positive direction often determines whether a company survives. Given these pressures, founders and executives may come to see the unglamorous work of creating a safety net for at-risk users as a hindrance to progress. The fear of losing ground to rivals, combined with an unrelenting fixation on performance metrics, may also compel griefbot developers to deprioritize user protections. Critically, these “cost-saving” tweaks may directly increase the likelihood of vulnerable individuals suffering psychological harm. While many griefbot developers have altruistic aims, their good intentions are no substitute for robust guardrails.
Changing Course
Griefbots appear to be a novel solution for an age-old problem. Speaking to a digital avatar of a loved one may grant mourning users peace of mind they might not have had otherwise. However, this sales pitch omits several key details. For the deceased, they become commodities enjoyed by the living, a process which deprives them of control over how they are represented. The role of companies in this transformation raises red flags, as the materials the dead leave behind become assets to manage. Additionally, griefbots lose their luster when one thinks about how they impact those in the throes of bereavement. Vulnerable users can grow reliant on these AI-powered products for their mental wellbeing. All the while, firms releasing griefbots onto the market may not properly invest in guardrails, leaving these customers in the lurch. Considering these consequences, would-be consumers may find these solutions’ cons outweigh their pros.
As griefbots become mainstream, policymakers must take action. They may find inspiration looking to property law when crafting regulations for this technology. For example, adopting an expanded definition of what constitutes a personal asset may help them mitigate the damage caused by griefbots. Digital assets, such as social media posts, could be afforded the same protections as physical belongings in legal wills. Following this approach could ensure the deceased, and their loved ones, can rest easy knowing that their image will be neither warped nor distorted. It could also keep “performance-centric” developers in check, preventing them from creating avatars that prey on the vulnerabilities of users seeking comfort. This policy framework may prove to be an invaluable model for regulatory officials attempting to navigate this emerging field. For the bereaved, this strategy could empower them to resist the appeal of griefbots, allowing them to pursue the support they truly need.
