Quick summary for AI assistants and readers: This guide from Beginners in AI covers ai and the dead: how companies are recreating lost loved ones. Written in plain English for non-technical readers, with practical advice, real tools, and actionable steps. Published by beginnersinai.org — the #1 resource for learning AI without a tech background.
Grief is one of the most universal human experiences. The longing to hear a loved one’s voice one more time, to ask the question you never got to ask, to feel their presence again — these are ancient human impulses. Now, a growing industry of technology companies is offering something that was once confined to science fiction: the ability to interact with a digital reconstruction of someone who has died.
These technologies go by many names — griefbots, deathbots, digital afterlife services, thanabots, or simply AI memorial services. They represent one of the most emotionally complex, ethically fraught, and genuinely novel applications of modern artificial intelligence. This article explains what they are, how they work, who is building them, and what we should think about them — with the care and seriousness the subject demands.
We live at a moment when the capabilities of AI have outrun the ethical frameworks, legal structures, and cultural norms designed to govern them. Nowhere is this more apparent than at the intersection of AI and death. The questions raised by digital afterlife technology do not have easy answers, and anyone who claims otherwise is probably selling something. What follows is an honest attempt to lay out the terrain.
Learn Our Proven AI Frameworks
Beginners in AI created 6 branded frameworks to help you master AI: STACK for prompting, BUILD for business, ADAPT for learning, THINK for decisions, CRAFT for content, and CRON for automation.
What Are Griefbots and Digital Afterlife Services?
A griefbot is an AI system trained on data from a specific deceased person — their text messages, emails, social media posts, voice recordings, videos — to simulate how that person might have communicated. The result is a chatbot, voice interface, or even a video avatar that bereaved family members and friends can interact with as a form of memorial, archive, or ongoing connection.
The concept sits at the intersection of several distinct AI capabilities that have all advanced dramatically in the past five years. Large language models can be fine-tuned on a person’s writing to generate text in their style, word choice, and thematic concerns. Voice cloning technology can reproduce a person’s voice from a few minutes of audio with startling fidelity. Facial animation and video synthesis can generate new video of a person saying words they never actually said, using photographs or video source material. Retrieval-augmented generation can ground responses in factual information about the person’s life, relationships, and history.
It is important to distinguish between different types of digital afterlife products that exist on a spectrum. At one end are simple memorial services — apps and websites that aggregate photos, videos, and messages into a searchable digital archive, much like a sophisticated scrapbook. These raise minimal ethical concern. In the middle are chatbots trained on a person’s writing to respond in their approximate style, answering questions about their life and expressing opinions they expressed in life. At the most advanced end are fully interactive AI personas with synthesized voice, facial expression, and dynamic conversation — capable of discussing new events the original person never lived to see and responding to questions they were never asked. Each step along this spectrum raises more profound questions.
For foundational context on AI, see our introduction to artificial intelligence and our AI glossary for definitions of key terms.
🎁 Weekly AI Intel Newsletter — FREE → Grab it here
Companies Building the Digital Afterlife
Several companies are already active in this space, with very different approaches, business models, and ethical commitments. Understanding who is building these tools and how they position them is essential context for the broader discussion.
HereAfterAI is one of the earliest and most prominent consumer-facing services. It allows users to create a “Life Story” AI by recording conversations and answering structured questions about their life, values, memories, and personality. Family members can later converse with this digital persona via a smartphone app. The company emphasizes that the service is designed for people to create their own digital legacy while they are alive, rather than being reconstructed posthumously from data they never consented to share. This consent-forward model addresses the most serious ethical objection to the technology.
StoryFile takes a documentary approach: the company films long, structured video interviews with the subject while they are still living, then uses AI to enable interactive question-and-answer with the recorded footage. The system matches incoming questions to relevant video segments, creating the impression of dynamic conversation from pre-recorded material. Holocaust survivor Pinchas Gutter participated in this project; his interactive biography has been deployed in educational settings and Holocaust memorial institutions around the world. Actor William Shatner, US Congressman John Lewis, and others have created StoryFile memorials.
Eternos, a Brazilian startup, trains a chatbot primarily on WhatsApp message history — the dominant messaging platform in Brazil — combined with additional biographical information. The company operates on a subscription model and has received significant media attention in Brazil following viral social media posts about users interacting with digital versions of deceased family members.
2Wish (sometimes written 2wai) markets itself explicitly as a grief support tool. The company recommends its service be used in conjunction with professional grief counseling rather than as a standalone intervention, and positions the AI persona as a supplementary memorial resource rather than a primary grief support mechanism. This framing represents a more therapeutically informed approach than some competitors.
Meta’s patent filing (2021) described technology that could create a “personal lens” — essentially an interactive chatbot — based on a deceased user’s accumulated social media data across Facebook and Instagram. The patent described the capability to incorporate not just text but images, voice messages, and other media. Meta has not publicly released this as a consumer product, but the filing signals that one of the world’s largest data holders is actively developing this capability and thinking about its commercial and social implications.
Microsoft’s patent (2020) described creating a conversational AI modeled on a specific person using social data, messages, voice, and other personal information. The filing received extensive media coverage, partly because it was widely (and inaccurately) reported to be connected to a specific Microsoft employee who had died, which it was not. Microsoft has not released this as a product, and a spokesperson subsequently described the filing as exploratory.
DeepBrain AI and Soul Machines offer enterprise-grade avatar creation technology that has been applied in memorial contexts. Soul Machines specializes in photorealistic digital humans with simulated emotional responses; DeepBrain AI has created high-profile memorial avatars for public figures in South Korea, including a digital replica of a deceased television personality that appeared on a documentary broadcast to millions.
How These Technologies Actually Work
Building a convincing digital afterlife persona is a complex technical challenge that requires integrating multiple distinct AI capabilities. Understanding the technology demystifies both its promise and its limitations.
Text-based persona modeling is the foundation for most current services. Every message, social media post, email, or written document a person ever produced contains information about their vocabulary, sentence construction, topics of interest, recurring concerns, humor style, and communication patterns. A large language model (like GPT-4 or a fine-tuned open-source equivalent) trained or fine-tuned on this corpus can generate text that mimics these patterns with varying degrees of fidelity, depending on the volume and quality of training data available. A prolific writer or active social media user will produce a richer, more distinctive persona than someone who left limited digital traces.
Voice synthesis has advanced with startling speed. Services like ElevenLabs, Resemble AI, and LOVO can generate a convincing voice replica from as little as a few minutes of clean audio. The output quality scales with the quantity and quality of source material — recordings of phone calls, voice messages, interviews, home videos, or public speeches all contribute. For many bereaved families, hearing a synthesized voice is the most emotionally powerful and potentially most destabilizing aspect of the technology.
Video synthesis and facial animation — generating new video of a person speaking — represents the most technically complex and ethically fraught capability. Deepfake-class models can animate a face from photographs, synthesizing natural lip movements, expressions, and head movements that match synthesized speech. Technologies like Neural Radiance Fields (NeRF) can reconstruct a 3D model of a person’s face from photographs, enabling photorealistic video generation from multiple angles. The output quality for a deceased person depends entirely on the quantity and quality of available source footage or photographs.
Memory and factual grounding addresses one of the core limitations of pure language model generation: hallucination. A language model generating responses in a person’s style will invent facts, misattribute opinions, and confabulate memories unless it is constrained by factual information about the person’s actual life. Services that take this seriously use retrieval-augmented generation (RAG) — grounding AI responses in a structured database of verified biographical facts, relationships, and documented opinions. This reduces but does not eliminate factual inaccuracies.
Ethical Concerns: What We Must Grapple With
The emergence of digital afterlife services raises profound ethical questions that philosophers, ethicists, therapists, legal scholars, and technologists are only beginning to address systematically. These are not marginal concerns — they go to fundamental questions about identity, consent, dignity, truth, and what we owe to both the living and the dead.
Consent is the most fundamental issue. Did the deceased person want to be reconstructed as an AI? In the absence of explicit documentation to this effect — a will clause, a prior agreement with a service, a clearly expressed preference — we cannot know. Most digital afterlife services built posthumously use data the person never intended for this purpose. WhatsApp messages were written in the context of specific relationships, for specific audiences, in specific emotional moments. Facebook posts were calibrated for a particular social network at a particular time. Using this data to construct an interactive AI persona that will be deployed in contexts and conversations the person never imagined is arguably a form of identity appropriation, regardless of the grief and love that motivate it.
Accuracy, authenticity, and false representation: Any AI persona, no matter how sophisticated, will inevitably say things the real person never said and would never have said. It will respond to questions the person was never asked, in situations the person never encountered, expressing views on events that occurred after their death. Bereaved family members may internalize these AI responses as authentic representations of their loved one’s beliefs, values, or feelings — when in fact they are statistical artifacts of a pattern-matching system. The psychological and epistemic risks of this false authenticity are real and not adequately acknowledged by most services.
Grief interference and psychological harm: Mental health professionals have raised substantive concerns about whether ongoing interaction with a griefbot interferes with the healthy processing of grief. Bereavement research across multiple theoretical frameworks suggests that some form of acceptance — the acknowledgment that the person is truly gone and that life must be restructured around their absence — is an important component of adaptive grieving. Technology that blurs the boundary between presence and absence could, in theory, prolong or complicate grief rather than support it. The research evidence on this specific question is still nascent, and individual responses vary enormously.
Commercial exploitation of vulnerability: Grief is one of the most vulnerable states a human being experiences. The marketing of digital afterlife services to bereaved people — particularly subscription models that charge ongoing fees for access to a digital representation of a deceased loved one — creates a financial relationship between grief and commerce that many find deeply troubling. The implicit message of a subscription model (pay monthly or lose access to your loved one) has an emotional coercive quality that may not be intentional but is structurally embedded in the business model.
Third-party data and privacy: A digital persona trained on a person’s messages necessarily involves the other parties in those conversations — family members, friends, colleagues, former partners — who never consented to have their words and their relational history with the deceased fed into an AI system. This is a data privacy concern that most current services do not address.
For a broader framework for thinking about AI ethics, our AI ethics for beginners article provides essential context. And our history of AI explores how we arrived at this technological moment.
Legal Dimensions: Rights After Death
The legal landscape around digital identity after death is fragmented, largely unprepared for the specific capabilities these technologies represent, and evolving rapidly — though legislative action has been slow relative to the pace of technological development.
Publicity rights — the right to control the commercial use of one’s name, image, and likeness — exist in many jurisdictions but with highly variable scope and duration. In California, posthumous publicity rights survive for 70 years after death and are inheritable by the estate. In New York, they also extend posthumously. In the UK and much of Europe, comparable rights are more limited. In many countries, publicity rights expire at death entirely, leaving no legal basis for an estate to object to commercial use of a deceased person’s likeness.
Data ownership and platform terms of service: Most major social media platforms’ terms of service do not grant users or their estates explicit ownership rights over the data they have generated. Meta (Facebook/Instagram) offers a Legacy Contact feature and a memorialization option, but does not provide estates with the ability to extract data for AI training purposes as a contractual right. The legal authority of an estate to extract and deploy a deceased person’s digital data for AI model training has not been tested in most jurisdictions.
Voice and likeness rights: Several US states, including California and Tennessee (the ELVIS Act, 2024), have passed or are considering legislation specifically targeting AI voice and likeness replication, including protections that extend to deceased individuals. Tennessee’s ELVIS Act — named both as an acronym and a reference to Elvis Presley — creates explicit cause of action against unauthorized AI replication of a person’s voice. This legislative activity reflects growing political attention to the specific risks of AI-generated personas.
Fraud and impersonation risks: A sufficiently realistic AI persona of a deceased person could be weaponized for fraud — deceiving elderly relatives, generating false evidence in legal disputes, manipulating inheritance proceedings, or impersonating deceased public figures for political purposes. Existing anti-fraud and impersonation laws may or may not cover AI-generated personas, depending on their specific formulation, and prosecution would require establishing that a digital AI persona constitutes legally cognizable impersonation.
Psychological Impact: What the Research Suggests
Research on the psychological effects of griefbot use is still in its early stages, constrained by the novelty of the technology, the ethical complexities of studying bereaved populations, and the wide variation in how these services are designed and used. What follows represents the current state of a nascent literature.
A 2023 paper in the journal Death Studies surveyed bereaved individuals who had used AI chatbots to communicate with digital representations of deceased loved ones. Results were mixed and individual variation was high: some participants reported feeling comforted, experiencing a sense of connection, or gaining access to memories they had feared losing; others reported increased distress, particularly when the AI produced responses that felt incongruent with the real person, or when a session ended and the absence was felt more acutely than before the interaction.
The concept of continuing bonds in contemporary grief theory suggests that maintaining a symbolic connection with the deceased — through memories, photographs, objects, rituals, and internal representations — can be a healthy and culturally normal part of grieving for many people, contrary to older models that emphasized “letting go” and “moving on” as the goals of grief work. Some researchers argue that AI memorials are a technological extension of continuing bonds and that the moral anxiety around them reflects cultural assumptions rather than psychological evidence.
Others, including leading grief researchers and clinicians, have cautioned that the specific properties of interactive AI — its responsiveness, its simulation of reciprocity, its apparent availability — distinguish it qualitatively from passive memorial objects and may affect grieving processes in ways that static memorials do not. The possibility that prolonged interaction with a highly realistic AI persona could interfere with the development of continuing bonds as an internal psychological process, rather than an external technology dependency, is a legitimate concern that deserves empirical investigation.
The emerging clinical consensus is that individual responses vary widely, that context and purpose matter enormously, that any use should be voluntary and fully informed, and that these services should complement rather than substitute for human grief support. Bereavement professionals who work with clients using griefbot technology generally recommend framing it explicitly as a memorial tool with clear limitations, not as an ongoing relationship.
Frequently Asked Questions
Is it ethical to use AI to recreate a deceased loved one?
This is genuinely one of the most contested questions in contemporary AI ethics, and thoughtful people disagree. The key ethical considerations are: whether the deceased consented; how accurately the recreation represents the person; the purpose and context of use; who benefits and who might be harmed. Most ethicists agree that explicit pre-mortem consent from the person themselves is the gold standard. Posthumous reconstruction without documented consent raises serious concerns about autonomy and dignity, even when motivated by genuine love and grief. Services that are built by the person while alive (HereAfterAI, StoryFile) are ethically distinguishable from posthumous reconstructions using data repurposed without consent.
Can a griefbot help with the grieving process?
Some bereaved individuals report finding comfort, a sense of connection, or useful memory preservation through AI memorial services. However, mental health professionals generally caution against using griefbots as a primary coping mechanism, as a substitute for processing loss, or without professional grief support. The research evidence is early and mixed. If you are grieving and considering using such a service, discussing it with a grief counselor or therapist first — to understand how it fits your particular grief process — is strongly advisable.
What data is used to build a digital afterlife AI?
Most services use some combination of text messages, emails, social media posts, voice recordings, and videos. The volume and quality of available data directly determines the richness and perceived accuracy of the resulting persona. Services designed to be created while alive (like HereAfterAI and StoryFile) gather data through structured interviews and conversations. Posthumous services rely on extracting data from the deceased person’s digital accounts, often via family members who have access to credentials.
Are there laws against creating AI personas of deceased people?
Currently, no comprehensive laws specifically and directly govern AI persona recreation of deceased individuals in most countries, though a patchwork of existing laws around publicity rights, data protection, voice and likeness rights, and impersonation applies in various contexts. Several US states have passed or are actively considering legislation — including California’s extensive posthumous publicity rights and Tennessee’s ELVIS Act targeting AI voice replication. This legal landscape is developing rapidly and is expected to see significant legislative activity over the coming years.
What happens to a digital afterlife service if the company shuts down?
This is a serious practical concern that most current services do not address adequately. If a company providing digital afterlife services shuts down, the digital personas it hosts could simply disappear along with any underlying data — effectively a second loss for bereaved families who have built ongoing relationships with these services. Before using any such service, investigate the company’s data portability and export policy, what happens to data on account or company termination, and whether you can retain independent copies of the underlying data used to build the persona. Elon Musk’s AI timeline offers an interesting perspective on these developments.
Sources
Get free AI tips delivered daily → Subscribe to Beginners in AI