Digital ghosts: AI companies create chatbots that simulate conversations with deceased people

Artificial intelligence technology is breaking one of the most sensitive barriers of human experience: death. Generative AI specialized companies train large language models with the vast digital trail left by deceased people — social media publications, text messages, emails, and even voice recordings — to create chatbots capable of simulating conversations with those who have gone. The phenomenon, nicknamed 'generative ghosts', is already marketed by startups such as American HereAfter AI and Chinese XiaoIce, and raises urgent questions about the use of posthumous data and the emotional impact on mourners.

How do companies create generic ghosts with AI?

The process begins with the collection and digitization of all content written, recorded or filmed by the deceased person. This data is then used to adjust (fine-tuning) a large language model (LLM), such as those that feed ChatGPT or Claude. The result is a custom chatbot that tries to reproduce the tone, opinions, humor, and even small speech habits of the deceased. Interaction can happen by text or voice, on dedicated platforms or integrated to virtual assistants.

Some companies go beyond: they offer animated avatars in 3D, who "gave life" on screens or augmented reality glasses. The promise is to allow the user to "convert" with the loved one as if he were still present. Despite the emotional appeal, experts warn that technology is still limited: the model has neither consciousness nor true memory — it only reproduces statistical patterns based on the data provided.

Why do generative ghosts generate ethical alerts?

The main concern is with consent. Unlike a lifetime data use contract, the deceased cannot authorize the re-creation of his personality. It is the family or close friends who decide — which can lead to uses that the deceased would never approve of. There is also the risk of perpetuating abusive relationships, emotional dependence or even fraud, with the creation of digital clones without the knowledge of the heirs.

Another critical point is posthumous privacy. Personal emails, intimate messages and even health records can be exposed during training. Current legislation, such as the General Data Protection Act (LGPD) in Brazil, does not clearly cover the use of data from deceased people — a gap that AI companies exploit freely.

The customer service industry with outsourced AI(/noticias/atendimento-ao-cliente-ferramentas-de-ia-terceirizadas-superam-chatbots-proprios) has faced similar privacy challenges, but the emotional impact here is much greater. Psychologists fear that chatbots may prolong pathological mourning, preventing acceptance of loss.

What do companies and users think about generative ghosts?

Startups argue that they offer comfort to those suffering, and that many users report temporary therapeutic benefits. American StoryFile, for example, created an interactive hologram of a Holocaust survivor, preserving her memory for schools. XiaoIce allows users to create "digital souls" of relatives, with millions of registered downloads.

However, critics point out that the business model depends on sensitive data that, once loaded into the cloud, can be used to train other systems or sold to third parties — even if the terms of use say otherwise. In an interview with CBS News, the original source of this report, a spokesman for one of the companies admitted that "the line between homage and exploration has not yet been drawn".

Frequently Asked Questions about Generating Ghosts

Are the 'generative ghosts' accurate regarding the real personality?

Accuracy depends on the quantity and quality of the available data. The more records (conversations, emails, publications), the more the model can imitate the style of communication. However, chatbot is a limited and statistical representation — it has no real memory and can generate answers the deceased would never give.

Is it cool to create a dead person's chatbot without his consent in life?

In most countries, including Brazil, there is no clear legislation on this. The authorization is usually given by relatives, but the deceased cannot have opposed himself in life. Experts argue that, as there are rules for organ donation, there should be for posthumous use of personal data.

How to protect the data of a loved one after death?

Currently, the best way is to explicitly define in a will or digital will document that data should not be used for AI purposes. It is also possible to ask the social media platform for permanent exclusion of profiles after death, before third parties can collect them.