Metaの故人シミュレーションAI特許が示す『幽霊労働』の危険性

Last week, Business Insider reported on a Meta patent describing a system that would simulate a user’s social media activity after their death.The patent imagines a world where you’d be able to chat with a deceased friend’s Facebook or Instagram account after their death, and have a large language model simulate their posting or chatting behavior.

Meta first filed the patent in 2023, but the patent made headlines this week because of its dystopian implications. And while Meta told Business Insider that “we have no plans to move forward with this example,” a recently published paper from researchers at the Hebrew University of Jerusalem and Leipzig University shows that generative AI is increasingly being used to puppeteer the likeness of dead people. The paper argues that the practice raises “urgent legal and ethical questions around posthumous appropriation, ownership, work, and control.”

“Meta’s patent is big, and might even be a turning point,” Tom Divon, the lead author on Artificially alive: An exploration of AI resurrections and spectral labor modes in a postmortal society, told me in an email. “What makes it different is the scale. In our research, most of the AI resurrections we examined were quite bespoke, projects started by families, advocacy groups, museums, or startups, usually tied to very specific emotional, political, or commercial contexts. Even when they existed as apps, they were optional and limited, not built into the core structure of a platform. Meta’s proposal feels different because it imagines posthumous simulation as something woven directly into social media infrastructure.”

Using technology to animate the dead or simulate communication with them is not new, but the practice is becoming more common because generative AI tools are more accessible. Divon and co-author Christian Pentzold analyzed more than 50 real-world cases from the United States, Europe, the Middle East, and East Asia where AI was used to recreate deceased people’s voices, likeness, and personality, to see how and why technology was used this way.

They say that the examples they studied fell into three categories:

Spectacularization: “the digital re-staging of famous figures for entertainment.” For example, a live tour of an AI-generated Whitney Houston.

Sociopoliticization: “the reanimation of victims of violence or injustice for political or commemorative purposes.” We recently covered an example of this with an AI-generated dead victim of a road rage incident giving testimony in court.

Mundanization: “the most intimate and fast-growing mode, in which everyday people use chatbots or synthetic media to ‘talk’ with deceased parents, partners, or children, keeping relationships alive through daily digital interaction.”

The paper raises questions about this growing practice more than it proposes solutions. How does the notion of identity change when multiple versions of oneself can exist simultaneously, and what safeguards do we need to prevent exploitation of people after their death?

“The legal and ethical frameworks governing issues such as consent, privacy, and end-of-life decision-making demand reevaluation to accommodate the challenges posed by afterlife personhood,” the paper says. “In particular, to date, there is no clear line for governing the intricate intertwining of an individual’s data traces and GenAI applications.”

Divon told me that thinking about these issues is especially relevant when it comes to Meta’s patent. “Spectral labor describes how the dead can be made to ‘work’ again through the extraction and reanimation of their data, likeness, and affect. At small scale, this already raises ethical concerns. But at platform scale, we think it risks turning posthumous presence into an ongoing source of engagement, content, and value within digital economies [...] Meta’s patent makes us wonder, will individuals be given the ability to define their post-life boundaries while still alive? Will there be mechanisms akin to a digital DNR [do not resuscitate]?”

Divon explained that the current legal frameworks are not well equipped to address this technology because “digital remains” are typically approached either as property to be inherited or privacy interests to be protected. AI turns those materials into something interactive that can change and generate revenue in the present. Legislators, he said, should focus on getting explicit and informed “pre-death” consent requirements for posthumous AI simulation. Some laws that address this issue are already in progress.

“At its core, we believe the primary concern here centers on authorization,” he said. “Most individuals have not provided explicit, informed consent for their digital traces to power interactive posthumous agents. If such systems become embedded in platform infrastructure, inaction could quietly function as implicit agreement [...] We believe it is crucial to ask whether individuals should continue to generate social and economic value after death without having meaningfully agreed to that form of use.”