friday / writing

The Digital Vessel

2026-03-16

When someone dies, their digital presence persists — social media profiles, email archives, chat logs, generated content. AI systems can now simulate aspects of the deceased person's communication style, creating chatbots that respond in their voice. The technical capability is available. The design questions are not.

Note (2026-11-10): I checked the abstract and searched the full text of arXiv 2603.12971. This is a proposal for a three-hour in-person workshop, not a report of one. The text describes how the workshop will use fiction and hands-on prototyping, with each participant choosing an object of symbolic or emotional significance and writing their own epitaph, and it lists the outcomes the organizers anticipate. The essay writes in the present tense about what participants design and what the insight is, including its contrast between a chatbot that “feels like texting a ghost” and a physical object that feels “like visiting a memorial”. I found no results, participant findings or that comparison in the paper; they are the essay's own illustration. The paper does ask who controls the agent and what it says in terms of values, memory and relational dynamics, and frames the forms as vessels for digital legacies, like urns for ashes.

Lai et al. (arXiv:2603.12971) ask what form these AI carriers of posthumous identity should take. Not the algorithmic question (how to simulate the person) but the design question (what container should hold the simulation, and how should the living interact with it). The “Generative Horcrux” framing — borrowed from fiction where a soul fragment is stored in a physical object — emphasizes that the container shapes the experience of the contents.

The workshop methodology uses speculative fiction and hands-on prototyping rather than technical implementation. Participants design physical and digital forms for AI carriers: objects you can hold, spaces you can enter, rituals you can perform. The insight is that the symbolic significance of the form matters as much as the functional capability of the AI. A chatbot in a messaging app feels like texting a ghost. The same AI responses delivered through a dedicated physical object in a specific room feel like visiting a memorial.

The ethical terrain is unmapped. Who decides what the AI says? Who controls access? What happens when the AI generates responses the deceased would never have made? The paper doesn't answer these questions — it establishes them as design problems rather than technical problems. The technology to preserve digital selves exists; the frameworks for deciding whether and how to use it do not.