If you could chat with an AI ghost, what would you want them to say? New study explores

· Source: News on Artificial Intelligence and Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

CU Boulder researchers Jack Manning and Jed Brubaker conducted the first scientific study on "generative ghosts," AI agents trained on data of deceased individuals. Published in the Proceedings of the 2026 Designing Interactive Systems Conference, the study involved 16 participants, aged 22-50, interacting with AI-generated representations of lost loved ones via Zoom. Key findings indicate participants preferred first-person "reincarnation" ghosts over third-person "representation," valuing accurate emotional tone, dialect, and conversational rhythm. While minor AI "hallucinations" were tolerated, incorrect terms of endearment caused discomfort, such as a ghost calling a participant "champ." All participants expressed willingness to use the technology again but feared its unsupervised use by grieving loved ones. Commercial platforms like Project December, Séance AI, and HereAfterAI already offer text, voice, and VR-based generative ghost services.

Key takeaway

For AI Scientists and Ethicists developing generative AI for grief support, prioritize designing "reincarnation" agents that speak in the first person and meticulously replicate personal linguistic nuances. Your focus should be on ensuring emotional accuracy and conversational rhythm, as minor inaccuracies like incorrect terms of endearment can severely disrupt user experience. Crucially, integrate robust safeguards to prevent unsupervised use, especially by vulnerable individuals, to mitigate potential psychological risks while maximizing therapeutic benefits.

Key insights

Generative AI ghosts offer profound emotional closure but require careful design to mitigate risks and maximize user preference.

Principles

Method

Researchers built real-time LLM-based ghosts using biographical data, then had participants chat with first-person and third-person iterations for 20 minutes, fine-tuning as needed.

In practice

Topics

Best for: AI Product Manager, AI Scientist, Research Scientist, AI Ethicist

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by News on Artificial Intelligence and Machine Learning.