Generative Ghosts: How Users Prefer AI Reincarnations of the Dead
A new study from arXiv (2605.21390) investigates how people interact with "generative ghosts"—AI systems trained on data of the deceased. Researchers compared two design modes: representation (AI speaks about the dead in third person) and reincarnation (AI speaks as the dead in first person). In a qualitative study with 16 participants, reincarnation was preferred for its immediacy, though participants feared over-reliance. Representation was favored for engaging with memory, but users often ignored the third-person framing and conversed with the AI anyway. Across both modes, participants valued affective resonance over factual accuracy. The study concludes that factors like tone, language, and conversational rhythm—unique to each user's memory—shape these interactions.
Key facts
- Study examines generative ghosts: AI trained on data of the dead
- Two design modes: representation (third person) and reincarnation (first person)
- Qualitative user study with 16 participants
- Reincarnation preferred for immediacy, but fears of over-reliance
- Representation preferred for memory, but users often conversed despite third-person framing
- Affective resonance prioritized over factual fidelity
- Tone, language, and conversational rhythm shape interactions
- Published on arXiv with ID 2605.21390
Entities
Institutions
- arXiv