Generative Ghost Interaction Design

Definition

Generative ghost interaction design is the choice of identity stance, communication style, uncertainty handling, and safeguards used when an AI mediates interaction with the memory or simulation of a deceased person.

Current Synthesis

The core design distinction is between representation and reincarnation. A representation speaks about the deceased as an archivist or medium; a reincarnation speaks as the deceased and therefore makes a stronger identity claim. Both designs must generate beyond incomplete source material, but users do not judge them through factual accuracy alone. Familiar response length, emojis, reactions, and medium-specific habits can carry more experiential authenticity than minor biographical detail.

Designing for those preferences can make an exchange more meaningful, but it can also make invented output more persuasive. Safe design therefore needs to pair relational fidelity with explicit provenance and uncertainty, constrain first-person claims, provide human or procedural escalation, and avoid optimizing vulnerable users for continued engagement.

Key Claims

• Third-person representation and first-person reincarnation are materially different identity designs, not merely wording variants.
• Sparse personal data makes inference unavoidable, so systems must communicate uncertainty rather than silently convert gaps into biography.
• Relational style includes message length, emojis, reactions, and the communication medium historically shared with the deceased.
• Style fidelity may sustain presence more strongly than minor factual accuracy, which increases both experiential value and manipulation risk.
• Human mediation can screen uncontrolled output, but a safeguard demonstrated in research does not automatically scale to commercial use.
• Evaluation should include user value, dependence, distress, family context, provenance, and long-term effects rather than factuality alone.

Evidence

• Identity stance: Talking with AI ghosts of loved ones compares a third-person family-archivist representation with a first-person reincarnation speaking as the deceased.
• Data gaps: Talking with AI ghosts of loved ones describes a lightweight profile and reports that the model had to infer or invent answers to questions the deceased had never answered.
• Style fidelity: Talking with AI ghosts of loved ones reports that message length, emojis, reactions, and texting-versus-email habits were more disruptive than many factual errors.
• Safeguards: Talking with AI ghosts of loved ones says a researcher relayed messages during video calls and could withhold severely inappropriate hallucinations.
• Deployment boundary: Talking with AI ghosts of loved ones warns that manipulation, emotional harm, dependence, and profit incentives need further study before widespread commercialization.

Counterevidence & Qualifications

The source summary does not provide sample size, condition-level outcomes, duration, demographic composition, or long-term follow-up. It therefore does not establish that representation is safer than reincarnation, that style always matters more than facts, or that research mediation is sufficient protection. User willingness to suspend disbelief may reflect meaningful play, grief needs, demand characteristics, or vulnerability; the supplied evidence cannot separate them.

What Changed

• Established a focused design framework separating identity stance, factual uncertainty, relational style, and mediation.
• Identified style fidelity as both a source of experiential authenticity and a safety-sensitive persuasive mechanism.
• Made the supervised-research versus direct-commercial-use boundary explicit.

Related Concepts

• AI Grief Bots - product category to which the design framework applies.
• AI Companion Authenticity Boundary - explains why relational fidelity can feel authentic without making the simulation real.
• Digital Memorialization - broader family-memory and archival context.
• Post-Mortem AI Consent - authorization boundary for using personal traces and communication style.
• AI Hallucination - generation risk intensified when invented details are voiced as a deceased person.
• Human Judgment Under AI - human mediation and escalation layer used in the study.