The Research Agenda

Six questions on computational fidelity.

Forever Spoken is advancing a new research domain focused on computational fidelity, provenance, and governance for AI representations of human identity. Our agenda is organized around six open questions, each a defensible research and societal problem.

// Proposed Research

A Computational Fidelity Framework for Truthful AI Representations of Human Identity

Forever Spoken is investigating how AI systems can represent a specific human identity truthfully when the available evidence includes verified facts, self-reported experiences, subjective memories, third-party observations, evolving beliefs, and conflicting perspectives.

Question 01

Representation

How should a human identity be computationally represented, so that values, judgment, personality, and lived experience are captured, not just facts and demographics?

Question 02

Authenticity Attributes

Which attributes most strongly influence whether a representation is perceived as authentic by the people who knew the person best?

Question 03

Truthfulness and Fidelity

How can an AI system determine whether an identity-related claim is sufficiently supported, appropriately qualified, attributable to a particular perspective, or too uncertain to assert?

This is the scientific heart of the proposed work. It is where provenance, corroboration, and abstention policy meet.

Question 04

Measurement

How can fidelity and perceived authenticity be measured reliably and reproducibly using complementary computational metrics and human evaluation?

Validating any such metric requires both technical benchmarks and human-subject study.

Question 05

Evolution

How should a digital identity evolve as new evidence arrives, while remaining faithful to the individual and never drifting into fabrication?

Question 06

Generational Trust

What governance mechanisms best preserve trust across generations, so a representation remains consented, accountable, and worthy of belief decades from now?

// The Claim-Level Model

Every response is a set of claims, not a monolith.

The framework treats an AI response as a collection of material claims. Each claim is evaluated using its evidence provenance, source authority, corroboration, temporal relevance, perspective status, uncertainty, and disclosure authorization. The system then determines whether to assert, qualify, attribute, present competing perspectives, or abstain.

Assert Qualify Attribute Present Competing Perspectives Abstain

Why It Matters

Continuity is only valuable if it stays truthful.

Every generation loses expertise, context, and judgment that never made it into a record. Preserving access to that knowledge matters only if the representation stays truthful and faithful to its source, not merely present. Advancing this agenda returns value across families, institutions, culture, and the broader field of trustworthy AI.

For families and aging adults

A relative's knowledge and stories stay reachable, with clear signals about what is verified, self-reported, or remembered differently by different people.

For institutions and the workforce

Critical expertise stays queryable after the expert retires, with claims traceable to their source rather than presented as uniform fact.

For culture and scholarship

Archives, founders, and elders enter living conversation with citation and appropriate uncertainty, not flattened into a single confident voice.

For trustworthy AI

A governed model for how persistent AI systems represent identity responsibly: evidence-based, auditable, and willing to qualify or abstain.