Right to Be Remembered (RTBR)

• RTBR is a normative framework ensuring digital contributions are preserved truthfully and inclusively, safeguarding both mainstream and marginalized narratives.
• It addresses AI-induced biases by promoting layered attribution, retrieval-augmented generation, and uncertainty calibration to fortify digital memory.
• The framework informs AI system design and policy reforms, balancing ethical imperatives with technical innovations for accurate and equitable historical records.

The Right To Be Remembered (RTBR) is a normative framework asserting that every contributor—whether scientist, artist, or citizen—retains the right for their digital achievements and records to be maintained in a maximally truthful manner. This paradigm extends digital memory from its conventional role as an incidental by-product of data aggregation and AI computation to that of a fundamental public good, preserving both prominent and marginalized narratives across the digital spectrum.

1. Formal Definition and Scope

The RTBR is defined as the right for digital contributions to be preserved comprehensively, with fair treatment and maximal truthfulness. It safeguards not only distinguished records, but also the diverse, less-visible knowledge contributed by individuals and communities. Three principal components delineate the RTBR:

• Guarantee of fair treatment: Digital narratives from marginalized or low-visibility groups must not be suppressed by dominant data trends.
• Maximal truthfulness: Records should be contextually accurate, exhaustive, and resistant to distortion from biased or oversimplified AI syntheses.
• Integrity of collective memory: Digital memory under RTBR serves as a foundational resource for future research, accountability, and cultural continuity.

2. AI-Driven Challenges to Digital Memory

Current LLMs pose unique risks to the preservation of digital memory:

• Risk of digital amnesia: LLMs consume heterogeneous, incomplete corpora. Underrepresented or decaying data (e.g., due to link rot) skews syntheses toward dominant narratives.
• Amplification of mainstream perspectives: Vendor-driven selection biases and statistical regularities in training datasets result in responses that blend multiple perspectives into authoritative, but potentially incomplete, answers.
• Structural bias: The inherent design of LLMs, which optimize for dominant data patterns, accelerates the erasure of underrepresented viewpoints.

These phenomena collectively undermine the breadth and diversity of digital memory, creating disproportionate visibility for established narratives and decreasing the representation of marginalized contributions.

3. Technical Mechanisms for Memory Preservation

To mitigate LLM-induced omission and bias, several strategies are advanced for preserving maximally truthful digital records:

• Layered attribution and provenance: Instead of merging sources into a unified synthesis, information systems should enable users to access detailed citation chains and attribution layers. This mirrors initiatives such as EKILA and standards like C2PA, which aim to preserve the visibility and traceability of original contributions.
• Retrieval-Augmented Generation (RAG): Coupling LLM outputs with retrieval subsystems grounds answers in verifiable external evidence. However, RAG is inherently limited: it cannot access data that was never digitized or has already faded from the digital corpus.
• Internal model analysis for truth detection: Truthfulness in responses is empirically associated with specific geometric properties of model activations. Notably, truthful outputs correspond to compact regions in the neural activation space, measured via local intrinsic dimension (LID). Formally,

$$\text{Truthfulness} \propto \frac{1}{\text{LID}(x)},$$

where $\text{LID}(x)$ denotes the local intrinsic dimension for the activation pattern $x$. Empirical evidence suggests

$\text{LID(truthful)} < \text{LID(hallucinated)}$

—truthful responses cluster in lower-dimensional subspaces than hallucinated ones.

• Uncertainty calibration and abstention: Incorporating abstention mechanisms allows models to convey epistemic uncertainty or refrain from producing answers when available evidence is insufficient.

4. Effects on Individuals and Narratives

The societal ramifications of RTBR are substantial. Digital erasure disproportionately affects scholars from the Global South, minority communities, and non-dominant language contributors—individuals and groups with reduced digital visibility. As LLMs continue to shape what is “remembered,” the absence of RTBR may result in:

• Skewed collective memory: Omission or misrepresentation of contributions from less digitally active segments distorts historical and scientific narratives.
• Suppression of diversity: The shrinking of the digital record to favor prominent voices leads to echo chambers and hinders future innovation.
• Compromised accountability: Gaps in the digital memory impact transparency in sciences, history, and societal discourse, ultimately impairing the reliability of knowledge infrastructures.

A plausible implication is that persistent neglect of RTBR mechanisms could precipitate long-term epistemic injustice, where entire scholarly trajectories or cultural legacies vanish from collective awareness.

5. Ethical and Societal Context

RTBR is anchored in justice, fairness, and epistemic responsibility:

• Ethical imperatives: There is a societal obligation to sustain recognition of all contributions, irrespective of initial visibility or digital footprint.
• Tension with the Right to Be Forgotten: Unlike the privacy-led doctrine of data deletion, RTBR promotes retention—even posthumously—to maintain historical continuity.
• Balancing remembering and forgetting: Protecting privacy requires selective erasure, yet systematic memory loss undermines shared cultural and historical authenticity.

This tension highlights the complex intersection in AI-driven information environments, where managing trade-offs between privacy and memory is integral to ethical information governance.

6. Quantitative and Model-Based Frameworks

Dimensional analysis and formal modeling underpin the technical aspects of RTBR:

• Geometric analysis: As outlined above, local intrinsic dimension metrics capture differences in activation space structure between truthful and hallucinatory responses.
• Consistency frameworks (e.g., MASK): Model “belief” is quantified by consistency across neutral and pressured prompts. Accuracy is formally defined by:

$$\text{Accuracy} = \frac{1}{N} \sum_i \mathbb{1}(\hat{Y}_i = Y_i)$$

where $\mathbb{1}$ is the indicator function, $\hat{Y}_i$ denotes the model’s output under neutral conditions, and $Y_i$ represents ground-truth. This quantitative framework aligns model responses with external standards of evidence, reinforcing the pursuit of maximal truthfulness.

7. Implications for AI System Design and Policy

The RTBR advocates for embedding fairness and inclusivity at the core of AI architectures. Technical innovation must be coupled with societal and legal reforms to ensure that digital memory reflects the full diversity of human history. A plausible implication is that future AI models and knowledge systems will be evaluated not only by their capacity for synthesis and reasoning, but also by their effectiveness in preserving the rights of contributors to be accurately and equitably remembered.