The Right to Be Remembered: Preserving Maximally Truthful Digital Memory in the Age of AI (2510.16206v2)

Abstract: Since the rapid expansion of LLMs, people have begun to rely on them for information retrieval. While traditional search engines display ranked lists of sources shaped by search engine optimization (SEO), advertising, and personalization, LLMs typically provide a synthesized response that feels singular and authoritative. While both approaches carry risks of bias and omission, LLMs may amplify the effect by collapsing multiple perspectives into one answer, reducing users ability or inclination to compare alternatives. This concentrates power over information in a few LLM vendors whose systems effectively shape what is remembered and what is overlooked. As a result, certain narratives, individuals or groups, may be disproportionately suppressed, while others are disproportionately elevated. Over time, this creates a new threat: the gradual erasure of those with limited digital presence, and the amplification of those already prominent, reshaping collective memory. To address these concerns, this paper presents a concept of the Right To Be Remembered (RTBR) which encompasses minimizing the risk of AI-driven information omission, embracing the right of fair treatment, while ensuring that the generated content would be maximally truthful.

• The paper introduces RTBR as a counterpoint to the right to be forgotten, emphasizing the ethical imperative to preserve digital memory in AI systems.
• The paper details how LLMs synthesize information, often amplifying dominant narratives and marginalizing less prominent perspectives, while proposing layered attribution.
• The paper advocates for integrating technical and regulatory measures that balance digital memory preservation with privacy concerns to ensure collective historical accuracy.

Preserving Maximally Truthful Digital Memory: The Right to Be Remembered in AI Systems

Introduction

This paper introduces and formalizes the concept of the Right To Be Remembered (RTBR) as a counterpoint to the established "right to be forgotten" in the context of AI-mediated knowledge systems, particularly LLMs. The authors argue that as LLMs become the dominant interface for information retrieval, the preservation of maximally truthful digital memory is not only a technical challenge but also an ethical and societal imperative. The RTBR is positioned as a normative claim: contributions to human knowledge—scientific, cultural, or social—should remain accessible and visible within digital infrastructures, ensuring fair treatment and minimizing AI-driven information omission.

The Dynamics of Digital Memory and Erasure

LLMs, trained on vast corpora, synthesize information rather than retrieve discrete facts, leading to the collapse of multiple perspectives into singular, authoritative responses. This process amplifies the risk of bias, omission, and the gradual erasure of less digitally prominent narratives. The paper highlights empirical evidence of "link rot" and the fragility of digital records, noting that a substantial portion of online content becomes inaccessible within a decade. The consequences are profound: marginalized communities, non-English scholarship, and contributions outside major institutions are disproportionately excluded from AI-mediated knowledge retrieval, entrenching epistemic inequities.

The authors emphasize that remembrance is not a neutral by-product but a process structured by institutional, economic, and technological forces. The RTBR is thus framed as a societal duty to maintain a truthful and inclusive record, recognizing that collective memory sustains accountability, enables progress, and fosters fairness.

Responsibility for Forgetting: Vendors, Models, and Users

The paper delineates the sources of bias and omission in LLM outputs. Vendors exert direct control over training data selection, filtering, moderation, and deployment design, shaping the boundaries of what models can "know." RLHF and other fine-tuning methods embed human judgments, potentially entrenching biases. LLMs themselves, due to their statistical nature, amplify dominant patterns and underproduce rare or marginalized perspectives, exhibiting topic-dependent biases even across architectures. User prompt phrasing further influences outputs, but vendor-imposed constraints remain primary.

The interplay between vendor choices and model tendencies creates a dual responsibility for the erasure or distortion of digital memory. The authors argue that accountability must be demanded from vendors, while technical solutions must address intrinsic model biases.

Ensuring Maximal Truthfulness in AI

Maximal truthfulness is conceptualized as a multi-layered construct encompassing statistical accuracy, epistemic relation to external evidence, and normative honesty. The paper reviews emerging benchmarks (e.g., MASK) that formalize distinctions between model "beliefs" and ground truth, and highlights geometric analyses of neural activations that reveal structural signatures of truthfulness. Notably, truthful responses are encoded in more compact, lower-dimensional activation patterns, while hallucinations are scattered across higher-dimensional manifolds.

The authors advocate for layered provenance and attribution mechanisms, drawing on standards like EKILA and C2PA, to preserve recognition of contributors without compromising efficiency. They note that current solutions capture data lineage and content authenticity but lack systematic epistemic acknowledgment of intellectual labor. Transparency about uncertainty is also emphasized; models should abstain from answering when evidence is insufficient, prioritizing honesty over fluency.

RTBR vs. Right to Erasure: Legal and Technical Tensions

The RTBR is contrasted with the "right to be forgotten," particularly as codified in GDPR Article 17. While traditional systems allow for discrete data deletion, LLMs embed knowledge in parameter space, making erasure technically challenging and potentially damaging to model integrity. The authors argue that, in foundational AI, the collective RTBR—ensuring maximal truthfulness and historical accuracy—should take precedence over individual claims to erasure, especially for deceased individuals whose digital legacy holds societal value.

The phenomenon of "machine unlearning" is discussed as an emerging but imperfect solution for data removal from models, often resulting in degraded utility and coherence. The authors assert that erasing model "memories" risks introducing gaps and distortions into the historical record, undermining the integrity of collective knowledge.

Implications and Future Directions

The paper reframes digital memory preservation as an ethical, technical, and societal imperative. It calls for a shift in AI design toward systems that explicitly optimize for epistemic integrity, incorporating layered attribution, traceable provenance, and uncertainty calibration as fundamental architectural components. Regulatory frameworks must evolve to balance privacy protection with the necessity of preserving collective memory.

Empirical findings on the geometric structure of truthfulness in neural networks invite further research into architectures that align internal representations with verifiable knowledge states. The challenge is to ensure that the archive of human thought remains inclusive, accurate, and representative, preserving meaningful human contributions rather than perpetuating all data indiscriminately.

Conclusion

The integrity of AI-mediated digital memory will increasingly define the boundaries of collective understanding. The RTBR articulates a new ethical foundation: digital memory should be preserved as a public good, and truthfulness in AI depends on the inclusivity and continuity of the human record. The authors argue that RTBR and maximal truthfulness are net positive for both individuals and populations, facilitating optimal human-AI convergence. The paper raises important philosophical questions regarding equal and fair treatment of individual data, the impact of digital immortality, and the future of human behavior in the age of AI, but maintains that the preservation of maximally truthful digital memory is essential to sustaining the conditions of human understanding.