Legal Recognition Of Digital Intellectual Heritage Restored Through Machine Learning

18 Mar 2026 --
0 Comments

1. Introduction

Digital intellectual heritage refers to creative, scientific, or cultural works preserved in digital form, which could include literature, music, films, historical documents, or art. With machine learning (ML), we can restore, reconstruct, or even enhance these works that might have been lost, damaged, or incomplete.

The question is: Once restored via ML, who owns the rights? Does the law recognize such works as intellectual property? This is complex because traditional IP laws were designed for human-created works, not machine-assisted or generated restorations.

2. Core Legal Principles

Copyright Law: Protects original works of authorship fixed in a tangible medium. The key issue is originality. ML restoration raises questions:
- Is the restored work sufficiently original?
- Who is the “author”: the human operator, the algorithm, or the dataset provider?
Moral Rights: Even if ML restores a work, original creators’ moral rights (right to attribution, right to prevent derogatory treatment) may apply.
Derivative Works: ML-restored works can be treated as derivative works if they are based on existing copyrighted works.
Database Protection: Collections of digital heritage might be protected as databases even if individual elements are public domain.

3. Case Law Examples

Case 1: Naruto v. Slater (2018) – US

Facts: This involved a monkey taking selfies with a photographer’s camera. Though not ML, it set a precedent for non-human authorship.
Principle: Only humans can claim copyright under U.S. law.
Relevance: If an ML system autonomously restores digital heritage, the copyright may not automatically vest in the machine itself. Human contribution is crucial.

Case 2: Infopaq International A/S v. Danske Dagblades Forening (2009) – EU

Facts: Infopaq used software to extract and summarize snippets from newspaper articles. The court asked whether these snippets constituted copyrightable reproduction.
Principle: Even small parts can be protected if they are original and have intellectual creation.
Relevance: ML-restored fragments of digital heritage could be considered original if a human curated or guided the restoration process.

Case 3: Bridgeman Art Library v. Corel Corp (1999) – US

Facts: Bridgeman claimed copyright over exact photographic reproductions of public domain artworks.
Outcome: Court held that exact reproductions of public domain works are not copyrightable, because there is no originality.
Relevance: ML restoration must add creative input to qualify for copyright; exact digital replicas of old works may not be protected.

Case 4: Authors Guild v. Google (Google Books, 2015) – US

Facts: Google scanned millions of books for indexing, including copyrighted books.
Principle: Fair use applied, as transformative use for search indexing did not replace original works.
Relevance: ML can restore digital heritage as a transformative use; if restoration enhances access, it may be recognized under fair use/fair dealing.

Case 5: SAS Institute Inc v. World Programming Ltd (2012) – UK/EU

Facts: Concerned software functionality and the copying of algorithms.
Principle: Ideas and functionality cannot be copyrighted, only expression.
Relevance: ML models that restore works may rely on algorithms trained on public domain datasets; only the output with originality may gain protection, not the model itself.

Case 6: Authors Guild v. HathiTrust (2014) – US

Facts: Universities digitized copyrighted works for preservation and accessibility.
Principle: Courts recognized the use as transformative and for non-commercial purposes, supporting public interest.
Relevance: Restoration of digital heritage via ML can be defended as transformative, especially for preservation.

4. Emerging Legal Trends

Human-in-the-Loop Requirement: Many jurisdictions (US, EU) emphasize human creativity as a prerequisite for copyright. ML-assisted works where a human supervises may qualify.
Derivative Work Recognition: Restored works can be considered derivative; original creators’ rights must be respected.
AI and Copyright Reform: Some countries (UK, EU proposals) are considering giving limited recognition to AI-assisted works if there is substantial human input.

5. Practical Takeaways

Originality Matters: The more human guidance in ML restoration, the stronger the copyright claim.
Attribution: Even for restored heritage, moral rights apply if the original author is known.
Derivative vs. Original: Distinguish between pure reproduction and transformative restoration.
Public Domain Awareness: Works fully in the public domain may have no copyright but can be restored for commercial purposes if sufficient creativity is added.

Summary Table: Key Cases and Principles

Case	Jurisdiction	Principle	Relevance to ML Restored Heritage
Naruto v. Slater	US	Non-human authors cannot hold copyright	ML-only restoration may not get copyright
Infopaq v. DD	EU	Short excerpts can be protected if original	ML-restored snippets may gain copyright
Bridgeman v. Corel	US	Exact replicas of public domain works not copyrightable	ML replicas need creative input
Authors Guild v. Google	US	Transformative use = fair use	ML restoration can be transformative
SAS Institute v. WPL	UK/EU	Software ideas not copyrightable	Algorithms alone don’t get copyright
Authors Guild v. HathiTrust	US	Non-commercial preservation can be lawful	Supports ML heritage preservation