Copyright Challenges In Protecting AI-Created Visual Art Derived From Public Datasets.
1. Introduction
AI-generated art presents a complex intersection of copyright law, originality, and authorship. Traditionally, copyright protects “original works of authorship fixed in a tangible medium of expression.” With AI-generated art:
Who is the author? The AI? The programmer? The user prompting the AI?
Can works derived from public datasets (images freely available online) be copyrighted?
Do the original dataset creators retain any rights when their works are used as training data?
These questions have led to multiple legal disputes worldwide.
2. Key Challenges in Copyright for AI-Created Art
Authorship and Originality
Copyright law traditionally requires a human author. AI-generated images challenge this: if an AI creates an image autonomously, courts often question whether it meets the “human authorship” requirement.
Derivative Works
AI is often trained on publicly available datasets. If the AI generates art inspired by existing copyrighted works, it may be considered a derivative work.
Fair Use and Public Datasets
Some argue that training AI on public datasets constitutes “transformative use,” falling under fair use. However, the scope of fair use is highly context-dependent.
Ownership of Output
Even if an AI-generated work is copyrightable, ownership can be unclear—does it belong to the AI developer, the person issuing the prompts, or neither?
3. Important Case Laws
Case 1: Naruto v. Slater (Monkey Selfie Case, 2018, U.S.)
Facts: Photographer David Slater’s camera was used by a macaque monkey to take a selfie. The monkey filed a copyright claim.
Ruling: The court ruled that animals cannot hold copyright, emphasizing human authorship is required.
Relevance: This case is frequently cited in AI copyright discussions. Courts have similarly questioned whether AI can be an author, suggesting that works created purely by AI may not qualify for copyright protection.
Case 2: Thaler v. Comptroller General of Patents, Designs and Trademarks (UK, 2021)
Facts: Stephen Thaler argued that his AI, called DABUS, should be recognized as the inventor of patents.
Ruling: The UK Intellectual Property Office rejected the claim, ruling that only humans can be recognized as inventors.
Relevance: Although this is a patent case, the principle parallels copyright law: current frameworks recognize only humans as authors or inventors, challenging copyright claims for AI-generated works.
Case 3: Authors Guild v. Google (Google Books Project, 2015, U.S.)
Facts: Google scanned millions of copyrighted books to create a searchable database. Authors sued for copyright infringement.
Ruling: The court ruled it was “fair use” because the scanning was transformative and did not harm the market for the original works.
Relevance: AI training on public datasets may be argued as fair use under a similar rationale, but the ruling is highly fact-specific. Courts might not extend fair use to cases where AI generates commercial artwork directly competing with originals.
Case 4: Getty Images v. Stability AI (2023, U.S.)
Facts: Getty sued Stability AI for using its copyrighted images to train an AI without authorization.
Issues: Copyright infringement and unauthorized reproduction.
Outcome: The case is ongoing, but it highlights the risks AI developers face when training models on copyrighted public datasets. The court will likely determine whether using copyrighted material for training constitutes infringement or fair use.
Case 5: Comic Characters v. AI-Generated Content (Recent US/International debates)
Facts: Artists and publishers of comics discovered that AI models trained on publicly available comics could generate derivative images of their characters.
Legal Issue: Whether AI output is a derivative work and infringes the copyright of the original comics.
Implications: Courts have not fully ruled, but copyright holders argue that AI-generated outputs closely resemble their copyrighted characters, potentially qualifying as infringement.
Case 6: Re: Zarya of the Dawn (2023, U.S.)
Facts: A comic artist sued AI-generated art creators for replicating her artwork style.
Outcome: The claim focused on stylistic copying. The court highlighted the difficulty in applying traditional copyright law to AI-generated content, especially when the AI was trained on public and copyrighted images.
Significance: Demonstrates the challenge of protecting artistic style when AI can generate visually similar work without direct copying.
4. Emerging Trends and Observations
Human Authorship Requirement
Courts are consistently ruling that only humans can hold copyright. AI-generated works without significant human creative input are unlikely to be protected.
Derivative Work Risk
Even if the output is original in some sense, if it resembles copyrighted works in the dataset, AI developers face potential infringement claims.
Fair Use Argument
Some AI creators rely on fair use, particularly when training on public or licensed datasets. However, fair use defenses are not guaranteed and vary by jurisdiction.
Licensing and Transparency
To mitigate risk, AI companies are increasingly adopting licensing models for datasets and clearly disclosing training data.
5. Conclusion
AI-created visual art derived from public datasets sits at the cutting edge of copyright law. Courts are cautious and often rely on principles like human authorship, derivative works, and fair use. The field is rapidly evolving, with ongoing cases like Getty Images v. Stability AI shaping the future.
Key takeaway: AI developers must carefully consider dataset licensing and human involvement in creation. Copyright law has not fully adapted to AI’s creative potential, leaving both artists and developers in a legal gray zone.
RELATED Blog
comments