Ai Training Datasets Copyright Infringement.
PART 1: AI TRAINING DATASETS AND COPYRIGHT INFRINGEMENT
1. Overview
AI systems, especially large language models, generative AI, and computer vision systems, require large-scale datasets for training. These datasets often contain:
Text from books, articles, and websites
Images, audio, and videos
Scientific papers, patents, or proprietary data
Legal concern: Copying copyrighted material without a license could constitute infringement, even if the AI model does not directly replicate the work.
Key Issues:
Training on copyrighted works: Is it fair use or infringement?
Output similarity: Does AI-generated content copy protected expression?
Derivative works: Can AI outputs be treated as derivative of copyrighted data?
Licensing obligations: Are dataset providers required to secure permissions?
2. Key Case Laws
Case 1: Authors Guild v. Google (2015)
Facts: Google scanned millions of books to create a searchable AI-like database.
Decision: Court held Google’s use was transformative fair use, as it did not substitute the books’ market.
Implication:
Training an AI on copyrighted material may be permissible if it is transformative.
Direct reproduction or substantial output duplication could still infringe.
Case 2: Authors Guild v. OpenAI (Ongoing, 2023–2025)
Facts: Allegation that OpenAI’s models were trained on copyrighted works without licenses.
Issue: Can AI model training constitute copyright infringement?
Status: Litigation ongoing; highlights risk of dataset scraping without consent.
Implication: Shows increasing scrutiny on AI dataset curation and licensing.
Case 3: Github Copilot / Microsoft v. Plaintiffs (2022–Present)
Facts: Copilot AI reproduces code from open-source repositories without attribution.
Decision: Pending, but raises issues of direct copying vs. model-assisted generation.
Implication:
Training datasets containing copyrighted content must consider output replication.
AI outputs that replicate large portions of copyrighted text/code may lead to derivative infringement claims.
Case 4: Authors Guild v. HathiTrust (2012)
Facts: HathiTrust created a searchable digital library of copyrighted books for research.
Decision: Courts held it was fair use due to transformative academic and research purposes.
Implication: Training AI for research or non-commercial purposes may be safer than commercial AI model deployment.
Case 5: Oracle v. Google (Java API case, 2014–2021)
Facts: Google used Java APIs without a license to build Android.
Decision: Supreme Court ruled fair use due to transformative purpose.
Implication for AI: Using APIs, datasets, or other copyrighted resources can be fair use if sufficiently transformative and non-market substitutive.
Key Takeaways for AI Datasets
Transformative Use: AI training must alter the original work sufficiently to reduce infringement risk.
Output Monitoring: AI output must not reproduce copyrighted material verbatim.
Licensing: Obtain licenses for high-risk datasets.
Derivative Risk: AI outputs closely resembling copyrighted works can still trigger liability.
Non-commercial research is safer under fair use than commercial deployment.
PART 2: SYNTHETIC BIOLOGY PATENT PROTECTION
1. Overview
Synthetic biology combines biology, engineering, and AI to create:
Genetically modified organisms (GMOs)
Synthetic DNA sequences
Novel biochemical pathways
AI-designed proteins and enzymes
Patent protection encourages commercialization and innovation but faces challenges:
Patent eligibility: Laws of nature and natural DNA sequences are not patentable.
Novelty: Must demonstrate non-obviousness and utility.
AI involvement: AI-designed molecules must meet traditional patent criteria.
2. Key Case Laws
Case 1: Diamond v. Chakrabarty (1980)
Facts: A genetically engineered bacterium capable of breaking oil spills.
Decision: Supreme Court allowed patenting genetically modified organisms.
Implication: Synthetic biology inventions created by humans are patentable, even if based on natural biology.
Case 2: Association for Molecular Pathology v. Myriad Genetics (2013)
Facts: Myriad patented isolated BRCA1 and BRCA2 genes.
Decision: Naturally occurring DNA is not patentable, but cDNA (synthetic) is patentable.
Implication: Synthetic DNA or AI-designed sequences can be patented, but raw natural genes cannot.
Case 3: Mayo Collaborative Services v. Prometheus (2012)
Facts: Patents for metabolite-based drug dosing.
Decision: Claims based solely on natural correlations are not patentable.
Implication: Synthetic biology patents must include inventive human intervention, not just discoveries of natural laws.
Case 4: Amgen v. Sanofi (2017)
Facts: Dispute over monoclonal antibody patents.
Decision: Focused on the scope of patent claims and enablement for synthetic biological molecules.
Implication: Claims must be fully enabled, especially for AI-designed molecules.
Case 5: University of California v. Broad Institute (CRISPR Patent Dispute, 2016–2022)
Facts: Patent battle over CRISPR-Cas9 gene editing.
Decision: Courts analyzed inventorship and contribution.
Implication: AI-designed gene editing methods require clear inventorship attribution and detailed patent disclosure.
Case 6: Synthetic Biology Startups and AI Molecule Design (Recent 2020s)
Facts: AI-designed enzymes and proteins patented by startups.
Outcome: Courts and USPTO generally grant patents if molecules are human-designed and non-obvious.
Implication: AI can assist inventorship, but human inventors must still be identified.
3. Key Takeaways for Synthetic Biology Patents
| Factor | Principle |
|---|---|
| Natural vs Synthetic | Natural DNA cannot be patented; synthetic sequences can |
| AI-Designed Molecules | Patentable if human-assisted and non-obvious |
| Inventorship | Human inventors must be listed |
| Enablement | Detailed disclosure of synthetic biology process is required |
| Utility | Must demonstrate functional utility in biotech or medicine |
4. Combined Lessons
AI in copyright vs synthetic biology patents:
AI training datasets risk copyright infringement if output reproduces copyrighted works.
AI-designed synthetic biology inventions can be patented if human inventorship is clear and inventive step exists.
Licensing & Access:
AI datasets require licenses or fair-use justification.
Synthetic biology patents grant exclusive rights to license biotech and pharmaceutical applications.
Strategic Protection:
For AI: Use proper dataset licensing and output control.
For synthetic biology: Secure patent protection early, especially for AI-designed molecules.
RELATED Blog
comments