Patent Frameworks For Decentralized Neural Networks Supporting Collective Intelligence.

21 Mar 2026 --
0 Comments

1. Decentralized Neural Networks and Collective Intelligence

A decentralized neural network (DNN) is a neural network where computation, data storage, or model updates are distributed across multiple nodes rather than centralized in a single server. These networks often leverage concepts like federated learning or blockchain-based coordination.

Collective intelligence refers to the emergent intelligence created when multiple agents collaborate or aggregate knowledge.
Combining DNNs and collective intelligence allows multiple participants (human or machine) to train models, share insights, and make decisions without a central authority.

Key patent issues:

Patent Eligibility: Is the invention “abstract” or “technical”? Courts often scrutinize AI and algorithms under patent law, especially in the U.S. under 35 U.S.C §101.
Novelty and Non-Obviousness: Decentralized neural networks must show a technical improvement over existing systems.
Ownership and Inventorship: Who owns an invention when multiple decentralized nodes contribute? This is complicated when contributors are autonomous AI agents.

2. Patent Frameworks

Patenting DNNs supporting collective intelligence can follow several frameworks:

Utility Patents: Protect the functional aspect of the DNN algorithms, data flow, and network coordination.
System Patents: Cover the architecture of the decentralized network, e.g., how nodes communicate, consensus mechanisms, or federated learning protocols.
Method Patents: Protect specific processes for aggregating collective intelligence, such as secure model averaging or incentive mechanisms for participation.
Software Patents: Focus on the software implementations of AI processes, though these are scrutinized for abstractness in many jurisdictions (e.g., U.S., Europe).

3. Key Case Laws in AI, Neural Networks, and Decentralized Systems

Here’s a detailed analysis of more than five landmark or relevant cases:

Case 1: Alice Corp. v. CLS Bank International (2014) – U.S. Supreme Court

Citation: 573 U.S. 208 (2014)
Summary: The Supreme Court ruled that abstract ideas implemented on a computer are not patentable unless they include an “inventive concept” that transforms the idea into a patent-eligible application.

Relevance:

Many DNN patents face scrutiny under Alice. Decentralized AI patents must show more than just “running a neural network on multiple computers.”
For instance, a patent that merely distributes computation without improving speed, accuracy, or security may be invalidated.

Case 2: Berkheimer v. HP Inc. (2018) – U.S. Federal Circuit

Citation: 881 F.3d 1360 (Fed. Cir. 2018)
Summary: Clarified that patent eligibility under §101 can involve factual determinations. If a patent describes a concrete improvement, it can survive Alice scrutiny.

Relevance:

A decentralized neural network with a unique way to combine node outputs securely could be patentable.
Supports patents claiming specific methods for collective intelligence aggregation.

Case 3: Enfish, LLC v. Microsoft Corp. (2016) – U.S. Federal Circuit

Citation: 822 F.3d 1327 (Fed. Cir. 2016)
Summary: The court found that software can be patent-eligible if it improves computer functionality itself rather than an abstract idea.

Relevance:

A DNN improving computational efficiency across nodes, reducing latency, or enhancing learning rates may be patentable under this reasoning.

Case 4: Diamond v. Diehr (1981) – U.S. Supreme Court

Citation: 450 U.S. 175 (1981)
Summary: Held that implementing a mathematical formula in a physical process (like curing rubber) can be patentable.

Relevance:

Analogous for neural networks: a decentralized neural network that improves collective decision-making in a tangible system (like IoT or distributed robotics) may qualify as patentable.

Case 5: Voter v. Google LLC – hypothetical federal court references

Summary: While no direct case exists yet about decentralized AI, courts have considered joint inventorship issues. When multiple nodes or contributors participate in neural network training, patent law may need inventorship attribution.

Key Takeaway:

Under 35 U.S.C §116, each inventive contribution must be acknowledged. The question arises: can autonomous AI agents be “inventors”? Most courts currently say no; a human must be listed as an inventor.

Case 6: European Patent Office (EPO) Guidelines – G 1/19 (2021)

Summary: AI-generated inventions cannot be named with an AI as an inventor. Human oversight is required.

Relevance:

For decentralized neural networks generating inventions collectively, human operators or designers must be listed as inventors for patent eligibility.

4. Synthesis of Legal Principles for DNN Patents

Patentability requires:
1. Technical contribution (speed, accuracy, security).
2. Concrete implementation (not abstract).
3. Clear inventorship (human contributors).
4. Non-obvious aggregation of collective intelligence.
Best Practices:
- Document how the decentralized network improves over conventional centralized models.
- Show specific technical improvements in data processing or model training.
- Clearly define human roles in design, training, or system coordination.

5. Practical Example of a Patentable Claim

“A system for decentralized neural network training comprising:
multiple nodes configured to train local models;
a secure aggregation protocol that weights contributions based on data quality;
a consensus mechanism that verifies model integrity;
wherein the system collectively improves prediction accuracy compared to conventional centralized training methods.”

This type of claim focuses on technical improvements, decentralization, and collective intelligence, which aligns with the case law principles above.