Procedural Fairness In Epo Examination Of Ai Algorithm Claims.
I. Meaning of Procedural Fairness at the EPO
Procedural fairness in EPO examination is primarily governed by:
Article 113(1) EPC – Right to be heard
Article 113(2) EPC – Decisions bound by text submitted or agreed by applicant
Article 116 EPC – Right to oral proceedings
Rule 111(2) EPC – Reasoned decisions
Article 117 EPC – Taking of evidence
Procedural fairness ensures:
Applicants are heard before adverse decisions.
Decisions are reasoned and transparent.
Applicants can amend claims.
Examination standards are applied consistently.
Evidence and objections are properly substantiated.
In AI algorithm cases, fairness is particularly sensitive because:
AI inventions often mix technical and non-technical features.
The “technical character” requirement can be complex.
Inventive step analysis under the COMVIK approach requires careful separation of technical and non-technical features.
II. Key Legal Framework for AI Algorithm Claims
AI claims are examined under:
Article 52 EPC – Patentable inventions
Article 56 EPC – Inventive step
Article 83 EPC – Sufficiency of disclosure
Article 84 EPC – Clarity
Article 123(2) EPC – Added subject-matter
Algorithms per se are excluded under Article 52(2)(c) (mathematical methods), unless they provide a technical effect.
The Boards of Appeal have developed key jurisprudence that shapes procedural fairness in AI examination.
III. Leading Case Laws on Procedural Fairness and AI/Algorithm Claims
1. T 641/00 – COMVIK (Two Identities/COMVIK)
Principle Established:
The “COMVIK approach” for mixed inventions (technical + non-technical features).
Facts:
The invention concerned a method involving mobile phone identification with technical and administrative aspects.
Legal Holding:
Only technical features contribute to inventive step.
Non-technical features (e.g., business rules, abstract algorithms) can appear in the claim but do not contribute to inventive step unless they produce a technical effect.
The problem-solution approach must include non-technical features as constraints given to the skilled person.
Relevance to AI:
AI algorithms often contain:
Mathematical training steps
Data classification rules
Business logic components
Under COMVIK:
Pure algorithmic features are treated as non-technical unless tied to technical implementation.
The examining division must clearly identify which features are technical.
Procedural Fairness Impact:
The EPO must explicitly separate technical from non-technical features.
Failure to properly reason this separation violates Rule 111(2) EPC (reasoned decision).
Applicants must be allowed to argue the technical contribution (Art. 113 EPC).
COMVIK remains foundational for AI inventive step examination.
2. T 258/03 – Auction Method/Hitachi
Principle:
A method involving technical means is not excluded under Article 52(2), but inventive step still requires technical contribution.
Facts:
An electronic auction method implemented on a computer.
Decision:
Any claim involving technical means avoids exclusion under Art. 52(2).
However, inventive step must be based on technical features.
Importance for AI:
Many AI claims:
Recite a computer
Recite a processor
Recite storage medium
Under Hitachi:
Merely implementing AI on a computer does not guarantee inventive step.
The technical contribution must go beyond generic computer implementation.
Procedural Fairness Angle:
Examiners must not reject AI claims solely as “excluded subject-matter” if technical means are present.
If inventive step is denied, proper reasoning must be provided.
Applicants must be informed which features lack technical effect.
Failure to follow this framework may violate Article 113(1) EPC.
3. T 1227/05 – Circuit Simulation I/Infineon
Principle:
A mathematical method can be patentable if it produces a technical effect.
Facts:
Simulation of electronic circuits subject to noise.
Holding:
The simulation method had technical character because:
It served a technical purpose (circuit design).
It reflected physical reality.
Importance for AI:
This case is critical for AI-based simulations and modeling:
AI used for physical system modeling
Neural networks simulating chemical reactions
AI in engineering optimization
The Board recognized:
A mathematical method is patentable if linked to a technical purpose.
Procedural Fairness Significance:
Examiners must:
Assess whether the AI method serves a technical purpose.
Avoid blanket categorization as “mathematical method.”
Applicants must be allowed to demonstrate technical application.
This case protects applicants from overly rigid exclusion.
4. G 3/08 – Programs for Computers
Principle:
Clarification of when computer programs are patentable.
Holding:
The Enlarged Board confirmed:
Technical effect beyond normal computer operation is required.
No radical departure from existing jurisprudence.
Relevance to AI:
AI software:
Must produce technical effect beyond data processing.
Training methods must solve technical problems.
Procedural Fairness Aspect:
The Enlarged Board emphasized:
Legal certainty.
Consistency in interpretation.
This ensures:
AI applicants are examined under established principles.
No arbitrary shift in examination practice.
5. T 1358/09 – Data Retrieval
Principle:
Information modeling per se is non-technical.
Relevance:
Many AI inventions involve:
Data structures
Information classification
Labeling systems
The Board held:
Cognitive content does not confer technical character.
Technical implementation must solve a technical problem.
Procedural Fairness:
Examiners must:
Distinguish cognitive content from technical implementation.
Provide detailed reasoning for inventive step refusal.
Failure to reason properly violates Rule 111(2) EPC.
6. T 161/18 – Classification Method
Highly Relevant AI Case
Facts:
Claim directed to a method for classifying data using machine learning.
Decision:
The Board held:
A machine learning method is not technical merely because it uses mathematical steps.
If classification serves a technical purpose (e.g., controlling a machine), it may contribute to inventive step.
Key Takeaway:
Technical application is decisive.
Procedural Fairness Significance:
The Board carefully analyzed each feature.
It did not dismiss AI claims categorically.
The applicant’s arguments were fully addressed.
This demonstrates proper application of Article 113 EPC.
7. T 0423/11 – Diagnostic Method
Importance:
AI frequently used in medical diagnostics.
The Board emphasized:
Technical character depends on technical implementation.
Mere automation of mental acts is not enough.
Procedural Aspect:
The Board required:
Clear reasoning.
Proper application of COMVIK.
AI diagnostic claims must show technical effect (e.g., improved imaging apparatus operation).
8. T 0704/12 – Right to Be Heard
Principle:
Violation of Article 113(1) occurs if decision based on grounds not communicated.
Application to AI:
If an examining division:
Introduces new reasoning at decision stage
Reclassifies features as non-technical without prior notice
It violates procedural fairness.
This case reinforces:
AI applicants must be informed of classification of features.
Surprise reasoning is impermissible.
IV. Procedural Fairness in Practical AI Examination
1. Proper Identification of Technical Features
Examiners must:
Explicitly identify technical contribution.
Avoid vague statements like “mere algorithm.”
2. Opportunity to Amend
AI claims often require:
Reformulation emphasizing technical effect.
Limiting to specific hardware implementation.
Article 123 EPC allows amendments; refusal without opportunity may violate fairness.
3. Sufficiency (Article 83 EPC)
AI training claims must:
Disclose training data structure.
Provide reproducibility.
However, overly strict standards may violate proportionality.
4. Clarity (Article 84 EPC)
Terms like:
“Trained model”
“Optimization”
“Neural network”
Must be assessed reasonably, not arbitrarily.
V. Common Procedural Pitfalls in AI Examination
Blanket rejection as mathematical method.
Failure to apply COMVIK properly.
Ignoring technical purpose.
Inadequate reasoning.
Surprise objections in final decision.
Boards of Appeal frequently overturn such decisions.
VI. Conclusion
Procedural fairness in EPO examination of AI algorithm claims requires:
Strict adherence to Articles 113 and 111 EPC.
Correct application of COMVIK (T 641/00).
Careful distinction between mathematical methods and technical applications.
Full reasoning in refusal decisions.
Opportunity to respond and amend.
Key cases shaping this area include:
T 641/00 (COMVIK)
T 258/03 (Hitachi)
T 1227/05 (Infineon)
G 3/08 (Computer Programs)
T 1358/09
T 161/18
T 0423/11
T 0704/12
RELATED Blog
comments