Excluding Artificial Neural Networks from Patentability under s1(2) of the Patents Act 1977

Comptroller General of Patents, Designs and Trade Marks v Emotional Perception AI Ltd ([2024] EWCA Civ 825)

Introduction

The case of Comptroller General of Patents, Designs and Trade Marks v Emotional Perception AI Ltd ([2024] EWCA Civ 825) represents a pivotal moment in the intersection of artificial intelligence (AI) technology and intellectual property law. The appellant, the Comptroller General, sought to overturn a High Court decision that had previously allowed Emotional Perception AI Ltd ("EPL")'s patent application concerning an AI-based system for media file recommendations. The central legal question revolved around whether the patented invention, which utilizes artificial neural networks (ANNs), fell within the exclusion from patentability outlined in section 1(2) of the Patents Act 1977.

Summary of the Judgment

The England and Wales Court of Appeal delved into whether an ANN, fundamental to EPL's invention, constitutes a "program for a computer" under s1(2)(c) of the Patents Act 1977 and thus is excluded from patentability "as such." Lord Justice Birss held that ANNs, regardless of their implementation through hardware or software, indeed qualify as computer programs. Consequently, the exclusion applies, leading to the upholding of the Hearing Officer's decision to reject the patent application. The court emphasized that the semantic aspects of file recommendations provided by EPL's system were of an aesthetic and cognitive nature, lacking a technical effect required to escape the exclusion.

Analysis

Precedents Cited

The judgment extensively referenced landmark cases that have shaped the UK's patent law regarding computer-implemented inventions:

• Aerotel Ltd v Telco Holdings Ltd [2007] RPC 7: Established a four-stage approach to determine patentability, focusing on technical contributions beyond excluded subject matter.
• AT&T Knowledge Ventures v Comptroller [2009] EWHC 343 (Pat): Introduced five signposts to assess the technical contribution of computer programs.
• Halliburton Energy Services [2011] EWHC 2508 (Pat): Highlighted the importance of the specific tasks performed by a program in determining patent eligibility.
• Protecting Kids the World Over [2011] EWHC 2720 (Pat): Demonstrated that technical superiority in monitoring electronic communications could render an invention patentable.
• HTC v Apple [2013] EWCA Civ 451: Endorsed and slightly recast the five signposts from AT&T Knowledge Ventures, reinforcing the assessment framework for technical contributions.

Additionally, the Court of Appeal referred to European Patent Office (EPO) decisions, such as Mitsubishi/Sparsely connected neural network (T 702/20), aligning the UK approach with broader European standards.

Legal Reasoning

The court employed the Aerotel four-stage approach to dissect the patentability of EPL's ANN-based system:

1. Properly construe the claim: The claims were understood to encompass both hardware and software implementations of ANNs.
2. Identify the actual contribution: The contribution was recognized as the training of an ANN to align property vectors with semantic distances, enhancing file recommendation accuracy.
3. Assess exclusionary grounds: The court scrutinized whether the contribution fell solely within excluded subject matter, focusing on the ANN's role.
4. Determine if the contribution is technical: It was evaluated whether the technical effect extended beyond mere data presentation or cognitive aspects.

Central to the judgment was the determination that the ANN's weights and biases constitute a computer program. Lord Justice Birss interpreted "computer program" as "a set of instructions for a computer to do something," irrespective of whether these instructions are generated by humans or through machine learning processes. This interpretation aligns with both traditional dictionary definitions and the jurisprudential stance of UK courts and the EPO.

The court rejected EPL's arguments that the ANN's ability to learn autonomously via training processes differentiated it from conventional computer programs. It was held that the origin of the program's instructions, whether human-written or machine-generated, does not exempt it from being classified under s1(2)(c) exclusions.

Furthermore, the court emphasized that the semantic qualities of the file recommendations—being aesthetic and cognitive in nature—do not qualify as technical effects. This lack of technical grounding means the invention does not meet the necessary criteria to escape the exclusionary provisions.

Impact

This judgment has profound implications for the patent landscape, particularly concerning AI and machine learning innovations:

• Clarification of Exclusions: Reinforces that ANNs, regardless of their implementation, are considered computer programs and are thus excluded from patentability "as such."
• Technical Contribution Requirement: Highlights the necessity for inventions to demonstrate a clear technical effect beyond data presentation or cognitive enhancements to be patentable.
• Scope of AI Patents: Signals increased scrutiny on AI-based inventions, ensuring that merely aesthetic or cognitively driven functions do not qualify for patents unless accompanied by a substantive technical contribution.
• Alignment with EPO Standards: Ensures consistency between UK and European patent law regarding computer-implemented inventions, providing clearer guidelines for applicants operating across jurisdictions.

Future patent applications involving ANNs will need to meticulously demonstrate technical advancements beyond algorithmic or data-driven enhancements to achieve patentability.

Complex Concepts Simplified

Artificial Neural Networks (ANNs)

An Artificial Neural Network (ANN) is a computing system inspired by the biological neural networks of animal brains. It consists of interconnected units called "neurons" organized in layers. Each neuron processes input data, applies specific weights and biases, and passes the output to subsequent layers. ANNs are trained using large datasets to recognize patterns and make decisions, which makes them essential in modern AI applications like image and speech recognition.

Section 1(2) of the Patents Act 1977

Section 1(2) of the Patents Act 1977 outlines exclusions from patentability. Specifically, it states that certain categories, including "a program for a computer as such," are not considered inventions eligible for patents. This exclusion is intended to prevent abstract ideas, mathematical methods, and purely cognitive processes from being patented.

Aerotel Four-Stage Approach

The Aerotel four-stage approach is a methodological framework used to assess the patentability of computer-implemented inventions:

• Stage 1: Properly construe the claims to understand the scope and technical features of the invention.
• Stage 2: Identify the actual (or alleged) contribution of the invention over the prior art.
• Stage 3: Determine if the contribution lies solely within excluded subject matter.
• Stage 4: If the contribution extends beyond exclusions, assess whether it has a technical character.

Technical Effect

A technical effect refers to a tangible and measurable outcome that results from the functioning of an invention. In the context of patent law, an invention must produce a technical effect to qualify for patent protection. Effects that are purely abstract, aesthetic, or cognitive are insufficient to confer patentability.

Conclusion

The ruling in Comptroller General of Patents, Designs and Trade Marks v Emotional Perception AI Ltd underscores the stringent criteria applied to AI-based inventions within the UK patent framework. By affirming that ANNs fall under the exclusion of "computer programs as such," the court has set a clear boundary, mandating that only AI systems demonstrating substantial technical contributions beyond data processing and cognitive functions can aspire to patent protection. This decision not only aligns UK patent law with European standards but also serves as a critical reference point for innovators and legal practitioners navigating the complex terrain of AI patentability.

Moving forward, developers and companies engaged in AI and machine learning must ensure that their inventions offer unequivocal technical advancements to secure patent rights. This judgment acts as a clarion call to deepen the technical aspects of AI innovations, ensuring they transcend abstract methodologies to achieve meaningful industrial applicability.