Guarding The Code: Evaluating Intellectual Property Frameworks For AI Algorithms

18th November, 2024 - 6:51 am Categories: Copyright 0 Comments

FROM TRADE SECRETS TO TRANSPARENCY (Introduction)

Artificial Intelligence (AI) has made far-reaching changes in innumerable industries, affecting critical decisions such as credit scoring, university admissions, and hospital discharges prompting the EU AI Act reflects these concerns by categorizing these applications as ‘high-risk’ and imposing strict requirements of transparency and accuracy on these applications.

As the role of AI rises, crucial legal and ethical apprehensions arise about how AI algorithms will be protected under IP law to determine “algorithmic accountability”. This question is particularly more pertinent in the EU (the pioneer) where current legal frameworks are still struggling to define appropriate protection frameworks. Should AI systems be considered as trade secrets, copyrights or patents? Clearing up these options will help to strike a fair balance between innovation and public interest.

Currently, AI systems are most often protected as trade secrets under the EU Trade Secrets Directive (EUTSD), which permits companies to circumvent the law to protect their proprietary algorithms and ensure their competitive edge. But that’s a conundrum; algorithmic opacity (lesser transparent working of algorithm) is enabled by trade secrets, which shields these systems in particularly sensitive fields like criminal sentencing or medical decision-making. For instance, we learned that the UK’s visa algorithm discriminated against people based on race, and yet trade secrets made these mistakes hazy and hard to challenge.

The problem with this growing reliance on trade secrets is that these laws weren’t made with complex technologies such as AI with ever-evolving algorithms in mind, but with static information that doesn’t make decisions on its own. Rise of autonomous algorithmic decisions has led to calls for “explainable AI” to ensure that these decisions are understandable and contestable, particularly as algorithmic decisions become more autonomous.

WHY COPYRIGHT IS AN “UGLY FIT” FOR ALGORITHMS (Main Blog) 

Another path under copyright law to protect AI systems is also available, but it’s circumscribed. The landmark case SAS Institute v World Programming Ltd (2012) confirmed this. The CJEU decided that only the source code is protected by copyright, not the method and algorithms underlying it, this is another instance stressing the complexity of algorithms as they struggle to satisfy the ‘author’s own intellectual creation’ standard set by copyright law, tending to appreciate more human-created expression. Thus, most types of AI including self-learning algorithms, are not covered by copyright protection. This is because self-learning algorithms are not static and evolve by themselves through new data inputs.  Having human creativity missing in algorithms and then often being perceived as purely functional (not serving creative expression standard) algorithms are not afforded within the copyright framework.

The Berne Convention specifically allows broader protection for algorithms as “scientific works” under the EU copyright framework; however, the Court of Justice of the European Union (CJEU) has consistently interpreted algorithms as technical systems and not creative works. It is unlikely that the functional processes of the algorithms can be protected under copyright. Moreover, the CJEU had ruled that technical conceptions of an algorithm such as code or logic are unlikely to satisfy copyright thresholds, so technical functionalities are inherently too complex to be encompassed in copyright law, which is rather focused on creative endeavors.

PATENTS AS PROTECTORS – AN IDEAL SAFEGUARD FOR ALGORTIHMS?

In particular, patents—specifically those protected as Computer Implemented Inventions (CIIs)—are the most promising route for protecting AI systems. Machine learning algorithms are now being accepted under patent law at the European Patent Office (EPO) as long as they meet the criteria of novelty, inventive step, and technical character. The international precedent for this approach is a well-known U.S. Supreme Court case, Diamond Diehr v. NASA. (1981), where it was determined that algorithms could be patented if they are integral to a larger process to produce a technical effect (practical application). This time it was an algorithm necessary to control a technical process (curing a rubber), providing a patentable reason for the algorithm to exist. Even though patents are a stronger IP right than trade secrets, they are faced with obstacles, for example, problems in terms of satisfying technical character (technical solution to a specific problem) requirements. Nevertheless, since the whole thing is publicly accessible, the patent system offers transparency and is unclouded on how AI technologies work, thereby enabling accountability for those responsible. The patent system stands out as a ‘good’ solution for AI protection and paves the way for innovation and transparency.

Additionally, China’s approach is intuitive, for it applies patent frameworks to allow the protection of AI algorithms through an “exchange of disclosure for protection”. It encourages developers to share their algorithmic details, clearly safeguarding innovation while encouraging transparency.

In India, the Telefonaktiebolaget LM Ericsson v. Intex Technologies (2015) case stressed trade secrets as a means of protection of algorithms. Software “per se” (software without technical application i.e. abstract programs without hardware relation) is explicitly excluded in sec 3(k) of Indian Patents Act thereby, it is not easy to obtain patent protection for algorithms. This means that in India, companies often resort to trade secrets to keep a “lid” on their proprietary algorithms rather than patents or copyrights. The exclusion of ‘mathematical methods’ and ‘algorithms’ from patentable subject matter under Section 3(k) of India’s Patents Act, 1970, undermines innovation in AI. While Indian courts, like in the Ferid Allani Vs Union of India case mirroring their European counterparts have allowed patentability for algorithms with technical applications beyond mere execution i.e a mathematical algorithm performing calculations to identify patterns in medical scans (application), though the patentability standards are markedly more stringent in India.

As in Indian market practices, proprietary algorithms in India are protected through trade secrets, this heavy reliance could bring in its own set of risks of creating a less transparent innovation ecosystem. Amending Section 3 (k) to provide for those AI-based inventions that have ‘technical effects’ (technical solutions to a specific problem) could make India better placed to compete in the global AI-driven world.

An expanded flowchart to help guide you towards picking the right IP protection for your AI algorithm – Copyright, Patent, Trade Secret, or Open Source Licensing. (Diagram Credits – Author) For Clearer Version – Here

THE CONUNDRUM OF SELF-LEARNING ALGORITHMS

Traditionally, patents are seen as the optimal avenue to protect technological innovations but falter in the face of self-learning algorithms. With the rise of machine learning, a significant portion of modern algorithms are self-learning and not static. An invention under European Patent Law must be novel, involve an inventive step yet self-learning algorithms, particularly those that use autonomous learning, frequently have difficulty satisfying these stringent requirements. In self-learning algorithms, the AI system itself might make changes without human intervention.

1. Technical Character

In the first place, a technical character requires that an invention represent a solution of a technical problem by technical means. Self-learning algorithms are self-directed and autonomously recognize patterns and make decisions autonomously without any explicit programming being required. Due to this autonomy, it is not possible to identify a problem that they technically resolve since their operations are not fully predictable and within the initial parameters. The ability of the algorithms to adapt and to modify themselves over time make it complicated to assess the technical contribution that they include at the date of the patent application. In the U.S. , it is difficult to patent self-learning algorithms, especially unsupervised algorithms, as it does not meet technical, inventorship, and disclosure requirements. These algorithms adapt autonomously, which cannot be pointed to any particular technical problem required by 35 U.S.C. § 101 of the United States. The adaptability complicates assessments of novelty and inventive steps since many of these self-learning techniques are based on “known mathematical models”.

1. Inventive Step

Secondly, the Inventive step requires that an invention is not obvious to a person who has ordinary skill in the art, and because many self-learning algorithms are built upon customary mathematical models and methods, it can prove difficult to show that they are non-obvious (innovative and not apparent in already established techniques). In addition, open-source algorithms and public datasets are used to counter such novelty and inventiveness claims. A USPTO report revealed that the number of companies currently filing for patents in AI drug discovery is over 135, Nevertheless, the stringent inventorship regime in the U.S. like the EU remains an impediment for a few innovators who resort to trade secrets as a prospective protection.

The AI can now make transformative amendments itself, going beyond its primitive programming to making intelligent and innovative outcomes thanks to self-learning algorithms. This evolution blurs the lines of inventorship, raising critical questions: Who should own the copyright—the developer, or maybe it’s the AI itself, or even a human-AI team? Notably, scholars like Ryan Abbott argue that AI should be recognized as a legitimate inventor in cases where its contributions go beyond passive tools to active, independent creators, thus challenging traditional concepts of intellectual property and inventorship rights.

CAN AI BECOME A CO-INVENTOR? (Conclusion)

Professor Ryan Abbott leads The Artificial Inventor Project aiming to recognize and register AI as inventors and co-inventors. At the heart of this matter is DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), an AI system applying for patents for innovations including a fractal-based food container and an emergency light signal to aid search and rescue.

The project has received widely varied response internationally. The South African system does not conduct a detailed inventorship examination, and the country became the first in the world to grant a patent listing DABUS as an inventor. The initial ruling by Australia’s Federal Court finding for DABUS’s recognition was overturned by the Full Federal Court in Australia. On the other side, the United Kingdom and the United States patent office believe that the inventors have to be ‘natural persons’ which means that AI can’t be the inventor.

But IP law is  attempting to integrate AI inventions, led by the European Patent Office (EPO). Its President António Campinos highlighted the need for legislative reforms that recognize the role of AI in producing inventions, yet with human oversight, a feature of a “sui generis” (category of its own) approach. Notwithstanding that, in J 8/20 (DABUS Case) EPO still ruled an AI system such as DABUS cannot be an inventor.

This ever-evolving landscape underscores the potential for a unique legal framework to recognize AI’s contributions and illustrates how differing international IP systems can shape global innovation outcomes. Of course, there’s got to be a responsible use of AI, and that includes suitable IP protection frameworks for that. Futurist Alvin Toffler once pertinently wrote: ‘The illiterate of the 21st century will not be those who cannot read and write, but those who cannot learn, unlearn, and relearn’ – and so the need for adaptable legal structures in artificial intelligence’s evolving landscape is clear.

Author: Rudraditya Singh Panwar, A  Student at Rajiv Gandhi National University of Law, in case of any queries please contact/write back to us via email to chhavi@khuranaandkhurana.com or at Khurana & Khurana, Advocates and IP Attorney.

References As Highlighted In The Blog (From Start to End)

1. https://artificialintelligenceact.eu/annex/3/
2. https://www.cambridge.org/core/books/abs/cambridge-handbook-of-the-law-of-algorithms/understanding-transparency-in-algorithmic-accountability/D355F8D31BF1778431D92D2E79917093
3. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32016L0943
4. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3388639
5. https://www.foxglove.org.uk/2020/08/04/home-office-says-it-will-abandon-its-racist-visa-algorithm-after-we-sued-them/
6. https://www.deloitte.com/nl/en/services/consulting/perspectives/a-call-for-transparency-and-responsibility-in-artificial-intelligence.html
7. https://eur-lex.europa.eu/legal-content/en/TXT/?uri=CELEX:62010CJ0406
8. https://eur-lex.europa.eu/legal-content/en/ALL/?uri=CELEX:62009CJ0393
9. https://www.cambridge.org/core/journals/asian-journal-of-international-law/article/copyright-protection-for-aigenerated-works-exploring-originality-and-ownership-in-a-digital-landscape/12B8B8D836AC9DDFFF4082F7859603E3
10. https://eur-lex.europa.eu/legal-content/EN/ALL/?uri=CELEX%3A32009L0024
11. https://www.hlk-ip.com/revised-epo-guidelines-ai-and-computer-implemented-inventions-emphasis-on-comvik-and-technical-detail/
12. https://www.marks-clerk.com/insights/articles/patenting-ai-the-epos-new-guidelines/
13. https://supreme.justia.com/cases/federal/us/450/175/
14. https://www.herbertsmithfreehills.com/insights/2023-03/the-ip-in-ai-can-ip-rights-protect-ai-systems
15. https://www.epo.org/en/boards-of-appeal/decisions/t170697eu1
16. https://drpress.org/ojs/index.php/fcis/article/view/5210/5045
17. https://www.wipo.int/wipolex/en/judgments/details/2173
18. https://www.wipo.int/ip-development/en/agenda/flexibilities/details.jsp?id=8825
19. https://www.scconline.com/blog/post/2023/05/17/need-to-reconsider-exclusions-u-s-3k-of-the-patents-act-in-view-of-growing-innovations-delhi-high-court-legal-news/
20. https://indiankanoon.org/doc/90686424/