Caspar Chen
Edited by Kira Small, Kolby Phillip, Mac Kang, and Jia Lin
The launch of ChatGPT has drawn immense attention to Artificial Intelligence (AI). AI can be traced back to the 1950s with the discovery of machine learning algorithms and the proposal of the Turing Test, a test of a machine's ability to simulate human intelligence and thoughts [1]. In the following decades, with advancements in computing power enabling data-driven AI, AI has evolved closer to passing the Turing Test, signifying an increasing ability for AI to conduct tasks once considered to be uniquely human. This fast growth of AI begins to raise concern among people as the maturation of AI may directly challenge conventional rules and structures of many existing fields. In the field of intellectual property law, for example, the ability of AI to generate inventions challenges traditional definitions of inventorship, raising the question: can AI be granted inventorship?
AI’s inventive capability stems from its foundation in machine learning and data analysis. Depending on the specific algorithm, each machine learning strategy generally will use mathematical calculations to train inputting data, find patterns in data, and verify the pattern’s accuracy. To increase accuracy, people frequently need to optimize and adjust the inputs and weights until they are satisfied with the model’s output. When the machine learning models reach certain accuracy scores, they will be used to predict and generate possible data to predict future patterns. The most popular machine learning method is neural networks. Neural networks are similar to the human brain: the input assembles dendrites, passes several hidden layers with set criteria like passing axon, and generates output at the axon terminals [2]. By adding more hidden layers, neural networks can process complex tasks with large input features such as facial recognition, known as deep learning [3]. When the input data becomes large enough – to train a large language model like ChatGPT would need data from nearly all online open sources, for example– AI can construct a model with enough complexity and data to generate responses to all types of questions, known as Generative-AI. When tasking to invent, AI can extract from its data and compile it into inventions through calculations.
While the above explanation may portray AI as a powerful and effective inventor, in the current U.S. legal landscape, there has not been a case that successfully grants AI inventorship. In Thaler v. Vidal (2022), Stephen attempted to patent his AI system DABUS for its independent invention of a fractal drink container and a lighting device [4]. Both The United States Patent and Trademark Office (USPTO), the U.S. District Court for the Eastern District of Virginia, and the United States Court of Appeals for the Federal Circuit ruled against Stephen with affirming fundamental reasons that reject AI inventorship: “the Patent Act requires an ‘inventor’ to be a natural person” [5]. The use of “natural person” highlights the nature of patents being to reward the creativity of humans. Although AI is meant to mimic human creativity, the Court believed it should not enjoy patentability, for its creativity derives from mathematical computation. Moreover, the term “natural person” reinforces a current limitation of AI: its lack of responsibility and liability. As required by 35 U.S.C. § 115, inventors should make an oath and declaration to affirm their originality of inventions and take legal duty with it [6]. However, for AI, the legal responsibility of inventors becomes unclear. For instance, who should receive the economic benefit that AI’s inventions bring about? If there is a negative impact, who should be held accountable ? Similar challenges arise in patent infringement cases. As previously described, AI’s inventive ability comes from mathematical calculations. Courts may not be able to question AI in litigation and thus struggle to decide whether its invention commits infringement with limiting testimony.
Another reason for rejection centers on the difficulty in determining the extent to which AI independently makes inventions. Different from the common patent of computer-assistance invention, where the contribution of AI can be quantified and proved, it is challenging to determine the extent to which AI invent independently the reliability of AI’s independent inventing process is in doubt. As the USPTO’s report regarding AI shows, people believe AI has a long distance till it attains intelligence akin to humans [7]. This question on the inventing ability of AI can be traced back to the mechanism of AI training. As mentioned, the production of a mature AI requires frequent adjustment and optimization with human intervention. Consequently, it is questionable whether AI becomes skillful enough to invent without further human help, bringing transparency concerns to the Courts.
To be more clear, no case in the U.S. has granted AI as a joint inventor either. Patenting computer-assistance (or computing algorithms involved) inventions only view it as a tool. In Diamond v. Diehr (1981), the Supreme Court ruled in favor of James, patenting his invention of the process for curing synthetic rubber using a computer-based algorithm [8]. The Court’s decision demonstrates that the use of computing processes to assist applications in specific industrial situations may be eligible for a patent. This shows the greatest difference between AI-generative inventions and computer-involved inventions: patent eligibility is only granted to computing systems used as a tool, not as independent creative innovators.
However, the exclusion of AI as an inventor can cause its own problems. By eliminating the patent eligibility of AI, other people with certain relations to AI—owner, developer, or user—may otherwise claim ownership of inventions. Their contribution to the inventing process may be minimal, but they could enjoy the lump financial benefit brought by the invention. For example, in Thaler v. Vidal (2022), what if Stephen claimed himself as the inventor? His contribution was only limited to coding and training AI. Even worse, given the striking computation ability and durability of AI, its inventing speed may be fast enough to cause patent floods that dramatically increase the numbers of patents in a field, leading individuals or companies to form a monopoly with a large number of patents in various fields.
The ban of AI inventorship may also decelerate the pace of innovation. Without protection for AI’s inventions, companies will be less supportive of using and training AI for complex analysis and calculation, causing a longer inventing time. As Abraham Lincoln stated, the patent system will add “the fuel of interest to the fire of genius” [9]. The ban of AI inventorship may otherwise deviate from the purpose of patents.
As a controversial and novel technology, AI lacks legal regulation on its potential and usage. The unclear position of AI in patents suggests a policy lag and space for change in law. As of October 2024, no case regarding AI has been accepted for review by the Supreme Court. The majority of the general public knows little about AI; no one at all knows the future of AI. As AI improves, laws clarifying its roles in invention and other intellectual processes will become more and more necessary.
[1]University of Washington. Machine Learning Exhibit. https://courses.cs.washington.edu/courses/cse490h1/19wi/exhibit/machine-learning-1.html
[2]Theobald, Oliver. Machine Learning For Absolute Beginners: A Plain English Introduction (Second Edition). Machine Learning for Beginners Book 1.
[3]Theobald, Oliver. Machine Learning For Absolute Beginners: A Plain English Introduction (Second Edition). Machine Learning for Beginners Book 1.
[4]Holland & Knight. Making the Case for AI Inventorship. June 2022. https://www.hklaw.com/en/insights/publications/2022/06/making-the-case-for-ai-inventorship
[5]Thaler v. Vidal, Federal Circuit Court Opinion. Aug. 5, 2022. https://cafc.uscourts.gov/opinions-orders/21-2347.OPINION.8-5-2022_1988142.pdf
[6]U.S. Patent and Trademark Office. Manual of Patent Examining Procedure (MPEP) §602. https://www.uspto.gov/web/offices/pac/mpep/s602.html
[7]U.S. Patent and Trademark Office. Public Views on Artificial Intelligence and Intellectual Property Policy Report. Oct. 7, 2020. https://www.uspto.gov/sites/default/files/documents/USPTO_AI-Report_2020-10-07.pdf
[8]Diamond v. Diehr, 450 U.S. 175. https://supreme.justia.com/cases/federal/us/450/175/
[9]Abraham Lincoln and the Patent System. https://www.abrahamlincolnonline.org/lincoln/education/patent.htm
Comments