AI-based business models (hereinafter BM) once again received a lot of attention as the AI language model ChatGPT3 was released. In particular, the whole world was seething beyond the industry when Open AI introduced ChatGPT4 not long after the ChatGPT3. How can the intellectual property rights of AI-based BM be protected? In this issue, we will introduce research on IP protection of AI-based BM.
 |
| clipartkorea.co.kr |
Intellectual property can be protected in both formal and informal ways. Formally it’s protected by patents, design, trademarks, and copyrights. Informally, trade secrets, strengthening product and complexity of manufacturing process to increase the difficulty of imitation, and lead-time advantage to secure competitive advantage through faster innovation than competitors. This study examines how formal and informal protection strategies can be applied to AI-based BM, with an emphasis on balancing open innovation with existing practices. Depending on the BM, optimization and supplementation of formal and informal IP protection strategies may be necessary to maximize value creation.
Challenges in applying a formal IP strategy
One of the major challenges for patent protection of AI-based inventions is that algorithms play an important role in designing AI concepts. Algorithms are considered mathematical methods “in their own right” under patent law and are therefore excluded from patentability. AI concepts are often aimed at automating or performing tasks or activities currently performed by the human mind, which fall under patent ineligibility because they are mere theoretical concepts or lack novelty. However, many AI-based inventions are currently based on and implemented in software, and over the past few decades, patent law and practice have been building around how software-based inventions are dealt with.
In Europe, AI as a mathematical method is exempt from patent, but if the method involves a technological means (e.g. a computer) or a device, it may be of a technical nature as a whole (computer-implemented inventions, CII). The European Patent Office (EPO) applies a 'two-hurdle approach' to 'Mixed-type inventions' when evaluating patentability to see if AI methods contribute to the technical characteristics of the invention. In this context, the EPO recently updated its screening guidelines with a specific section on ‘Artificial Intelligence and Machine Learning’. In this context, the EPO recently updated its screening guidelines with a specific section on ‘Artificial Intelligence and Machine Learning’. These Guidelines provide guidance on how to assess whether inventions relating to artificial intelligence and machine learning are based on the 'technical characteristics' required to be patentable, and provide detailed information about how to evaluate relevant cases and the CII's technological prowess as determined by the EPO Appeals Board.
US
In the United States, abstract ideas cannot be patented. Also, just using a computer to implement an abstract idea to inventions is not enough to qualify for a patent. Law firm Baker McKenzie explains that perhaps the biggest legal hurdle to patenting AI inventions in the US is §101 US Federal Regulations (35 U.S.C.). This is because patentable subjects are limited to ‘process, machine, manufacturing, or composition of matter’, and abstract ideas, natural laws, and natural phenomena are excluded from patentability. The criteria for eligibility for these patents were further strengthened in the 2014 U.S. Supreme Court decision for Alice Corporation vs CLS Bank, which applied a more stringent two-step test for software and computer-implemented inventions.
Current status of AI-related innovation patents
AI-related innovations are often based on software and computer-implemented inventions. These inventions may be directed towards one or more specific AI applications. Based on previous and current laws, various companies and research institutes have begun to apply for patents in the field of AI as well. As shown in the figure below, approx.. 340,000 family patents have been filed and published since the 1960s. Additionally, by mid-2018, a total of more than 1.5 million scientific publications had been published, indicating that AI has become a major field of science. By the early 2000s, scientific publications had grown significantly (nearly doubling at an average annual growth rate of 18% from 2002 to 2007), but it took another decade for patent applications to skyrocket (CAGR of 28% from 2012-2017). It is reasonable to interpret that this is because basic research is usually published as scientific publications first, whereas R&D related to industrial use takes a considerable amount of time and, leads to patent applications by its nature.
 |
| Trend in the number of AI family patent/ scientific publications by year of first publication (WIPO 2019) |
As seen in the figure below, patent applications
for specific application field have emerged since the mid-1990s. Those are mainly
transportation and communication field, and artificial intelligence-related
inventions are constantly being filed across multiple application fields.
 |
| Trend in the number of family patents with the earliest priority year by application field (WIPO 2019) |
IP protection strategy for AI-based BM
AI research and innovation requires significant investment. The European Union (EU) aims to invest at least EUR20 billion per year in AI after 2020 (Euro Commission 2018). According to the World Intellectual Property Organization (WIPO), more than 3,000 AI-related companies have received funding worth $46 billion, and M&A has become a means of securing AI technologies, data access, and related patent portfolios. Nearly 500 companies have been acquired, more than half of them have been happened since 2016.
Given the high level of investment in AI technologies and applications, companies and investors use a variety of strategies for protecting AI-related intellectual property to protect these investments and generate returns. These strategies include formal IP measures, such as patents and copyrights for AI algorithms and codes, and informal IP measures, such as trade secrets for AI data.
 |
| Formal/ Informal protection |
In addition, ③ there is a method based on standardization through disclosure (Public Domain). There are two main challenges when developing AI technologies. (a) developing AI systems and algorithms from a technical perspective; and (b) accessing suitable data sets to train optimized AI algorithms or AI systems. Access to data sets is an important issue not only for competitive advantage across legal systems, but also for investors investing in startups. Issues include available datasets, costs associated with data quality, and more. In public institutions, access to data is more difficult due to budget constraints or data protection regulations. Especially in the life sciences, where regulations are more stringent, accessing data can be more difficult. In other words, companies can collect/expand datasets while accelerating development by utilizing standardized open sources in AI R&D, and gain a competitive edge by simultaneously using official/unofficial IP protection strategies.
Implication
There are various studies that suggest that formal/informal protection strategies are complementary to each other in AI-based BM. Since the current system mainly regards AI as software and clearly stipulates the criteria for patent registration applying AI algorithms, legal, practical, and ethical challenges remain regarding the patent protection of AI-based methods and systems.
--------------------------------------------------------------------------------------------------------
< Source >
The above is excerpted from;
Patent Management: Protecting Intellectual Property and Innovation 2021, Oliver Gassmann, Martin A. Bader, Mark James Thompson, Springer.
No comments:
Post a Comment