Preview: The report highlights the growth of AI patent applications at the European Patent Office (EPO) and how the pace of growth has slowed, indicating that the AI industry is maturing. Chinese applicants at the EPO have overtaken Japanese applicants in 2020, with per capita applicants from the Republic of Korea becoming the largest filers. The report also notes that computer vision technology has seen an accelerated increase in patent applications, while speech processing applications have been in decline, indicating that computer vision technology has reached maturity. Additionally, the report mentions that the rate of growth in AI filings at the EPO is slowing and the EPO has attempted to reduce the average age of AI applications, but with limited success.
AI report 2022: Patents and the EPO — a long-term trend analysis
Preview: The report highlights the growth of AI patent applications at the European Patent Office (EPO) and how the pace of growth has slowed, indicating that the AI industry is maturing. Chinese applicants at the EPO have overtaken Japanese applicants in 2020, with per capita applicants from the Republic of Korea becoming the largest filers. The report also notes that computer vision technology has seen an accelerated increase in patent applications, while speech processing applications have been in decline, indicating that computer vision technology has reached maturity. Additionally, the report mentions that the rate of growth in AI filings at the EPO is slowing and the EPO has attempted to reduce the average age of AI applications, but with limited success.
1 Introduction
The number of AI patent applications at the European Patent Office has grown at a convincing rate in recent years, but the pace of growth has slowed, suggesting that the AI industry is beginning to mature.
Patents have become an increasingly key part of defending business strategy, with AI patents in particular becoming integral to businesses' commercial strategies.
Chinese applicants at the EPO have overtaken Japanese applicants in 2020, but per capita, applicants from the Republic of Korea have become the largest filers.
In the last five years, computer vision technology has seen an accelerated increase in patent applications, while speech processing applications have been in decline. This indicates that computer vision technology has reached maturity.
The rate of growth in AI filings at the EPO is slowing, with previous years seeing materially higher growth rates. In 2020, overall patent filings are expected to fall by 0.7%.
The EPO has attempted to reduce the average age of AI applications at closure over the last few years, but the rate of decrease slowed over this year.
The number of pending AI applications nearly doubled between 2016 and 2019. It will be interesting to see how the slowdown in the rate of closure affects the overall number of pending cases.
The EPO opposition procedure allows a third party to challenge the validity of a European patent. The EPO aims to give a decision on straightforward opposition cases within 15 months.
The proportion of AI patents that are opposed by the EPO is relatively low, at around 1%, but is in line with the proportion of patents in the electrical engineering field.
There is an increase in the number of oppositions filed in the AI space in 2020 and 2021, and a small increase in the opposition rate, suggesting that patents are becoming increasingly important to commercial strategy.
Under the current system, companies must file oppositions at the EPO. However, under the new system, companies can bring revocation actions in the Unified Patent Court (UPC) without initiating any opposition procedure before the EPO.
Marks & Clerk's AI applications had a higher allowance rate than those filed by other European firms from 2015 onwards.
The allowance rate for AI applications has continued to increase over the past ten years, but remains below the average for all applications at the EPO.
Figure 1.4.2 shows how the allowance rate changes with the age of closure. The allowance rate is relatively stable, but there is a slight decrease in allowance rate for applications that undergo examination for a longer period.
The Enlarged Board of Appeal issued a decision on 10 March 2021 on a case relating to computer implemented simulations. The decision is particularly important since there has only been one previous referral relating directly to patentability of software. A computer-implemented method of modelling pedestrian crowd movement in an environment was refused by the Examining Division in August 2013 and subsequently appealed to the Board of Appeal, which concluded that a referral to the Enlarged Board of Appeal was necessary.
2 AI technologies
The largest technology sectors were Life & Medical Science (15%), Telecommunications (12%) and Physical Sciences (11%), with Industry and Manufacturing jumping into the ranks of the top sectors.
The trend of increasing numbers of AI applications was maintained across all but one category, with Transportation seeing a notable decrease in the number of publications.
Transportation was the only sector to experience a decrease in publications, while Telecommunications, Security and Energy Management experienced growth rates reduction.
We looked at the technology area split for applicants based in different countries. For Chinese applicants, Telecoms was by far the largest sector, accounting for 30% of AI applications filed.
Japanese applicants are increasingly interested in applying AI in the transportation sector, perhaps due to a favourable policy environment in Europe for testing automated transport technologies.
The allowance rate between different technology sectors varied widely, with the highest allowance rate being for "Transportation".
Technical and non-technical sectors 2.1
The following categories had above average allowance rates: transportation, agriculture, energy management, telecoms, security, life and medical science, physical sciences, arts and humanities, and cartography.
The EPO has historically deemed AI applications relating to planning and scheduling to be technical, while AI applications relating to music are more likely to be deemed "non-technical" by the EPO.
Some sectors have seen a significant increase in allowance rate in recent years, such as the Security sector and the Networks sector. The allowance rate for the Transportation sector has remained relatively flat.
We determined the type of AI technology for each application based on keyword searching of the text of the applications themselves.
The proportion of publications of applications in the computer vision field has increased since our previous report, while the proportion of publications in the speech processing field has decreased. This likely indicates that speech processing as a technology has reached a maturity commensurate with the current market opportunities.
The proportion of Speech Processing applications has declined, but not that of Natural Language Processing. This could be because Speech Processing is now sufficient for most applications, or because applicants are placing more value on being able to understand the meaning of language than in detecting it.
The allowance rate for applications related to computer vision has seen the sharpest increase, and the allowance rate for natural language processing has remained fairly stable.
Patent Title Analysis
A Sentence-BERT model was used to compare the semantic similarity of patent publications. The chart below shows how similar or dissimilar each publication is with respect to each other.
When embedding text using a BERT model, the resulting vector has high dimensionality. We perform dimensionality reduction using t-distributed Stochastic Neighbour Embeddings (t-SNE) to enable visualisation in two dimensions.
The embedding vectors produced by our model seem to cross validate one another, with robotics and control methods clustering near one-another, speech processing and natural language processing clustering around the bottom centre, and predictive analysis distributed throughout.
Advancements in AI have made it possible to accurately assess the meaning of text with the assistance of very large and complex models.
3 AI Applicants
The number of AI patent applications filed at the EPO has increased in recent years, with some countries seeing this growth accelerate. The US continues to be the individual country filing the most AI applications at the EPO.
South Korean applicants file the smallest number of AI applications per head of population, but they are overtaking Japanese, US and EP applicants to become the largest filers of AI applications per-capita.
While US applicants continue to make up the largest proportion of AI applications filed at the EPO, the proportion of applications from other countries is increasing. China continues to enjoy the largest growth in share of AI applications in recent years.
We noted in our last report that US applicants had a lower success rate than applicants from any of the other top 5 countries.
In the period 2000-2006, Chinese applicants had a success rate of around 20%, compared to 50% for applicants from South Korea. In 2007-2014, Chinese applicants enjoyed the highest success rate of each of the top 5 source countries.
The number of AI patent applications continues to increase in Europe, with South Korean applicants becoming the largest filers of AI applications on a per-capita basis.
4 Conclusion
Computer vision has shown a striking increase in patent publications over the last five years, while speech processing has been in decline. This indicates that computer vision has a still significant runway of potential applications.
In recent years the EPO has been effective in reducing the average length of examination for applications in the AI space, and the allowance rate varies widely between different sectors.
5 Methodology
To produce the data analysed in this report, we used the IPC code and keyword definitions used for patent data in the WIPO Technology Trends 2019: Artificial Intelligence report. We then wrote custom formulae using the raw data to generate our own fields for the analysis.
Mike Williams
Mike is an expert in patent matters relating to digital technologies, and in particular artificial intelligence (AI).
Mike has extensive experience in patent matters relating to all aspects of computer science, including signal processing, image analysis, communications protocols, computer graphics.
Lara Sibley
Lara advises clients on patenting AI innovations in areas such as natural language processing, medical imaging and home automation, and she has given numerous presentations in Japan on the EPO approach to AI.
Lara qualified as a UK Chartered Patent Attorney in 2016 and a European Patent Attorney in 2016, having achieved the highest mark in Europe for the English language Opposition paper.
Matthew Jefferies
Matthew advises a wide range of clients in high tech fields, including telecommunications, machine learning, medical devices and non-volatile semiconductor memory, and has assisted with appeals and oppositions before the European Patent Office.
Matthew has experience in neural networking and has completed the Stanford University certificate in Machine Learning.
Matthew graduated from Durham University with a first class Masters degree in Physics. He received the Certificate in Intellectual Property Law from Queen Mary University and qualified as a UK and European patent attorney in 2015.
Full Report: https://www.marks-clerk.com/media/trgmnz4m/ai-report-2022-marks-clerk-pdf.pdf