Executive Summary
This data brief informs policymaker audiences who desire to understand how they might use patent data in planning for the quickly advancing impacts of Artificial Intelligence (AI). Such data can provide policymakers with insights into which areas of AI are rapidly developing, which countries are especially active in AI research, and which organizations are responsible for key AI inventions.
In this primer, we report analytic results on worldwide trends in AI patenting and suggest options for how these results might be interpreted and leveraged.
Key findings presented in this primer include:
- There were 10 times as many AI patent applications published worldwide in 2019 as in 2013, most of which have yet to be examined.
- Patent applications increased by 500 percent from 2009 to 2019 within the Chinese patent office—90 percent were domestic applications. The U.S. patent office has seen a 35 percent increase in applications during the same time, 48 percent of which were domestic.
- While the quality of Chinese patents has been repeatedly called into question, there are signs that the situation may be improving.
- Large companies—notably IBM, Microsoft, and Google—dominate AI patenting among U.S. organizations. Meanwhile, Chinese AI patenting is distributed much more broadly across companies (e.g., Ping An, Baidu, Tencent), government organizations (e.g. State Grid), and universities (e.g., Electronic Sci/Tech, Zhejiang, Xidian).
- China focuses AI patenting on Computer Vision, Japan on Control Systems, and Korea on Speech Processing. The United States is more evenly distributed across research fields.
We also place these analytic results in the wider context of patenting and note limitations and considerations with respect to AI patents. Key contextual points found within this primer include:
- Patents are an exchange between the inventor (publicly shared insights) and society (protected rights).
- A patent has a pre-examination “published application” and, if approved, a “granted patent” that confers rights.
- AI patents are strongly dependent on mathematical relationships and algorithms, which are considered abstract ideas under patent law and therefore restrict what can be patented. This concept has and will continue to evolve.
- Patent applications have very different meanings depending on where they are filed, which should be considered when comparing innovation trends.
Finally, patent data offers a useful measure of inventive activity across companies, regions, or countries. That said, care should be taken in using AI patent data to support policy decisions, since patenting in AI is growing so quickly. With many recent patent applications yet to be examined, quality and impact are hard to predict.