search

liaochris / USPTO_AI

Project on the USPTO AI dataset
0 stars 0 forks source link

Exercise on Explaining Figures #2

Open liaochris opened 2 years ago

liaochris commented 2 years ago
  1. Mean citations made per patent by AI category over time Up until around 2014, citations for AI patents overall increased. This number plateaued at around 26-27 in 2014 and has stayed at that figure since. This suggests that over time, patent complexity increased before plateauing. A notable outlier we can observe from this graph is that in recent years, speech-recognition patents have skyrocketed in complexity.
  2. Domestic and International Institutions: This figure describes the proportion of AI patents made by American vs. Non-American institutions. A lot has been said about how other countries are beating America in the AI space, in fact, patents filed at the USPTO by American institutions dropped to its low in 1990 before rising throughout the 2010s and remaining at that same level ever since.
    • A statistic by AI Type exists and from what we can see, the compared to it's global competitiveness in terms of patenting, it lags behind in computer vision compared to other countries.
  3. Publications by Top 20 Publishers: This figure describes the proportion of patents published in each year by the top 20 patent publishers. The proportion of patents published by this group peaked at over 30% in 2009/2010 and has steadily declined ever since.
    • This reflects decreasing market concentration and increasing market concentration by other companies. This is important because it suggests that now, more than ever, it is easier to break into the patenting field
    • This trend is also reflected in the figure by AI Type, which suggests that there is not a particular field of AI where smaller companies are particular disadvantaged. At the same time, it suggests that the top companies are patenting in many AI fields. We also see that while there used to be great heterogeneity in the proportion of patents published by the top 20 publishers across fields, this heterogeneity has decreased over time - this suggests horizontal integration, or that firms are beginning to innovate in many fields
  4. There has been a loss in market share by the top 20 firms, most of whom entered the ai patenting scene decades ago, but the proportion of patents published by institutions who entered the market less than 5 years ago has not increased dramatically. However, what we can see from this graph is that the proportion of patents from these "new" firms has stagnated, rather than declined over the past few years. This suggests that in fact, since as time passes, firms exit from this "new firm" category, that the number of patents produced by new companies has aligned with overall patenting growth and that the number of patents produced by new companies is on the rise. Thus, the economy is conducive to the entrance of new firms in the AI space
    • Interestingly, there is also little heterogeneity with regards to the proportion of patents being produced by "new" firms, which aligns with our earlier result that there is not a particular field of AI where smaller companies are particular disadvantaged.
  5. What we see in trends about growth in new patents aligns with the USPTO's technical guide about complements - that hardware and vision are complements and grow together. There is also visual evidence to suggest that this is the case between planning and knowledge recognition patents.
  6. Finally, I examine heterogeneity across firms of different ages. Older firms are the most productive - as measured by the mean number of patents produced by firms in their age bracket. However, what we also see is that among older firms, in the blue and purple lines, there is one uber-productive firm that has been there since the beginning that plays a large role in patenting and as time has passed, it's contemporaries have lagged behind. Its contemporaries used to not lag behind so much - which is why in the turquoise line, the downwards dip after ten years is not as large as it is in the purple and blue lines.
    • We also see this reflected in measures of market concentration - this one uber-large firm's transition between decades is causing huge declines in the share of patents produced by institutions in their n-th decade, although as time has passed, its overall share of the market has decreased.