Artificial Intelligence Patent Dataset (AIPD)
U.S. Patent and Trademark Office
Machine-learning labels flagging which of 13 million U.S. patents involve AI, by technology type.
USPTO's data set classifying 13,244,037 granted patents and published applications from 1976 through 2023 into eight AI component technologies, using machine learning models. Each row is a document with a flag and prediction score for each technology, such as machine learning, planning, and computer vision.
Access
- Cost
- FreeNo charge; open government data.
- License
- Public domainPublic Domain Mark 1.0
- Sign-up
- No sign-up
- How you get it
- Bulk downloadCSV, TSV, Stata
What's in it
- One row is
- One row per patent or published application
- Covers
- United States
- Years
- 1976-2023
- Updated
- Irregular
- Size
- 13,244,037 documents
Quality
A research data set with a published method and a comprehensive scope. Labels come from a model, not from human review, so errors exist.
- Labels come from a machine learning model
- Technology flags overlap
Connects to
- PatentsView: Join on patent number for inventors, assignees, and citations
- Higher Education Research and Development Survey: Compare AI patenting with university R&D
- County Business Patterns: Relate AI patent activity to local industry
Questions it could answer
- Which technology areas of AI patents grew fastest after 2012?
- Which states have the highest share of AI patents?