In 2021, California passed a law requiring its 58 counties to create programs to find and remove racial covenants from property deed records. These covenants were used in the past to restrict property ownership or use based on race. However, this task is enormous—Santa Clara County alone has 24 million deed records spanning 84 million pages, dating back to 1850.
To tackle this, some counties hired commercial vendors. Los Angeles County, for example, contracted a company for $8 million to complete the task over seven years. In other places, volunteers have come together to sift through deed records. But not all counties have the resources for such efforts.
A different approach
Santa Clara County, home to Silicon Valley, took a different approach. They partnered with Stanford University’s Regulation, Evaluation, and Governance Lab (RegLab), which used artificial intelligence (AI)—specifically large language models—to help with the project.
Led by Stanford’s RegLab and Princeton Professor Peter Henderson, the team compiled racial covenants from different jurisdictions across the country. They trained a cutting-edge AI model to detect these covenants in property records with near-perfect accuracy.
The team has made their AI model available to help other counties do the same, and the details can be found in their paper at https://reglab.github.io/racialcovenants/. Their approach is expected to save about 86,500 hours of labor and cost less than 2% of what comparable proprietary systems would require. They focused on 5.2 million deed records from 1902 to 1980, the most relevant period for this issue.
The team also cross-referenced historical maps to locate properties with racial covenants. By matching descriptions from old maps with administrative records, they were able to geolocate specific properties in Santa Clara County.
Important insights
Their findings revealed important insights into the use of racial covenants in the county. By 1950, around one in four properties in Santa Clara had these covenants. Just 10 developers were responsible for a third of the identified covenants, showing the significant role developers played in shaping the county’s racial landscape. Interestingly, African Americans and Asian Americans were excluded at the same rate, even though the local African American population was much smaller.
The team also discovered a case where a cemetery owned by the city of San Jose had burial deeds restricted to “Caucasians,” challenging the usual understanding that racial covenants were only used by private parties after racial zoning laws were struck down by the Supreme Court.
The researchers believe this project is a powerful example of how academic-government collaborations can make it easier to comply with new laws and shed light on historical housing discrimination. “This partnership dramatically reduced the time needed to review historical documents,” they explain, highlighting how technology can streamline the process of identifying, mapping, and removing racial covenants.
