Introduction
A collaboration between Stanford University’s Regulation [2], Evaluation [2], and Governance Lab (RegLab) and Santa Clara County has led to the development of an advanced large language model. This model is designed to assist in complying with California’s AB 1466, which requires the removal of discriminatory language, such as historical racial covenants, from property deed records [2]. The initiative addresses the enduring legacy of housing discrimination.
Description
A collaboration between Stanford University’s Regulation [2], Evaluation [2], and Governance Lab (RegLab) and Santa Clara County has developed a finely tuned large language model to assist in complying with California’s AB 1466, which mandates the removal of discriminatory language [1], including historical racial covenants [1], from property deed records [2]. This initiative aims to address the legacy of housing discrimination, as these covenants have been constitutionally unenforceable since 1948 [2].
The AI model has demonstrated near-perfect accuracy in detecting racial covenants [2], achieving a false positive rate close to zero [1], which significantly reduces the time and cost associated with the redaction process [1]. It processed 84 million pages of property records in just a few days [1], saving the county over 86,500 person hours and costing only a few hundred dollars, compared to an estimated $1.4 million for manual reviews [1]. Traditional methods [1], such as keyword searches [1], often resulted in high false positive rates and missed instances of racial covenants, highlighting the advanced capabilities of this model.
In addition to redaction, the technology converts images of property deeds into text [1], analyzes them for discriminatory language [1], and highlights relevant sections for legal review [1]. Historical mapping techniques were employed to geolocate properties affected by these covenants [2], revealing that one in four properties in Santa Clara County were subject to such restrictions as of 1950 [2]. The model effectively understands context from various documents and cross-references textual clues with administrative datasets to pinpoint property locations.
The findings and the model, along with a web interface tool for training data [1], will be made available for free to government agencies [1], facilitating human review while preserving original racist language for reference [1]. This initiative exemplifies the effective use of technology in fulfilling legislative mandates and addressing historical patterns of housing discrimination [2], with the potential for adaptation to identify other forms of discriminatory language in the future [1].
Conclusion
This initiative not only exemplifies the effective use of technology in fulfilling legislative mandates but also significantly impacts the efficiency and accuracy of identifying and removing discriminatory language from property records. By making the model and its findings freely available to government agencies, it facilitates a broader effort to address historical patterns of housing discrimination. The potential for adapting this technology to identify other forms of discriminatory language further underscores its importance and utility in promoting equity and justice in property law.
References
[1] https://statescoop.com/stanford-university-santa-clara-county-clerk-racial-covenants-ai/
[2] https://hai.stanford.edu/news/stanford-reglab-princeton-and-county-santa-clara-collaborate-use-ai-identify-and-map-racial
