Xiaozhong Liu (left) and Devan Donaldson
Two researchers from the Luddy School of Informatics, Computing, and Engineering have earned an award from IU’s Racial Justice Research Fund to study biases in artificial intelligence algorithms for legal prediction.
The project, [Chinese Student = Spy?] [Black Man = Predator?] Reducing Racial Bias in Legal Artificial Intelligence, will be led by Associate Professor of Information Science Xiaozhong Liu and Assistant Professor Devan Donaldson. It will focus on the fairness and trustworthiness of legal artificial intelligence, a special track in AI that is playing an increasingly important role in addressing various legal needs, including helping clients, lawyers, and judges access, understand, and generate legal domain knowledge.
“In the textual content of a legal case, racial information, such as Asian female, dark skin, slanted eyes, black young man, etc., may be memorized by AI models for prediction, which can lead to racial biases,” Liu said. “For this project, we will propose novel algorithms to de-bias the racial information in the legal documents to ensure AI fairness and trustworthiness.”
Liu and Donaldson hope to learn what kind of information can cause racial biases for legal AI, how racial information can be detected and removed from AI models for prediction, and how algorithms can be developed to remove potential bias from large legal corpuses or digital libraries. The researchers will develop open-source algorithm APIs for legal AI fairness and generate a De-biased Legal Digital Repository to ensure racial fairness in future legal AI research.
“We will collect large amounts of legal cases for database generation and recruit students to annotate those cases for model training,” Liu said. “We also will supervise a series of 15 semi-structured interviews to understand the perceptions of the value of the De-biased Legal Digital Repository for educational, research, and business purposes.”
IU’s Racial Justice Research Fund was created in June 2020 to provide seed funding for projects from across IU that address the systemic conditions fostering racism in the United States. The Luddy researchers will receive just short of $15,000 in funding for their work.
“The Luddy School was founded on the idea of developing technology that can solve real-world problems, and this project is the perfect example of that vision in action,” said Kay Connelly, the associate dean for research at the Luddy School. “The development of a de-biased legal AI will be crucial to improving outcomes for users, and the expertise of Xiaozhong in AI and Devan in metadata and the effective use of digital repositories makes this the perfect collaboration.”