Less Discriminatory Algorithms
The article discusses the phenomenon of model multiplicity in machine learning, arguing that developers should be legally obligated to search for less discriminatory algorithms (LDAs) to reduce disparities in algorithmic decision-making.
The article explores the concept of model multiplicity in machine learning, which suggests that multiple algorithms with equivalent performance often exist for a given prediction problem. These alternative models can have different impacts on demographic groups, and one of them is typically less discriminatory. The authors argue that developers should be legally required to search for and implement these less discriminatory algorithms (LDAs) as part of the model-building process, especially in domains like housing, employment, and credit where civil rights laws apply.
Share this Resource:
