Zachary Lipton: Where Machine Learning Falls Short

ML Solutionism

Zachary Lipton critiques the over-reliance on machine learning to solve complex social issues, a concept known as ML solutionism. He argues that this approach often ignores the limitations of algorithmic formalism, especially when applied to fairness and policy. Lipton emphasizes the importance of understanding the context and data before making determinations, drawing parallels to scientific inquiry:

Science doesn't work by people just saying, just give me a hard drive, and then I'll tell you the laws of electromagnetism or something.

--- Zachary Lipton

This highlights the need for a nuanced approach that considers the broader social and political structures involved 1 2.

   

Fairness & Identification

Lipton discusses the challenges of identifying causal relationships in data, particularly in the context of algorithmic fairness. He points out that many ML fairness efforts assume a simplistic view of data, ignoring the complexities of how data is collected and the social processes it represents. This can lead to misguided conclusions about fairness:

The idea that you could just sort of say in a sort of generally agnostic way, like I've got two demographics, I've got some features, tell me what's fair.

--- Zachary Lipton

Lipton stresses the importance of understanding the underlying variables and processes to make informed decisions 3.

   

US Law Relevance

Lipton analyzes how algorithmic fairness relates to US legal standards, particularly the Civil Rights Act of 1964. He critiques the simplistic application of mathematical expressions to legal doctrines, which can lead to a disconnect between ML models and the law's intent. Lipton highlights the importance of understanding the law's complexity:

It's understanding the way these things relate to each other and being able to kind of sit right in that binding and understand and articulate.

--- Zachary Lipton

This approach requires bridging the gap between legal scholarship and quantitative research to ensure that ML applications align with legal standards 4.