Book Review
Weapons of Math Destruction (by Cathy O' Neil)
|
- Moreover, O’ Neil appears to be making a social critique at various points in the book. Take the chapter on policing. A main argument is that recidivism-predicting algorithms are detrimental because they reinforce unviable political positions (like heavy policing). But that social critique isn’t supported...O Neil is not citing philosophers or political scientists, or theoreticians of justice. Readers are merely expected to agree with her assumptions. The upshot (for even a liberal like me who detests policial heavy-handedness) is a mathematician who appears to be talking about math but really committing herself to certain social positions — in an unsubstantiated way.
- In addition, O’ Neil’s arguments pertain to models which explicitly encode variables, like regressors, classifiers, and other first-wave AI technologies. As a result, they are inapplicable to deep neural networks that learn features on their own. These latter algorithms are arguably more relevant today and any argument for why they — which discover their own causal framework — are Weapons of Math Destruction will have to change.
- Finally, I felt the writing quality in this book fluctuated quite a bit. At some occasions, the sentence-level wording was quite strong. At others, I felt not only that it was weak, but also that the paragraphs or sub-sections weren’t structured in a way that advanced a singular intellectual thought or idea, an argument for or against something. Sometimes, this discontinuity left me scratching my head and wondering: “Where is this book going?” But that’s probably a function of the book’s being very research-heavy, and the quality of that research is very good indeed.