Weapons of Math Destruction

The book reads like a continuation of Ted Kaczynski's manifesto ‘The Industrial Society and Its Future'. This time focusing on machine learning and its use in coercing behavior change as well as discriminating the poor and disadvantaged. From the examples provided in the book, there are three categories of Weapons of Math Destructions (WMD).

First one is poor statistics. These are incorrectly calculated stats which are used to infer human behavior and performance. In them are lack of understanding of how certain statistics are interpreted or validated. A good example are proxy variables such as geography used to infer purchase power, reoffending propensity et cetera.

The second WMD are correct statistics that are misused. These seem to be the majority of the cases. It is more of an ethical issue rather than machine taking over of lives. When a company utilizes zip code to steer customers to high interest loans, that qualifies us an ethical use of machine leaning output and no necessarily anything wrong with the machine leaning process.

The last WMDs are dataset. From the book, certain attributes within data should never be used for prediction purposes, e.g race, gender, income, and zip code since they likely to correlate with outputs connected with discrimination.

In the end, machine learning is hailed as tool that can be used for social good - with several examples provided.