Weapons of Math Destruction are mathematical models that influence many lives, such as the examples shown by O’Neil- college rankings, online advertising that leads people to take loans, use of algorithms to sentence criminals, use of crime predictor to allocate police faster, and use of algorithms and biased models for employments. These WMDs produce negatives feedback loops, such as an individual unable to find a job because the model decided that the individual’s credit score was too low, thus further putting that person into lower credit score because of unemployment.
WMDs works for efficiency while at the cost of fairness for the society. How can WMDs be less destructive and minimize negative feedback loops? O’Neil discusses the baseball model extensively. Baseball models are fair because they do not use proxies (inaccurate or predictions of a player skills) and instead records hard evidence (number of hits a player makes). Another lesson we can take from baseball is that the models can be adjusted based on the results. WMDs like the teacher assessment and the recidivism model does not learn from the teachers and the potential criminals after the models produce the results. They do not know how the results affects the teacher and thus the models continue to spit out results without adjusting or learning. WMDs are hard to prevent as they made by humans so the models can never be perfect. The best thing we can do is to study WMDs and try to make them less destructive and more beneficial for society.
David J's LC Blog 
I like how you consider the feedback loop and WMDs by using many examples of several chapters. Even though you used the baseball model as a good model, do you think it’s possible to collect all huge data/information in other firms by human hands like the baseball model? Also, will people prioritize the accuracy and fairness of the models than efficiency after studying WMDs?
LikeLike
I believe transparency is needed for a model to obtain a large amount of data and to be able to constantly update. With understanding WMDs and studying the construction of models, I think accuracy and fairness would become as important if not more than efficiency but I think that’s quite a reach.
LikeLike
David, your explanation of the feedback loop is really thorough and clear, and the proposal for preventing WMDs seems very effective. I was convinced of it. One thing I wondered was if we can completely eliminate proxies from a model because it seems difficult for me when modelizing something for the sake of efficiency. But I totally agree with your opinion that we all should know the fact WMDs are prevailing, and I liked your way of concisely describing explanations and opinions. Thank you!
LikeLike
Thank you! In order to eliminate as much proxies as possible, I believe have a database that is large and constantly updating would help, but that would effect the efficiency and quickness of the model. One thing for another I guess.
LikeLike
Very good insight on the topic! The examples that helped grasp the topics were clearly evident and helped me make sure that I entirely understood the difference. Although I understand that proxies are generally problematic and can be harmful to the models that use them, sometimes there truly are no ways to go around it, especially when it comes to important or confidential information that is not usually accessible to others/the public. In these cases, would it be better to use proxies, or instead just find another way around that also might not be as effective to a specific model?
LikeLike
Thank you. I believe using proxies that shows strong correlations would be much better than proxies to just fill in the blanks. Ideally, having a large enough data would minimize the need for proxies.
LikeLike