User:Andrew.Greer123/sandbox

Hello! This is in addition to the previously existing Weapons of Math Destruction article. However, I've rewritten several of the sections to accommodate the increased scope of the article. I have made only minor edits to the previously existing infobox on the page to bring its topics more in line with the subject matter.

Weapons of Math Destruction is a book about the social injustices present within contemporary algorithmic models, published by American mathematician and data scientist Cathy O'Neil on September 12th, 2016. Within this book, O'Neil explores how many big data algorithms are creating repeating cycles of social inequality and codifying past discrimination into modern data models. O'Neil refers to these algorithms as 'Weapons of Math Destruction' or 'WMDs' for short, a play on of the term 'Weapons of Mass Destruction', which is commonly used to refer to nuclear, biological, and chemical weaponry. These data models are characterized by three factors; the scale at which they are implemented; how opaque their inner workings are; and the scope of the damage they inflict on society. Within this book, O'Neil seeks to dispel the myth of the unbiased, efficient mathematical model by exposing the biases that can be ingrained within a system despite the intentions of its creator. To this end, it examines the usage of these algorithms in several industries, including college applications, online advertising, policing, job applications, credit scores, insurance, shift scheduling in the workplace, and political engagement.

In 2016 it was added to the National Book Awards for Nonfiction longlist  and won the Mathematical Association of America's Euler Book Prize in 2018.

Summary
Throughout the book, O'Neil illustrates the varying forms Weapons of Math Destruction can take through a series of case studies. This includes:


 * Targeted advertising that is directed using algorithms at those who may be vulnerable to its message and promised service, often charging significant fees.
 * Algorithms used to calculate recidivism utilized in criminal sentencing and predictive policing which creates a self-sustaining cycle of targeting specific communities.
 * College ranking systems that fail to account for factors such as tuition and incentivize manipulating the system at the expense of student finances.
 * Personality tests meant to screen out applicants with undesired qualities such as mental health and preventing them from getting work anywhere that utilizes the system.
 * Shift schedule optimizing algorithms that maximize profit and efficiency at the expense of employee health and schedules, increasing strain on their personal lives and making it more challenging to raise a family or attend college.
 * Models that evaluate customers and assign them an insurance score that utilizes questionable proxy values to increase the cost of attaining insurance for specific groups.
 * Tools for bracketing like-minded voters into categories to be targeted with specific messaging tailored for their beliefs.

Key Components of a Weapon of Math Destruction
O'Neil's 'Weapons of Math Destruction are algorithms that share the following traits, scalability, opacity, and damage. Scalability refers to the algorithmic models being potentially scalable to fit with any size of data set. Opacity denotes whether or not the exact variables the algorithmic models measure are available to public scrutiny. A lack of opacity makes the outputs difficult to contest for the general public, as well as being challenging to regulate. Damage refers to their ability to deal significant harm to society, especially vulnerable and minority populations.

Chapter One
In this chapter, O'Neil examines the components of her 'Weapons of Math Destruction', these being opacity, scalability, and destructive potential. She speaks favorably of the statistical models utilized by data analysts in baseball. These algorithms are transparent, as everyone has access to all of the statistics that are being examined. As well, the data sets utilized directly relate to the behavior they're trying to explore. These models have to continually evolve in response to feedback they receive from the testing of their predictions playing out in real-time. Because of this, they serve as a positive contrast to the negative models she showcases throughout the book.

Chapter Two
Within this section, O'Neil recounts her journey through the world of data analytics and her disillusionment with it following the Financial crisis of 2007–08.

Chapter Three
In chapter three, the author makes the argument that college rankings within American news publications are Weapons of Math Destruction. This is due to their reliance on ambiguous proxy values to create their rankings and their widespread popularity spurring a competition between institutions to manipulate the algorithms to drive their rankings higher. O'Neil believes that this has contributed to the dramatic increase in the cost of tuition in America over the past several decades, with colleges passing the costs of this competition onto their students.

Chapter Four
In this chapter, O'Neil explores online advertising algorithms and their ability to identify and exploit vulnerable members of society that may be susceptible to their messaging. She determines these algorithms to be Weapons of Math Destruction based on their flexible scalability, opaque nature as they carry no specific guidelines or criteria for inclusion, and are dangerous as they target those who are either desperate or uninformed with overpriced services.

Chapter Five
Chapter five features the author examining recidivism models utilized during criminal sentencing in the American justice system, as well as predictive policing algorithms. Despite the claim that these systems are racially blind, O'Neil argues that cities are often segregated along lines of income and race, making proxy values such as zip codes and crime rates accurate predictors of race. When an algorithm treats all individuals within one of these areas as potentially dangerous, O'Neil argues that the community suffers from disproportionate police presence.

Chapter Six
Within this chapter, O'Neil explains that algorithms utilized during automated hiring processes are Weapons of Math Destruction as well. Personality tests are scalable to any company's demands, their function, as well as their assumptions, are unknown to the participants, and being labelled as unsuitable can lead to a candidate being deemed unfit for work across several industries.

Chapter Seven
Continuing from her examination of hiring processes, O'Neil explores the damage that scheduling optimization algorithms can have upon employees by forcing erratic working hours, stifling their ability to pursue a life outside their work.

Chapter Eight
In this chapter, O'Neil performs an analysis of various 'e-scores' utilized to evaluate individuals. Models that employ vast quantities of data available about an individual online to determine their potential risk, in this case, for determining their available credit. As these are often factored within the hiring process, individuals with poor credit can find themselves unable to find work and improve their credit rating.

Chapter Nine
Continuing from the last chapter, O'Neil examines 'e-scores' from the perspective of insurance premiums. These algorithms are Weapons of Math Destruction as individuals cannot view their score to correct mistakes, dangerous due to the additional financial burden placed upon them, and scalable because of their widespread usage.

Chapter Ten
In this final case study, O'Neil applies her arguments towards data models and their applications in political campaigns. Algorithms evaluating the data present on social media sites allow for politicians to create highly-customized messages tailored to a voting group's preferences. These algorithms are Weapons of Math Destruction, as they create multiple hidden versions of a politician's message, they are scalable to any voter demographic, and the diverging narratives presented to voters are dangerous for the democratic process.

The Myth of Big Data
A cornerstone of the defense of these models is that because they are mathematical formulas, they cannot be biased as they cannot do anything but analyze data sets. However, O'Neil explains that while the math might not be biased, these algorithms were crafted with the assumptions of its writer. She argues that no single model can factor in all of the nuance and complexity of human beings, and as such, will always have blind spots. Because of this false belief in the unbiased nature of algorithms, she theorizes that oversight of these models is severely lacking. As well, she states that the skill-sets required to analyze mathematical models for potential discriminatory factors are lacking within the typical education of a programmer.

Proxy Values
To create reliable models, data scientists require accurate measurements of whatever phenomena they are looking to quantify for analysis. However, it is frequently the case that direct measurements are either insufficient or difficult to quantify. To remedy this, data scientists utilize proxy values to achieve a measurement of values for their algorithms. These proxy values often look for generally shared characteristics amongst a group, ignoring the broader context of an individual's life. These proxy variables can produce negative, self-sustaining feedback loops and are often drawn from flawed statistical techniques. In this way, algorithms can create a reality that reflects and reinforces the biases present within their code. For example, when attempting to determine whether or not an individual is a flight risk, algorithms will look for factors that are associated with past flight risks. As a result, individuals can be judged as a flight risk through the similarities they share with others. By ignoring causal relationships and weighing all data points equally, depending on the model, this can result in large swathes of a given group being flagged as dangerous despite the intentions of a data scientist.

A critical factor in this phenomenon is what O'Neil describes as 'Birds of a Feather' thinking. A principle that refers to the idea that individuals who share characteristics such as background, personality, or interests will act similarly to one another. This allows algorithms to maximize their efficiency at which they remove individuals perceived to be dangerous or harmful, but fails to evaluate them based on their own merits.

Black Box Algorithms
O'Neil notes that data scientists rarely evaluate and update their models to improve their accuracy. Due to the secretive nature of an algorithm, it can be challenging to determine if there is a flaw or bias in the way the data is evaluated without a glaring issue appearing within the output. As well, because algorithms utilizing machine learning to adjust their criteria; over time, they will inadvertently recreate the conditions that led to their conclusions in the first place, creating a self-sustaining cycle. Algorithms require feedback to succeed and reduce errors, without any corrections, the system will continue to produce potentially false analysis with significant repercussions for those it's evaluating.

Inequality
Through the usage of case studies, O'Neil examines how big data algorithms do not affect all individuals equally. The poor, marginalized, and vulnerable are all significantly more susceptible to the damage an algorithm can inflict. This is due to their challenges often being ignored or unknown to the designers of an algorithm. As well, these communities lack the resources and expertise to challenge the results of a data model. Although O'Neil argues that algorithms are rarely created with the intention of being prejudiced against minority populations, however, they often create additional difficulties through reinforcing social assumptions that result in discrimination. As an example, O'Neil investigates the usage of algorithmic models in determining where law-enforcement should concentrate their efforts. Although these models may lack identifying information such as race, gender, or religion, proxy variables such as zip codes, education, and reported crimes can still create a similar effect. Marginalized and impoverished communities typically concentrate within an area, are generally under-educated, and are likely to have higher crime rates. As well O'Neil argues that as police officers are drawn to one of these areas, the more likely they'll continue to make arrests within that area, further reinforcing the data being inputted to predictive policing algorithms.

As well, O'Neil observes these difficulties following these communities into the workplace, as automatic hiring software and models that are utilized to filter potential candidates during the hiring process based on keywords. Those with access to the knowledge and resources to tailor their resumes to the demands of these systems will have a distinct advantage to those without. Sometimes, personality tests are utilized to assess if a candidate is ideal for the position. However, this can discriminate against candidates with mental disabilities. Additionally, an individual denied a job based on their credit score is likely to fall into a cycle of being unable to find high paying work, ensuring that their credit score will stay low for the foreseeable future. Frequently, the historical data sets utilized by algorithms are reflective of the discriminatory biases of the time. As they seek to replicate ideal candidates from this data, they reinforce the prejudices of the past.

Another significant factor O'Neil notes throughout the book is, typically, impoverished communities have less access to the education, skills, and resources to become aware of the inner workings of algorithms in the first place.

Proposed Solutions
O'Neil presents a variety of solutions to the issues presented in the book. She stresses the addition of a human ethical element to the process of creating algorithms, putting the impetus on data scientists to check their behaviour and establish a model of best practices. To this end, she quotes Emanuel Derman and Paul Wilmott's 'The Modeler's Hippocratic Oath' from their work, The Financial Modelers' Manifesto,
 * I will remember that I didn't make the world, and it doesn't satisfy my equations.
 * Though I will use models boldly to estimate value, I will not be overly impressed by mathematics.
 * I will never sacrifice reality for elegance without explaining why I have done so.
 * Nor will I give the people who use my model false comfort about its accuracy. Instead, I will make explicit its assumptions and oversights.
 * I understand that my work may have enormous effects on society and the economy, many of them beyond my comprehension.

O'Neil recommends that algorithms and their outputs are regularly evaluated to ensure that they are functioning appropriately and without undue prejudice against certain groups or communities. As well, she urges data scientists to challenge their assumptions and to be cautious whenever the results of their algorithms correlate with their own beliefs. She calls for the raising of awareness amongst the public to rally against those who utilize algorithms wrongly, and for the integration of positive feedback loops to allow for oversight. Additionally, O'Neil advocates for the implementation of a comprehensive regulatory system for algorithmic models, as current systems allowing for transparency are without a consistent code of conduct. One that would measure the hidden costs of data models, such as the impacts they have upon the vulnerable and poor. Finally, O'Neil recommends a shift towards a European style model of personal data sharing, where all information being divulged is optional for the customer. As well, she suggests a non-re-usability clause in regards to personal data, disincentivizing the storage of data.

Popular Reception
Many have praised the book for its accessibility, including Wendy M. Grossman, writing for ZDNet, describes the book as "O'Neil does a particularly fine job of explaining the basis for that contention -- and does it without formulas, in plain, accessible language." Evelyn Lamb commended the book in Scientific American for shining a light on this issue in an accessible fashion.

Several commentators have highlighted the book for its concise writing style. Richard Beales writing for BreakingViews recommended the book for its succinct presentation of its argument, calling it "a thought-provoking read for anyone inclined to believe that data doesn't lie." Canadian-British blogger and digital copyright activist Cory Doctorow of the blog Boing Boing described the book as "a vital crash-course in the specialized kind of statistical knowledge we all need to interrogate the systems around us and demand better."

Daniel Gutierrez of InsideBigData commented that the book raises important questions about the work of a data scientist, saying "I think all data scientists should read Weapons of Math Description in order to add an important filter for judging how their work may be misused." Data scientist Hafidz Zulkifli writing on Medium's Towards Data Science publication praised the book for its unique take on the ethical concerns of data ethics, writing "I find the angle about bias and self reinforcing bias quite interesting as it does have a far wider and continuing impact to the victims of the model." Danny Dorling writing for Times Higher Education praised the book for its exploration of the hidden technologies behind so many obstacles and questions in modern life and praised for its overall excellence, "Weapons of Math Destruction is a well-written, entertaining and very valuable book."

Will Rhinehart writing for The Technology Liberation Front criticized the book for failing to examine the broader political and social motivations beyond the algorithm. He described the book as something that "continuously toys with important questions regarding the moral agency of technologies but never explicitly lays them out."

The Financial Times' Federica Cocco writes that although O'Neil's proposed solutions are "are not highly imaginative", she nevertheless commended the book, describing it as "a manual for the 21st-century citizen, and it succeeds where other big data accounts have failed — it is accessible, refreshingly critical and feels relevant and urgent."

Scholarly Reception
Reception within academia has been largely positive to O'Neil's work. Professor Mary Poovey praised the work in American Mathematical Society, writing "O’Neil’s book may seem too polemical to some readers and too cautious to others, it speaks forcefully to the cultural moment we share.” However, Poovey has some criticism for the book, citing O’Neil’s failure to fully grasp “the full extent of what we stand to lose as we transfer the very processes of judgment from humans to algorithms.” Professor Thomas Woodson writing for the Journal of Responsible Innovation, applauded the book for its accessible nature, noting that,

"“This book may not delve deep into the theoretical scholarship on science, technology, and inequality, but it does highlight the pervasiveness of inequality and many interconnections between innovation and society. Thus, for members of the general public as well as for various technical communities, reading Weapons of Math Destruction should serve as a compelling introduction into the idea that science and mathematics are never value-free.”"

In The Social Science Journal, Professor Zaki Eusufzai proclaimed the book as “hard-hitting” and written by “an insider with impressive academic credentials.” He recommending it his fellow social scientists stating,

"“Beyond the hype however, for social scientists, there are two compelling reasons to read the book. First, the case studies are excellent examples of how behind-the-scenes technical algorithms can have important negative effects on society at large when proper modeling procedures are not followed. Second, social scientists have long been aware of the difficulties involved in extracting solid knowledge out of social data. Confounding variables, feedback effects, necessary use of proxies have all complicated empirical work in the social sciences. This book simply confirms that even with these new DS tools, those problems have not gone away. To ignore those problems and proceed anyway is to indulge in hubris and inflict real costs on society.”"

Doctoral Scholar Shikha Verma in Vikalpa, commending the book for being “a much-needed attempt to highlight the perils of Big Data” and recommended the book as a “enlightening read for students, policy makers and practitioners who want to understand the repercussions of treating Big Data as a panacea for all organizational and social problems.”

Awards
In 2016, the book was placed on the National Book Awards for Nonfiction’s longlist.

In 2019, the book won the Euler Book Prize of the Mathematical Association of America.