Book Review: Cathy O’Neil’s “Weapons of Math Destruction”

To read Cathy O’Neil’s Weapons of Math Destruction is to experience another in a line of progressive pugilists of the technological age. Where Tim Wu took on the future of the Internet and Evgeny Morozov chided online slactivism, O’Neil takes on algorithms, or what she has dubbed weapons of math destruction (WMD).

O’Neil’s book came at just the right moment in 2016. It sounded the alarm about big data just as it was becoming a topic for public discussion. And now, two years later, her worries seem prescient. As she explains in the introduction,

Big Data has plenty of evangelists, but I’m not one of them. This book will focus sharply in the other direction, on the damage inflicted by WMDs and the injustice they perpetuate. We will explore harmful examples that affect people at critical life moments: going to college, borrowing money, getting sentenced to prison, or finding and holding a job. All of these life domains are increasingly controlled by secret models wielding arbitrary punishments.

O’Neil is explicit about laying out the blame at the feet of the WMDs, “You cannot appeal to a WMD. That’s part of their fearsome power. They do not listen.” Yet, these models aren’t deployed and adopted in a frictionless environment. Instead, they “reflect goals and ideology” as O’Neil readily admits. Where _Weapons of Math Destruction _falters is that it ascribes too much agency to algorithms in places, and in doing so misses the broader politics behind algorithmic decision making.

For example, O’Neil begins her book with a story about Sarah Wysocki, a teacher who got fired from the D.C. public school system because of how the teacher evaluation system ranked her abilities. O’Neil writes,

Yet at the end of the 2010-11 school year, Wysocki received a miserable score on her IMPACT evaluation. Her problem was a new scoring system known as value-added modeling, which purported to measure her effectiveness in teaching math and language skills. That score, generated by an algorithm, represented half of her overall evaluation, and it outweighed the positive reviews from school administrators and the community. This left the district with no choice but to fire her, along with 205 other teachers who has IMPACT scores below the minimal threshold.

In the ensuing pages, O’Neil describes the scoring system, how it was designed, and how it affected Wysocki. But the broader politics behind the scoring system that ousted Wysocki are just as important.

Why, for example, was the value-added score such a prominent feature in the teacher evaluation as compared to administrative and parent input? Well, research from the Bill & Melinda Gates Foundation found that a teacher’s value-added track record is among the strongest predictors of student achievement gains. So, the school district changed around their evaluations to make it a central feature. As Jason Kamras, chief of human capital for D.C. schools, told the Washington Post, “We put a lot of stock in it.” But that decision wasn’t without its critics, including Washington Teachers’ Union President Nathan Saunders who said, “You can get me to walk down the road with you to say value-added is relevant, but 50 percent is too weighted.”

Moreover, the weights changed in 2009 because the Chancellor of D.C. public schools, Michelle Rhee, had negotiated a new deal with the teachers union. In exchange for 20 percent pay raises and bonuses of $20,000 to 30,000 for effective teachers, the district was given more leeway to fire teachers for poor performance, which they did using the IMPACT system. In part, this fight was spurred on because Obama-era Education Secretary Arne Duncan was doling out $3.4 billion in Race to the Top grants that focused on teacher effectiveness measures. Moreover, Rhee was a Chancellor because D.C. Mayor Adrian Fenty had passed legislation that would bypass the Board of Education and give him control of the schools.  

Yes, Wysocki might have been a false positive, but what about all of the poor performing teachers that the previous system hadn’t let go? By focusing on the teachers, O’Neil steers the conversation away from what should be the central concern, did the change actually help students learn and achieve?

Truth be told, my quibbles with Weapons of Math Destruction fit into two types. The first class relates to questions of emphasis and scope, which become important when the reader tallies off the costs and benefits of algorithms. Perhaps it is the case that “The U.S. News college ranking has great scale, inflicts widespread damage, and generates an almost endless spiral of destructive feedback loops.” But on the other hand, lower ranked colleges have decreased their net tuition and accepted a larger share of applicants. Yes, credit scores “open doors for some of us, while slamming them in the face of others,” but in which proportion? In Chile, for example, credit bureaus were forced to stop reporting defaults in 2012. The change was found to reduce the costs for most of the poorer defaulters, but raised the costs for non-defaulters, leading to a 3.5 percent decrease in lending and a reduction in aggregate welfare. It could be case that “the payday loan industry operates WMDs,” but it is unclear where low-income Americans will find short-term loans if they are outlawed.

Second, Weapons of Math Destruction continuously toys with important questions regarding the moral agency of technologies but never explicitly lays them out. How much value should be ascribed to technologies? To what degree are technologies value-neutral or value-laden? All technologies, including the algorithms that O’Neil describes, are designed and implemented for certain kinds of instrumental outcomes by companies and government agencies. An institution has to take on the task on adopting an algorithm for decision-making purposes, and thus, the algorithm reflects the institutional goals.

Should the algorithm be blamed, the institutional structures that put it into place, or some combination of the both? Reading with a careful eye, one will easily see that this is the fundamental question of the book, especially since O’Neil wonders whether “we’ve eliminated human bias or simply camouflaged it with technology.” But the real answer isn’t in this binary. Algorithmic problems are pluralist.