Tag: cathy o’neil

Notes on O’Neil, Chapter 2, “Bomb Parts”

Continuing with O’Neil’s Weapons of Math Destruction on to Chapter 2, “Bomb Parts”. This is a popular book and these are quick chapters. But that’s no reason to underestimate them! This is some of the most lucid work I’ve read on algorithmic fairness.

This chapter talks about three kinds of “models” used in prediction and decision making, with three examples. O’Neil speak highly of the kinds of models used in baseball to predict the trajectory of hits and determine the optimal placement of people in the field. (Ok, I’m not so good at baseball terms). These are good, O’Neil says, because they are transparent, they are consistently adjusted with new data, and the goals are well defined.

O’Neil then very charmingly writes about the model she uses mentally to determine how to feed her family. She juggles a lot of variables: the preferences of her kids, the nutrition and cost of ingredients, and time. This is all hugely relatable–everybody does something like this. Her point, it seems, is that this form of “model” encodes a lot of opinions or “ideology” because it reflects her values.

O’Neil then discusses recidivism prediction, specifically the LSI-R (Level of Service Inventory–Revised) tool. It asks questions like “How many previous convictions have you had?” and uses that to predict likelihood of future prediction. The problem is that (a) this is sensitive to overpolicing in neighborhoods, which has little to do with actual recidivism rates (as opposed to rearrest rates), and (b) e.g. black neighborhoods are more likely to be overpoliced, meaning that the tool, which is not very good at predicting recidivism, has disparate impact. This is an example of what O’Neil calls an (eponymous) weapon of math destruction.(WMD)

She argues that the three qualities of a WMD are Scale, Opacity, and Damage. Which makes sense.

As I’ve said, I think this is a better take on algorithmic ethics than almost anything I’ve read on the subject before. Why?

First, it doesn’t use the word “algorithm” at all. That is huge, because 95% of the time the use of the word “algorithmic” in the technology-and-society literature is stupid. People use “algorithm” when they really mean “software”. Now, they use “AI System” to mean “a company”. It’s ridiculous.

O’Neil makes it clear in this chapter that what she’s talking about are different kinds of models. Models can be in ones head (as in her plan for feeding her family) or in a computer, and both kinds of models can be racist. That’s a helpful, sane view. It’s been the consensus of computer scientists, cognitive scientists, and AI types for decades.

The problem with WMDs, as opposed to other, better models, is that the WMDS models are unhinged from reality. O’Neil’s complaint is not with use of models, but rather that models are being used without being properly trained using sound sampling on data and statistics. WMDs are not artificially intelligences; they are artificial stupidities.

In more technical terms, it seems like the problem with WMDs is not that they don’t properly trade off predictive accuracy with fairness, as some computer science literature would suggest is necessary. It’s that the systems have high error rates in the first place because the training and calibration systems are poorly designed. What’s worse, this avoidable error is disparately distributed, causing more harm to some groups than others.

This is a wonderful and eye-opening account of unfairness in the models used by automated decision-making systems (note the language). Why? Because it shows that there is a connection between statistical bias, the kind of bias that creates distortions in a quantitative predictive process, and social bias, the kind of bias people worry about politically, which consistently uses the term in both ways. If there is statistical bias that is weighing against some social group, then that’s definitely, 100% a form of bias.

Importantly, this kind of bias–statistical bias–is not something that every model must have. Only badly made models have it. It’s something that can be mitigated using scientific rigor and sound design. If we see the problem the way O’Neil sees it, then we can see clearly how better science, applied more rigorously, is also good for social justice.

As a scientist and technologist, it’s been terribly discouraging in the past years to be so consistently confronted with a false dichotomy between sound engineering and justice. At last, here’s a book that clearly outlines how the opposite is the case!


Reading O’Neil’s Weapons of Math Destruction

I probably should have already read Cathy O’Neil’s Weapons of Math Destruction. It was a blockbuster of the tech/algorithmic ethics discussion. It’s written by an accomplished mathematician, which I admire. I’ve also now seen O’Neil perform bluegrass music twice in New York City and think her band is great. At last I’ve found a copy and have started to dig in.

On the other hand, as is probably clear from other blog posts, I have a hard time swallowing a lot of the gloomy political work that puts the role of algorithms in society in such a negative light. I encounter is very frequently, and every time feel that some misunderstanding must have happened; something seems off.

It’s very clear that O’Neil can’t be accused of mathophobia or not understanding the complexity of the algorithms at play, which is an easy way to throw doubt on the arguments of some technology critics. Yet perhaps because it’s a popular book and not an academic work of Science and Technology Studies, I haven’t it’s arguments parsed through and analyzed in much depth.

This is a start. These are my notes on the introduction.

O’Neil describes the turning point in her career where she soured on math. After being an academic mathematician for some time, O’Neil went to work as a quantitative analyst for D.E. Shaw. She saw it as an opportunity to work in a global laboratory. But then the 2008 financial crisis made her see things differently.

The crash made it all too clear that mathematics, once my refuge, was not only deeply entangled in the world’s problems but also fueling many of them. The housing crisis, the collapse of major financial institutions, the rise of unemployment–all had been aided and abetted by mathematicians wielding magic formulas. What’s more, thanks to the extraordinary powers that I loved so much, math was able to combine with technology to multiply the chaos and misfortune, adding efficiency and scale to systems I now recognized as flawed.

O’Neil, Weapons of Math Destruction, p.2

As an independent reference on the causes of the 2008 financial crisis, which of course has been a hotly debated and disputed topic, I point to Sassen’s 2017 “Predatory Formations” article. Indeed, the systems that developed the sub-prime mortgage market were complex, opaque, and hard to regulate. Something went seriously wrong there.

But was it mathematics that was the problem? This is where I get hung up. I don’t understand the mindset that would attribute a crisis in the financial system to the use of abstract, logical, rigorous thinking. Consider the fact that there would not have been a financial crisis if there had not been a functional financial services system in the first place. Getting a mortgage and paying them off, and the systems that allow this to happen, all require mathematics to function. When these systems operate normally, they are taken for granted. When they suffer a crisis, when the system fails, the mathematics takes the blame. But a system can’t suffer a crisis if it didn’t start working rather well in the first place–otherwise, nobody would depend on it. Meanwhile, the regulatory reaction to the 2008 financial crisis required, of course, more mathematicians working to prevent the same thing from happening again.

So in this case (and I believe others) the question can’t be, whether mathematics, but rather which mathematics. It is so sad to me that these two questions get conflated.

O’Neil goes on to describe a case where an algorithm results in a teacher losing her job for not adding enough value to her students one year. An analysis makes a good case that the cause of her students’ scores not going up is that in the previous year, the students’ scores were inflated by teachers cheating the system. This argument was not consider conclusive enough to change the administrative decision.

Do you see the paradox? An algorithm processes a slew of statistics and comes up with a probability that a certain person might be a bad hire, a risky borrower, a terrorist, or a miserable teacher. That probability is distilled into a score, which can turn someone’s life upside down. And yet when the person fights back, “suggestive” countervailing evidence simply won’t cut it. The case must be ironclad. The human victims of WMDs, we’ll see time and again, are held to a far higher standard of evidence than the algorithms themselves.

O’Neil, WMD, p.10

Now this is a fascinating point, and one that I don’t think has been taken up enough in the critical algorithms literature. It resonates with a point that came up earlier, that traditional collective human decision making is often driven by agreement on narratives, whereas automated decisions can be a qualitatively different kind of collective action because they can make judgments based on probabilistic judgments.

I have to wonder what O’Neil would argue the solution to this problem is. From her rhetoric, it seems like her recommendation must be prevent automated decisions from making probabilistic judgments. In other words, one could raise the evidenciary standard for algorithms so that they we equal to the standards that people use with each other.

That’s an interesting proposal. I’m not sure what the effects of it would be. I expect that the result would be lower expected values of whatever target was being optimized for, since the system would not be able to “take bets” below a certain level of confidence. One wonders if this would be a more or less arbitrary system.

Sadly, in order to evaluate this proposal seriously, one would have to employ mathematics. Which is, in O’Neil’s rhetoric, a form of evil magic. So, perhaps it’s best not to try.

O’Neil attributes the problems of WMD’s to the incentives of the data scientists building the systems. Maybe they know that their work effects people, especially the poor, in negative ways. But they don’t care.

But as a rule, the people running the WMD’s don’t dwell on these errors. Their feedback is money, which is also their incentive. Their systems are engineered to gobble up more data fine-tune their analytics so that more money will pour in. Investors, of course, feast on these returns and shower WMD companies with more money.

O’Neil, WMD, p.13

Calling out greed as the problem is effective and true in a lot of cases. I’ve argued myself that the real root of the technology ethics problem is capitalism: the way investors drive what products get made and deployed. This is a worthwhile point to make and one that doesn’t get made enough.

But the logical implications of this argument are off. Suppose it is true that “as a rule”, the makers of algorithms that do harm are made by people responding to the incentives of private capital. (IF harmful algorithm, THEN private capital created it.) That does not mean that there can’t be good algorithms as well, such as those created in the public sector. In other words, there are algorithms that are not WMDs.

So the insight here has to be that private capital investment corrupts the process of designing algorithms, making them harmful. One could easily make the case that private capital investment corrupts and makes harmful many things that are not algorithmic as well. For example, the historic trans-Atlantic slave trade was a terribly evil manifestation of capitalism. It did not, as far as I know, depend on modern day computer science.

Capitalism here looks to be the root of all evil. The fact that companies are using mathematics is merely incidental. And O’Neil should know that!

Here’s what I find so frustrating about this line of argument. Mathematical literacy is critical for understanding what’s going on with these systems and how to improve society. O’Neil certainly has this literacy. But there are many people who don’t have it. There is a power disparity there which is uncomfortable for everybody. But while O’Neil is admirably raising awareness about how these kinds of technical systems can and do go wrong, the single-minded focus and framing risks giving people the wrong idea that these intellectual tools are always bad or dangerous. That is not a solution to anything, in my view. Ignorance is never more ethical than education. But there is an enormous appetite among ignorant people for being told that it is so.


O’Neil, Cathy. Weapons of math destruction: How big data increases inequality and threatens democracy. Broadway Books, 2017.

Sassen, Saskia. “Predatory Formations Dressed in Wall Street Suits and Algorithmic Math.” Science, Technology and Society22.1 (2017): 6-20.