Wednesday, November 30, 2016

Weapons of Math Destruction


Weapons of Math Destruction: How Big Data increases inequality and threatens democracy.
Cathy O'Neil (Crown, 2016)

   Cathy O'Neil is a math nerd who, in 2007, left her academic career to do math for a major hedge fund. She thus had a front row seat for the financial catastrophe of 2008, which was in large part a product of the kind of mathematical processes people like her were engaged in. "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 that I now recognized as flawed."

   After leaving D. E. Shaw, O'Neil recast herself as a data scientist, working on models in the Big Data economy, and once again privy to the magic formulas that are taking over in so many areas of our lives, from college admissions to parole recommendations. She started her blog, MathBabe, "to mobilize fellow mathematicians against the use of sloppy statistics and biased models that created their own toxic feedback loops."

   In Weapons of Math Destruction, she expands that warning to general readers, including the mathematically challenged. In fact, for the kind of algorithm she describes as a WMD, the math is usually opaque, anyhow: it's proprietary to the company that is profiting from it, whether by sorting your resume by the ZIP code you come from, or selling your search clicks to on-line marketers. The black-box quality is part of what makes WMDs so dangerous: the numbers they spit out are nearly impervious to challenges, even though the numbers that went into them may be biased, false, or completely spurious.

   Another dangerous aspect of WMDs is its potential to damage people's lives. O'Neil has a case study about the Washington, D.C., school department's program of rewarding teachers whose students improve, and firing those whose students don't. Such a practice can (and did) lead to the firing of good teachers, if the previous year's teachers cheated by padding their students' scores. The school hierarchy got what it wanted, to be seen as weeding out underperforming teachers, but since they didn't do any external checking to see if that's actually what they had done, they don't know how many good teachers they lost in the process. A healthy model would have an independent way of looking at the results to see if they made sense.

   That's unlikely to happen, however, when the third destructive effect starts to work: scale. If your credit report were managed by your own bank, you could speak with them about it face to face, and presumably establish that you were not the same John Bradshaw who had defaulted on that electric bill three years ago. But scale it up to the size of the big three credit bureaus, over billions of data points, and you are unlikely to find a person who can fix mixups; but you may very well pay for the errors not only in higher interest rates, but also in difficulty getting a loan, or a job, at all.

   And that's to mention only the official credit bureaus, which are governed by requirements that let you see the data they're using, and challenge it. E-scores generated by studies of internet use, or assumption about the street you live on, are under no such constraint, and their feedback loops tend to make unfairness worse. "There's a very high chance that the e-scoring system will give the borrower from the rough section of East Oakland a low score. A lot of people default there. So the credit card offer popping up on her screen will be targeted to a riskier demographic. That means less available credit and higher interest rates for those who are already struggling."

   Most WMDs embody corporate goals such as efficiency and profit; if corporations are persons, they tend to be sociopathic ones. Human beings are much better at thinking about justice than computers are, still, and perhaps always–if we choose to, and if we know what we're up against. Weapons of Math Destruction is disturbing, and distressing, but I couldn't put it down. Cathy O'Neil is a warrior for economic justice, and we ignore her at our peril.

Tuesday, November 1, 2016

Bad English


Bad English: A History of Linguistic Aggravation
Ammon Shea (Perigee, 2014)

   The t-shirt that says "I'm silently correcting your grammar" has my name on it. I come by it honestly: I have linguistically sensitive forebears on all sides. My father winced at sentence adverb 'hopefully'; my maternal grandmother loathed the word 'tasty'; and I have recently felt utterly compelled to fix the number disagreement in the last sentence of the Lord's Prayer.

   But I really don't want to be a jerk about it, so I'm delighted to add Ammon Shea's Bad English to my 'language wars' shelf. Shea has taken a serious historical look at a the usage rants and grammar guides of the past century and a half. English, it turns out, changes so quickly that no guide can hope to be the last word. The peeves of the nineteenth century very often look strange to us now; the expressions that were then considered beyond the pale have either sunk out of sight or become commonplace and unobjectionable.

   Such a change doesn't even need centuries. Remember 'Winston tastes good, like a cigarette should'? "Here we have an extraordinarily clear line of linguistic demarcation. People either feel that using like as a conjunction marks one as essentially subliterate or they have absolutely no idea what you are talking about and fail to see why this would be a problem of any sort." There were many more of the first group around when I was little, and there are many more of the latter now. 
 
   We will always have doomsayers, and yet, says Shea, "English is not dying. It is behaving exactly as it should, which is to say that it is changing. All living languages change–it is one of the things that indicate that they are still in use by a large number of people. The problem is that, while many people accept that our language is subject to change, they want to dictate what sort of changes will take place and that is a very difficult thing to do."

   The rules and roadblocks set up by fourth-grade English teachers frequently have perverse effects, as either the teacher or the student remembers the rule but not the principle. Years of drilling students not to say 'Jimmy and me are going to the pool' has led to generations of people who say 'Between you and I.' Shea makes an amusing example of George Orwell's famous essay, "Politics and the English Language," which promulgates six reasonable-sounding rules, and breaks five of them with abandon. There must be some sort of rule about that...

   Shea is not out to stop you speaking English the way you want to. If you prefer never to split an infinitive or strand a preposition, more power to you; but don't imagine that you are defending rules handed down on stone tablets. And feel free to set a picket fence around unique and perfect, but you may also want to "accept that certain words... are used by some people in a less semantically exact manner than you would yourself employ and hope that they have some other redeeming qualities that make up for this lapse."

   Shea is being contrarian here, and a little argumentative; but I think I prefer that attitude to the certainty and superiority of the self-appointed guardians of the language. English is doing just fine; it can look out for itself.


Email edition 11/1/16