Pretending (Not) to Count

Datafication, online societies, and the war on academia.

Earlier this month, I participated in a lively discussion with Marion Fourcade about her excellent new book with Kieran Healy, The Ordinal Society. This was part of a series of engaging AI & Society salons organized by Irene Chen, Colleen Chien, and Deb Raji at Berkeley. I intended to blog about this when it happened, but got sidelined by a combination of some fun extraacademic activities and a head cold.

The Ordinal Society is primarily a critique of platform capitalism, the sorting it induces upon us, and how people act to please this algorithmic distortion of their behavior. But, wearing my predilections on my sleeve, I read it as an insightful qualitative critique of quantitative social science. Certainly, online platforms are implementing behavioral economics and social regression at scale. But issues with this quantified mindset manifest with small n too. The societal structures induced by deep learning algorithms and their classification errors differ in degree from those of antecedent regression methods. It is the quantification and quantization that are the root cause of the troubling feedback loops examined in the book.

One passage from the book that captures the heart of the critique makes an argument only controversial to those who have been too steeped in the mindset of overquantification

“As social phenomena, naming and ranking are much more general than quantification. Catechisms, shibboleths, purity regimes, ritual compliance, degradation ceremonies—in short, symbolically infused distinction making in all its forms—can serve in the place of numerical measures of position. Insofar as they are about distinguishing better from worse, as opposed to simply affirming the uniqueness of every single thing in the world, qualitative modes of classification can be as powerfully disciplining as quantitative ones.”

A quantitative mindset would assert instead that naming and ranking are quantitative. This mindset asserts that elicitation of preferences is quantifying. Preferences are just another way of describing “utility,” and we can construct optimal actions just by laying out our rankings in a systematic way.

This is the modern economic view, crystallized in post-war applied mathematics. In the computer age, we decided that ranking things, listing preferences, and then acting on those preferences was formalizable as mathematics. But Fourcade and Healy remind us that people and societies rank, classify, and choose without the formal scaffolding of Bayesian decision theory. And they have done so throughout history.

Healy and Fourcade continue:

“What is of real interest is the fusion of socially fundamental processes of naming and ranking with novel tools and data for carrying out those tasks. Large-scale measurement allows for thinking about scores and ranks through the lens of small differences and high dimensionality. What does it mean for computers to intervene in the business of seeing and organizing society?”

It’s always interesting how the quantitative person sees paradoxes as a math challenge and not a suggestion that quantification is a language game, and that the mathematical language is limited just like qualitative language.

I was thinking about this recently as I was working my way through Paul Meehl’s posthumous monograph, The Seven Sacred Cows of Academia, summarizing Meehl’s thoughts on how to make higher education more cost effective. Given the vicious war on higher education being waged by the Trump administration, I’ve been reading more about the long history of right-wing agitation against the academy and the associated academic responses. Every time, you’ll find academics writing their version of the Onion article “The Worst Person You Know Just Made A Great Point.” Here’s a great recent example by Anna Krylov, where I agree with everything the author writes, even though saying these things out loud right now only further imperils our institutions. I’ll blog about Meehl’s version of this meme once I’ve worked my way through the book (it’s 180 rambly pages).

I got stuck at the end of the introduction, where Meehl writes a six-page screed against those unwilling to quantify academic research merit, entitled “Miscounting by Pretending Not to Count.”

“When one suggests that some sort of composite index of these plausible criteria of scholarly merit should be used by the institutional decision-maker in sorting departments into the pure teaching and the mixed research and teaching categories, I find many persons, including psychologists (not, I am happy to report, all of them), express objections on the grounds that these things cannot reliably be quantified.”

For Meehl, numbers are more objective, even if the measures are unreliable. He rails against the imprecision of “a subjective, impressionistic pseudo-quantifying, instead of counting and measuring.” He firmly believes that using numbers and systems of numbers brings more clarity to decision making, and doing otherwise is irrational.

“The basic mistake here is not realizing that when you’re talking about things that exist in degree or frequency (e.g., how often, or how strong, or how intense), you are engaging in a tallying or measuring operation whether you use words or numbers. The main difference is that the words are less precise and less objective than the numbers.”

The idea that counts are more “objective” than words is a classic blindspot of the sciences, and I’m sure many people here reading it agree with Meehl. “We want data, not anecdotes” is the motto of quantified America. Fourcade and Healy write a well-needed reminder that organizing a society around unreliable counts bakes in a huge helping of irrationality.

And even the most adherent capital-R Rationalists know this to be true if you press them on it a bit. Assuming that classification and ranking are quantitative and building a mathematical framework for decision making on top leads to a system riddled with paradoxes and small-mindedness. We choose to ignore the paradoxes because we normatively decide that a quantified system is better than the alternatives. But this faith in quantification leads to institutional inertia that makes the university miserable for all of us (Meehl) or a society with pernicious hidden hierarchies that we all tailor our behavior to appease (Fourcade and Healy).