Are Humans Atoms?

supercrunchers2.jpegIan Ayres’s Super Crunchers lionizes a new set of heroes for the information age: data-driven analysts who take large sets of figures and come up with counterintuitive conclusions. Bill Henderson suggests that Ayres is part of a worthy tradition:

To my mind, the principles of Moneyball-Freakanomics-Super Crunchers are hard to dismiss. Indeed, over the last several years, I cannot recall a single frontal attack on any of these ideas. Instead, it is my observation that most people (and thus organizations and institutions) never take the time to think through their simple, stark implications–i.e., don’t be complacent, look at the data, or eventually you are going to get smoked by the competition.

I think there is a good sociological critique of Freakonomics, which I’ll quote at the end of this post. But since Ayres’ book is just published, let me heretically point out the possibly self-defeating and value-corroding aspects of the data-driven approaches he extols.

1) From self-defeating to stratifying?: Ayres is pretty enthusiastic about a program called “Title Scorer,” which lets you “predict the likelihood of [a novel] becoming a best-seller” via statistical analysis of past best-sellers. It “guessed right in 70 percent of cases” in a given study. But now that it’s public knowledge, isn’t everyone going to start using it? And as they do, won’t their advantage relative to others narrow?

This is a key difference between the human and natural sciences: I can observe regularities in nature for some time and (pace Heisenberg) my publishing my observations is not going to cause molecules or plants to act any differently. But humans can change their behavior on the basis of knowledge about how others behave. As Jon Elster states, “In parametric rationality each person looks at himself as a variable and at all others as constants, whereas in strategic rationality all look upon each other as variables.” More strategically rational individuals will undermine Title Scorer’s ability to predict.

Of course, one solution here is to further tweak the equation and then keep it secret. But do we really want a world of hidden persuaders supercharged by Super Crunching? And let’s not forget how differential access to such trade secrets can exacerbate positional competitions.

That leads to a more serious concern about this data-driven analysis. . .

2) Bringing out the worst in us?: Ayres also enthuses about a pair of hedge fund managers who want to build a fund that underwrites movies based on Epagogix–a “neural network [that] . . . predict[s] a movie’s receipts primarily on characteristics of a script.” (145). I’m happy that Epagogix is advising studios to spend less money on stars and more on unknown actors. But think about how this black-box predictor may work. What if words like “murder,” “violence,” or various epithets or sex, really drive up sales? Do we really want an ever more violent and sexualized movie biz? Moreover, we know that tastes are endogenous to current media products–those of the young and impressionable can be as much driven by as driving content.

We often hear about worries that individuals who would never tell a pollster that they would never vote for women, nevertheless may allow that secret prejudice to influence their decision in the privacy of the polling booth. Will Ayres-style number-crunching allow such prejudices to enter more and more spheres of life–spheres where once people had to “launder preferences” by giving some explicit reason for action, rather than basing it on a black box numerical prediction?

Ayres ends his vignette with the hedge fund managers behind Epagogix chuckling over the movie industry’s outdated resistance to their program. This brought to mind to me David Geffen’s falling out with Rick Rubin over a controversial new act. Here’s the story, as related by the New York Times:

At Def American, Rubin concentrated on a harder rock sound: Slayer’s “Reign in Blood,” which is considered to be a heavy metal classic, or the Geto Boys, whose rap song “Mind of a Lunatic” depicted vivid scenes of necrophilia and murder. “I just couldn’t put out a record about sex with dead bodies and cutting off women’s breasts,” said David Geffen, whose company Geffen Records was the distributor of Def American. “I begged Rick not to put out the Geto Boys. In the end, I lost. He left and went to Warner Brothers.”

Rubin said he had “artistic reasons” for pushing the album. But if for some reason the gruesome music he was supporting resonates with individuals’ darker sides, do we really want a supercrunching program to promote such music and excuse its supporters from giving reasons for it?

PS: By the way, here’s that excerpt from Kieran Healy’s review of Freakonomics:

In an ideal world, Levitt would represent a new breed of empirically oriented researcher working across the boundaries of social science disciplines, expanding our knowledge of a wide variety of social processes and perhaps laying the foundations of a new kind of social science, with the help of people working at equivalent boundaries in other fields such as cognitive psychology and economic sociology. We can hope for the best, but there is cause for pessimism. Economics has a bad habit of routinely rediscovering (and taking credit for) ideas that are well-established elsewhere.

Sometimes, whole fields are victimized in this way—social networks,institutional analysis, and culture—as smart economists assume an idea that is new to them, is new to everyone, and go off and reinvent some wheels. Freakonomics does have some sociology and psychology lurking in its footnotes. A stronger engagement with Stanley Lieberson’s work on shifting trends in baby-naming practices would have made Chapter 6 a lot more interesting, for instance. There are one or two instances where a well-researched idea is presented as though no one had ever thought of it before. Roland Fryer’s working paper on ‘‘The Economics of Acting White’’ is cited as the source of the notion that ‘‘some black students ‘have tremendous disincentives to invest in particular behaviors … due to the fact that they may be deemed a person who is trying to act like a white person.’’’ This idea was articulated just in this form in the mid-70s by John Ogbu (as the ‘‘oppositional culture’’ hypothesis) and arguably also by James Coleman in The Adolescent Society, published in 1961. Perhaps a mention of the large literature on the topic would have detracted from the Freakonomic buzz.

Frank Pasquale

Frank is Professor of Law at the University of Maryland. His research agenda focuses on challenges posed to information law by rapidly changing technology, particularly in the health care, internet, and finance industries.

Frank accepts comments via email, at All comments emailed to may be posted here (in whole or in part), with or without attribution, either as "Dissents of the Day" or as parts of follow-up post(s). Please indicate in your comment whether or not you would like attribution, or would prefer your comment (if it is selected for posting) to be anonymous.

You may also like...

3 Responses

  1. Dudley says:

    This highlights the concern I have about “Hume’s law”–that is does not imply ought, or as phrased by GE Moore, that the good is a non-natural kind. In other words, just because a social phenomena is recognized as emergent, doesn’t end the inquiry–we need to ask whether this is the preferred outcome, or whether we can take steps to reach a different equilibrium.

    A fascinating case study was provided by Thomas Schelling (in Micromotives and Macrobehavior), describing the city of Oak Park’s efforts to maintain a diverse neighborhood by prohibiting ‘for sale’ signs, and insuring home owners from any decline in their home value. Free marketers would’ve hated the supposed ‘government intervention,’ but Oak Park’s actions actually helped increase its economic competitiveness, while serving laudible moral goals.

  2. Frank,

    Great post. My only comments is that your critique is primarily normative — that reliance an data will erode our humanity. In some cases that this surely right.

    But in a world that rewards results, isn’t the genie out of the proverbial bottle? The tools themselves are here to stay. So let’s use them appropriately.

  3. There’sa lot of problems here:

    1) Main insight: It’s easy to overpay for something.

    This is true for stocks, baseball players, and movie stars. And ebay auction items.

    Knowing this, is a good thing.

    But it helps less that you might think – there’s a lot of social factors that push people to overpay.

    It’s not a bad insight – but using it is harder than it looks. Like “buy low, sell high”.

    2) Stories. Beware the data-miner with a story. Too many of these stories turn out to be excuses for bad policy.

    3) Sample limitations. With data-mining, often you get a lot of me-too’s since the model replicates the successes of the sample set. But it doesn’t tell you what other models might work. It turns into “formula”.

    And regarding: “do we really want a supercrunching program to promote such music and excuse its supporters from giving reasons for it”

    You don’t need any sort of complicated mathematics to knowthat sex and violence sell.