Thursday, February 26, 2015

The Ethics of Prediction

Predicting the future is probably the most valuable trick in the book. Well, maybe ACCURATELY predicting the future is the most valuable trick in the book, though a lot will still depend on who you’re trying to sell the book to. But if we knew the future, we would know not only that a sucker is born every minute, but also where and how fast. This is known as Heisenberg’s "circussy" principle.

We’re all familiar with one of the oldest tricks in the book: Insurance. Hell yeah, we want to know when we’re going to die so someone can bet us we’re not gonna. But despite the fact that your genetics and your family history are no longer admissible factors in eligibility for health insurance and employment, those snake eyes can see you right out of life and long term disability insurance, so you would have to be some of the people all of the time to play these odds.

This kind of gaffe really only works in your favor if you have the kind of dark secret Daphne Du Marier would have considered melodramatic. Insurance companies want to know how much they are probably going to have to shell out for you, how much more you are going to shell out to them, how hopeful they can be that you will be stabbed to death in the shower (which in insurance circles is considered the best possible outcome for everyone but you, and is thereby an ethical object of hope), and how long they can keep this outrageous racket up in a society of polysyllabic individuals, many of whom are capable of using tools and detecting the drawbacks of pissing into a sieve. At least I used to think so.

Some of the predictions that can be used to charge you money for insurance may incidentally be relevant to your health. This may be the one thing you want to know about as much as the people who want to sell you discount pacemakers and cremation packages. Unfortunately, these predictive tests are not as reliable as we would like. We can’t predict all outcomes with consistent regularity, and often we’re not quite sure which variables contribute to health outcomes. What happens when we tell you you’re going to die and you don’t? Talk about disappointing. People could go to jail for less! 

Which reminds me that they do.

Of course crimes are downright terrible. Obviously all crime should stop immediately, because it is against the law. Sure, it could make a lot of difference if we knew what caused a crime to be committed. Predicting incidence and recidivism makes a lot of sense. We could stop bad things from happening! The idea of taking a preventative approach to crime rather than a punitive one sounds lovely in theory, until you think about how we would actually do it. It certainly doesn't make sense to lock people up before they can commit crimes; and following an indicated individual until they eventually offend smells a little like entrapment, and definitely strikes me as unsportsmanlike. And would it even work?

How well can we predict these things? An integral aspect of any ethics assessment is a frank exploration of the efficacy or reliability of what is being assessed. In the case of predictive data, like with other prognostication, the answer is: It depends. I know, sorry. But if you say it in a fun mystical voice, it helps.

To make predictions for insurance agencies, we turn to a cold and ancient race of humanoids called “actuaries.” (Editor’s note: There are no facts in this paragraph.) Actuarial science is the inevitable product of ghastly pale dweebs rubbing their long fishy fingers together and hoping a piano falls on your head. At least I think so, and I already looked up way too much information for this post to bother checking this one out. Get out of here, you bloodless fiend! Anyone who is delighted to give me odds against my own defenestration or gleefully monitoring my day drinking can just take whatever numbers they’re excreting and shove them back up their asses. No thank you! 
(UPDATE: Whoops, looks like I confused "Actuary" with "Uriah Heep"? Not sure why. Sorry if you are an actuary, or whatever.)

Discussing predictors in health is trickier than just summarily dismissing an entire profession, because I happen to know slightly more than jack shit about it. Genetic testing and sequencing can provide insights into conditions that you may develop later in life that you might not have now, such as Huntington's, breast cancer, Alzheimer’s, and Parkinson’s. However, some of these traits are incompletely understood but known to be multi-factorial (involving more than one gene or exposure), which makes a reliable prediction of risk complicated. This makes many providers reluctant to offer these tests, since there is concern that these results could adversely impact health by erroneously influencing decision making, or causing undue alarm when there is no cause for concern.

Our method for predicting crime is even more bizarre, because we have been using measures about as efficacious as a splint on a shit for DECADES and we don’t even care. Moreover, we have more recently begun employing these variably useless measures to determine who can just get out of an overcrowded jail free, and who needs to serve out their sentences, as if this makes any goddamn sense at all. Most surveys designed to predict repeat criminal behavior are reliant on self-report data, both for the individual they are assessing, and the aggregate patterns they identify. This means that all of the things we think we know about the behavior of people in jail comes from what people in jail tell us.

N.B. I am not suggesting that people in jail are less reliable than other people (I’ll leave that to the courts), but I do recall that in high school I took a fiendish delight in derailing official surveys, and I don’t think that it is beyond the bounds of possibility that incarcerated people might demonstrate at least comparable delinquency to your average high school student. 

So just how accurate are these predictors? One moribund study from an era in which LEAD WAS STILL PLAYING A SIGNIFICANT ROLE IN CRIMINALITY suggests that these surveys can predict repeat crime with up to 70% reliability, though it is worth noting that the crime rate in the 1980s was the highest in U.S. history and so it was a pretty safe bet to assume that EVERYBODY was going to break the law eventually. Given that inconsistent variables, methods, and most importantly, transparency have made an overall judgment of this predictive measure impossible, it remains unclear whether or not these analytics contribute anything of value, or if, perhaps, they result in greater damages than they prevent. 

Prisons are horrifically over-crowded, contributing to rising costs that are exceeding 70 BILLION DOLLARS per year. A 2007 National Institute of Corrections sponsored policy analysis suggests that prison populations could be reduced by up to 50% by limiting incarceration for parole and non-violent violations, and reducing mandatory sentencing for low level crime like drug sales. Why are so many people in jail?

Some schools of thought surrounding criminality assume a significant genetic component to deviant behavior. While there are some genetic disorders that are associated with violent behavior, such as the Y chromosome, there is no evidence of any "aggression" gene that operates to escalate violent tendencies in individuals. This makes sense, since as noted above, most complex traits are multi-factorial. But aside from the factual irrelevance, could genetic predictions about criminality be harmful?

The concerns that we have about predictive tests in health are analogous to many of the concerns that arise about predictions anywhere. First of all, what if the test is wrong? If the outcome we predict is “Criminality,” an individual may be unjustly stigmatized. We also worry about compromising what is known as an “open future.”

To have an “open future” means to have one that has not been narrowly defined by external forces, or rather, MORE narrowly defined than it is already. Our ascribed characteristics such as our race, our class, and our social context can define our future more cogently than a deadly disease, of course, but there is something unnecessarily fatalistic about forecasting a hideous demise. One of the primary arguments against testing children for adult onset disorders like Huntington’s is the concern that this information will circumscribe a child’s options, and that a life may be lived differently with this knowledge. (Of course, if you use slightly different words this is also an argument for testing children, and now you know why people in ethics are such assholes all the time.)

There is also the concern that these predictions contribute negatively to the self-concept and identity of the individual, by defining them as “sick,” “criminal,” or “disabled.” We know that labeling can be harmful. By predicting relapses in criminal behavior, are we also encouraging it? An individual whose future has already been defined by poverty and racism may have no choice but to accept an identity further narrowed by predictions of criminal behavior.

It is also worth noting that predictive tests have implications for family members as well as individuals. Genes and environment are often shared, so it is not unlikely that these predictions could have relevance to others. 

Many, many others. 

The United States has more incarcerated criminals than any other country. With over 2.2 million individuals in prison, that's a lot of families, and a lot of stakeholders. That's a lot of people that these arbitrary predictions may be influencing, and it's worth thinking about who those people are. 

Of the 2.2 million people in our prison system today, an overwhelmingly disproportionate number are minorities. While there are numerous factors advanced for this disparity, the fact remains that the prison population is not representative of the overall population, and instead is inequitably representative of minorities and persons living in poverty.  With predictions that can influence such a disproportionate number of individuals, and their families, it makes sense to wonder who is making these predictions, and how they are being made. 

This is a process that is contributing to the subjugation of a lot of people. Who is responsible for it? In a way, we all are. We pay the taxes that fuel the justice system (over 70 BILLION dollars a year!), and we elect the judges that keep the system running. We pay taxes that support universities and public institutions, but the research that they produce often sits behind a pay wall, or is presented in a forum we don't have access to. But this research is informing these policies, and we are at a loss to understand why. We can ask. We can ask our elected representatives to be more transparent about the policies and practices behind parole and risk assessment, and we can look at statistics like this one, that say the average prison time served by an African American for a drug offense (58.7 months) is nearly the equivalent of the average prison term served by whites for violent offenses (61.7 months.) 

To say that this is an appalling disparity is anemic. We are not required to passively allow a harmful, discriminatory power differential to perpetuate itself, and we are certainly not required to pay for it. Furthermore, if we can predict trends in crime, doesn't that suggest that we are developing a more nuanced understanding of the factors that influence crime? Why wouldn't we focus on mitigating those factors as opposed to predicting recidivism in individuals? if we already have this information, it seems to me that we could be finding more responsible applications for it.