In Steven Spielberg’s movie “The Minority Report” (with Tom Cruise) and in the original 1956 science-fiction novel by Philip K. Dick, mutants called “precogs” can predict future crimes – “precrimes” – so that their authors can be arrested in advance. “In our society,” says the head of the precrime police, “we have no major crimes. But we have a detention camp full of would-be criminals.”
This is not only science fiction. Preventing precrimes is only one application of a more general phenomenon: statistical justice, that is, treating individuals according to the intentions or actions of some statistical groups they are put into.
“Big data” is helping to fuel statistical justice. Big data describes the process whereby masses of data are assembled and analyzed by computers to find hidden correlations. A new academic field, data science, has been born.
Big data is useful in many private undertakings, from predicting the most efficient baseball strategies, like in the movie Moneyball, to insurance companies predicting which ones of their applicants will be most likely be victims of the insured peril. Insurance predictions are often made on the basis of correlations with data that are not causally related (your credit history, for example).
The problem of statistical justice occurs when big data is used by government to discriminate against individuals on the basis of mere statistical probabilities, without the alternatives that competition provides on the market. The ultimate goal would be to predict that a specific individual will commit a crime, to find him guilty of a precrime.
Bloomberg recently reported that a big-data precrime system is under development in China, aimed at predicting who will become a terrorist, by comparing the characteristics and circumstances of “pre-terrorists” with those of past terrorists.
In America, the Los Angeles Police Department and other police agencies already use bid-data systems to forecast when and in which area crimes are more likely to happen. This is not necessarily problematic, assuming that the police is after real crime, as opposed to victimless ones. What is more problematic is when certain American judges use big-data systems for sentencing and making parole decisions in individual cases: it amounts to predicting who will reoffend.
In May, the White House published a report on big data and civil rights. Its chapter on criminal justice is full of the usual platitudes, such as the goal “to develop additional best practices for fair and ethical use of big data techniques” and “to reduce discrimination and advance opportunity, fairness, and inclusion.” Missing is the principle of not treating an individual on the basis of a statistical group he is in. This is alas not surprising because the very opposite principle is what today’s public policy is based upon.
Statistical justice is not justice. An individual cannot be morally responsible for the deeds of a statistical group – past criminals in similar circumstances – on which he has no control and of which he may not even be aware.
An example? Using FBI statistics, we can calculate that if all males were incarcerated from their 13th birthday to their 30th, the number of homicides would drop by at least 35%. Incarcerating only black males in this age range would reduce murders by at least 23%. This of course assumes that those who would not have committed murders don’t learn crime in jail. It ignores other possible unintended consequences.
Incarcerating only those are actually precriminals is of course impossible because uncertainty is, by definition, part of statistical analysis. The prediction that a specific individual will commit a crime will always have a probability of less than 100%. Not to mention that an individual has free will.
Statistical justice also creates perverse incentives. If you punish all individuals of a group for something that other individuals have done, you don’t give proper incentives to any individual not to do it.
We must also consider the incentives of government, that is, of the individuals – mainly politicians and bureaucrats – who actually adopt and enforce laws. If they have the power to control whole groups of individuals on the basis of statistical analyses they make and interpret themselves, they are likely to often use this power in their own interest.
The more government attempts to predict and control pre-crimes, the more crimes government agents will commit.