In science's hands.

Photographer: Frank Perry/AFP/Getty Images

The Prosecution Cannot Rest on a Trade Secret

Noah Feldman is a Bloomberg View columnist. He is a professor of constitutional and international law at Harvard University and was a clerk to U.S. Supreme Court Justice David Souter. His books include “Cool War: The Future of Global Competition” and “Divided by God: America’s Church-State Problem -- and What We Should Do About It.”
Read More.
a | A

On the surface, TrueAllele Casework, a computer program that extracts genetic profiles from DNA samples, would seem to mark an advance in criminal justice technology. But defense lawyers say it shouldn’t be allowed in court, because Cybergenetics Corp., the firm that owns the program, won’t reveal the software’s source code, which it considers a trade secret.

The resulting conflict, which is presently playing out in a Pennsylvania murder trial, poses fascinating and important questions: Do we need to know exactly how a given technique works to consider it scientifically reliable and admissible in court? And is it democratically right to convict, and possibly execute, someone based on a secret process the defendant isn’t allowed to know?

Start with the science. To oversimplify a bit, ordinary DNA analysis depends on qualitative comparisons made by human beings. Typically, a technician will type a defendant’s DNA, then compare it to DNA in samples found at the crime scene. By comparing peaks and valleys in the statistical representation of the DNA sequences, the technician determines the likelihood that the two samples match.

This technique doesn’t work if multiple samples are mixed together, because the comparison is too difficult to make. The same is true when the DNA sample is too small.

Quantitative analysis of the samples avoids these problems. If performed correctly, it should yield more robust conclusions, because a computer can use all or nearly all the information contained in the sample. And it’s free from the subjective biases of a human interpreter.

TrueAllele is a commercially available program that teases out different DNA signatures from mixed samples. It uses a simulation of the kind first invented for detecting high-energy particles in physics experiments, then deploys statistical analysis to offer a prior probability of what results would be expected in a given population and a likelihood of what’s in the mixture.

Scientifically, this is all perfectly reasonable. A New York state court that reviewed TrueAllele earlier this year noted that there have been many peer-reviewed articles published in connection to the program, and also pointed to 20 unpublished validation studies and six published ones.

There’s no particular reason to think these studies were flawed, although it isn’t ideal that many were performed by state police, who have a vested interest in finding that the program works.

The New York court concluded that TrueAllele followed generally accepted scientific methods and approved its use in court. But there is nevertheless a serious and important objection to such use.

The problem is that no one outside Cybergenetics knows the statistical parameters that the program uses. Without knowing those parameters, it’s almost impossible to criticize how the program functions and argue that, for example, it’s too lax in determining what counts as a match.

Put another way, seen from the outside, TrueAllele is a black box: A sample goes in, and results come out. But the company won’t reveal the black box’s inner workings.

In the context of academic scientific experiments, the black box would be a problem for accepting the validity of results. For one thing, reproducing experimental results requires actually doing the experiment again -- which means knowing how to run the experiment in the first place.

It isn’t really the same thing to run the same data through the black box and get the same result -- because you don’t know what happened inside.

What’s more, other scientists don’t simply want to reproduce experiments; they want to know the scientific mechanism that’s in play. A black box doesn’t tell you what the mechanism is, except very generally.

TrueAllele’s owners have a ready answer: They say their method has been tested, and that it works. According to their theory, it shouldn’t matter if defendants don’t know how it works. All that counts is that it does.

Who’s right? The New York court accepted the company’s answer, a pragmatic approach that determines admissibility based on practical efficacy. Other courts, and states, may follow suit by allowing TrueAllele on the grounds that it seems to work when tested.

But there’s something unsatisfying about this explanation when it comes to criminal justice. In other realms of life, we live and die by statistical probability. As Justice Oliver Wendell Holmes once put it, “all life is an experiment,” one in which we constantly play the odds.

Criminal justice is supposed to be held to a higher standard. We want proof beyond a reasonable doubt before convicting you, much less executing you. We should therefore be worried about the outlying case -- the statistically improbable result that nonetheless sends the bad-luck fall guy to prison.

In order to know that TrueAllele’s probability distributions don’t produce statistical outliers that we wouldn’t want to tolerate, we need to know what its inner process really is.

Democratic values reinforce this conclusion. The state, speaking as the people of Pennsylvania or New York, punishes crime. You shouldn’t be convicted based on a secret process that you are not permitted to know.

Trade secrets are important to many businesses. But when it comes to the life-or-death business of criminal justice, it’s reasonable to demand that they be exposed -- or that the secret techniques they use be inadmissible.

This column does not necessarily reflect the opinion of the editorial board or Bloomberg LP and its owners.

To contact the author of this story:
Noah Feldman at nfeldman7@bloomberg.net

To contact the editor responsible for this story:
Brooke Sample at bsample1@bloomberg.net