Looking for the Lie

This polygraph debate reminds me of one closer to my area of academic expertise: computer vision. One of the current goals in computer vision is to develop a system where a computer recognizes, identifies, and tracks the individuals captured by a security camera. Ideally, the system should be able to identify every man, woman, and child in the United States – or maybe even in the world. These systems have actually gotten quite impressive. The two main mechanisms by which they work are facial and gait analysis. They are able to identify and track groups of people (say 10 known researchers) with a good degree of accuracy (say 95%). However, what happens in an unusual situation, like when one of the researchers shaves his moustache, ages 10 years, and is now in a wheelchair?

The human brain, the most amazing computer ever constructed, would still be able to recognize the researcher. But the video tracking system would be stumped. Therefore, although the video tracking system might be a useful tool in some situations, it would be foolish to trust it when national security is at stake. This is what we are doing with polygraph machines—machines with less underlying scientific support than the video tracking systems.

After the industrial revolution, our society couldn’t trust a result if it didn’t come from a machine. In the information age, we can’t trust anything that isn’t processed by a computer. Machines with a high degree of accuracy, especially those that have been around for awhile, are the most insidious. Humans don’t trust humans, and it is easy to put our blind faith in these machines. However, computers and machines are only as good as their weakest designers and operators. They results may depend on the environment or situation. Algorithms can introduce artifacts. As another poster mentioned, people have been wrongfully convicted on the basis of polygraph results.

The construction and use of lie-detection machines has important implications politically (the War on Terror), legally (Gitmo, false imprisonment), socially (how to tell if someone is lying), and in “adult” pop culture (Big Brother, Minority Report). These are all important issues to the reader of the NYT magazine. The reader also wants to be informed about the current state of scientific progress, especially on hot topics, like fMRI. They may want a few more scientific details than the “Esquire” readers, but not too many. They also treat issues more contemplatively than the Esquire readers.

I doubt the NYT Magazine readers are too concerned about mind-detectors being unleashed on the public, stealing our free will. Not yet, anyway (cross your fingers). I think this is just a bit of media sensationalism used to spice up the article. The author truly hopes to show that the polygraph has no scientific basis, and it is in use more out of habit than anything else. However, making a “real” lie-detecting machine is a complex, unsolved problem. We know almost nothing about the brain, and building things without this basic research is like “trying to get to the moon by climbing a tree.” Maybe in 500 years, people will laugh at our clumsy polygraph machines and primitive video-trackers as abominations of science, and the fact that they were actually used for security purposes as abominations of justice.