Thursday, September 22, 2011

Could you really predict crime with a computer program?

From Io9: Could you really predict crime with a computer program?
In new JJ Abrams show Person of Interest, airing tomorrow night on CBS, a mysterious man named Finch (Lost's Michael Emerson) has become disenchanted with his government job designing a system that sifts through massive amounts of data to find terrorism suspects. Now Finch is a rogue outsider with a backdoor to the system he created — nicknamed "the Machine." He's able to pilfer hints about upcoming (non-terrorist) crimes from his Big Data creation when it feeds him the social security numbers of people who are soon to be at the center of a crime. Along with his partner Reese (Jim Caviezel), Finch tries to stop these crimes before they happen.

The Machine is based on real Big Data technologies, software that can sift through huge volumes of information from security cameras, social networks, credit records, or anything else to find patterns. But could a software program, even a really sophisticated one, actually spit out the social security number of a person at the heart of a future crime?

We asked Arnab Gupta, CEO of Opera Solutions, a Big Data company that designs software for the government and industry that in some ways resembles what Finch creates in Person of Interest. Gupta said the idea for his company came from reading about "trying to find patterns in human behavior" in Isaac Asimov's Foundation series. And in some ways, Asimov's dream in those novels has come true: datamining tools can already predict some kinds of future events. But they don't do it the way Finch's Machine does.

Gupta's company helps intelligence agencies discover what he calls "a signal," or pattern, in publicly-available data from social networks and other sources. That data is then combined with the government's classified data, from surveillance or communication records. Then his company's software looks for what he calls "anomalous patterns." Gupta elaborated:

How do you know if a threat is emerging if you don't know what the threat is? We look for pattern anomalies, after we've established what typical patterns are from a dataset. We can say if something is off kilter.

For example, software like Gupta's might be able to predict social unrest before a series of events like Arab Spring. But, Gupta cautions:

The machine alone cannot predict what's going to happen. Only a human can draw that conclusion. You have to have human insight to understand that signal we're getting from the data.

So could Finch's machine exist in real life? Gupta says no. Certainly there are programs that can predict crime, but not at the granular level we'll see on Person of Interest. You'd never have a program that could offer a specific person's social security number as a clue to a future crime. And, says Gupta, you can't forget the human factor. "This kind of data analysis will become more and more powerful, as we integrate our data sources," he says. "But in the end you'll always need a human analyzing the patterns you find. A machine couldn't do it on its own."

In other words, Person of Interest is doing what science fiction has always done. Show creators Jonathan Nolan (Memento, Dark Knight Rises) and JJ Abrams are taking an existing technology and extrapolating what it might do sometime in the future, or in a parallel present

No comments:

Post a Comment