Prism doesn't scare me.
On 9/11, my office was on the 39th floor of One World Trade. I was one of the many nameless people you saw on the news running from the towers as they collapsed.
But the experience didn't turn me into a hawk. In fact, I despise the talking heads who frame Prism as the price we pay for safety. And not just because they're fear-mongering demagogues.
I hate them because I'm a technologist and they're giving technology a bad name.
Let's start with the basics.
[ Meet IBM's "Engagement Advisor," a computer that can take customer complaints. Read Watson Gets Call Center Job. ]
What is Prism? If you're the vendor that sold it to the National Security Agency, Prism is a proprietary black box that applies state-of-the-art predictive analytics to big data to infer relationships between known terrorists and their social networks. That's marketing jargon, so let's break it down.
Note that the only thing proprietary in that last paragraph is the vendor's hokey sales pitch. Everything mentioned there can be built with open-source tools, specifically a scalable distributed graph such as Neo4j and some natural language processing (NLP) libraries from Stanford University. So if you're in government IT or purchasing, don't buy the vendor BS.
First, the graph ...
In theory, every person in the world can be a node on a graph. And every communication between two people is just a relationship between those two unique nodes. So if you were able to compel Verizon and every carrier in the world to give you their complete call records, you could create the world's largest game of Six Degrees of Kevin Bacon.
Supplement those phone records (as the thing that connects two people) with emails, instant messages, known aliases and financial transactions, and your ability to infer relationships dramatically improves.
That, by the way, is the same kind of inference engine that companies such as Amazon use to figure out which products to suggest you buy. It's a more sophisticated way of asking if you want fries with that. Only in this case, instead of advancing commercialism, law enforcement gets to quickly determine the social networks of known terrorists.
This isn't some dystopian Minority Reports-like future. This is good old-fashioned policing supplemented by technology. Instead of manually sifting through phone records and drawing lines on a whiteboard between grainy pictures of suspects (a la every serial killer movie you've ever seen), the NSA is using a graphing engine.
And for the best reason possible: to speed up the narrowing of the search.
Next, the NLP ...
So now you know who's communicating with whom. How can you make sense of content: the billions of hours of real-time voice and email exchanges between people? You certainly don't want to hire tens of millions of analysts to listen, translate and raise their hands whenever someone that's two degrees away from some blind sheikh uses the word jihad.