Teaching Machines How to Spell Will Help Catch Terrorists – Defense One
The Boston Marathon bomber Tamerlan Tsarnaev could have been caught before his attack if only his name had been properly matched to a terrorist watch list.
In a Defense One byline, Basis Technology VP of Engineering David Murgatroyd outlines how more robust software can match names with spelling variations as well as tackle other information processing problems faced by security agencies.
While the system that failed to identify Tsarnaev is classified, we can learn from public examples about how they work. Murgatroyd takes the example of IBM’s Watson on Jeopardy. Despite an overall impressive performance, Watson answered “Toronto” to a question in the category “U.S. Cities.” Using 200 million pages of content via 90 powerful servers, Watson stumbled not on compiling a list of potential answers, but on removing one answer that most humans would instantly know to be false.
This might seem like a minor example, but as Murgatroyd writes “when we move beyond game shows to stopping terrorists, curious errors can have fatal consequences.”