There's a tendency among tech elitists to view IT as the solution to an industry's or a company's complex challenges. The secret to complying with SOX and myriad other regulations, we're told, can be found in a nifty software tool. The future of supply chain management rests on the tiny shoulders of an RFID chip. If only health care, heavy manufacturing, and other agnostics got the high-tech religion, costs would fall, quality would improve, and those troubled industries would become much more stable.
Yes, yes, and yes ... but only to a degree. It's ludicrous to suggest that any vertical industry or horizontal process can be transformed with technology alone. If you're FedEx, giving customers real-time information on their packages in transit is indeed vitally important, but you'd better be at least as expert at actually delivering the goods. If you're Dell, supply chain automation and efficiency are key to your success, but you'd better make a dependable computer packed with the latest features.
Technology initiatives are often an enabler and a driver of monumental changes, but they're rarely the sole lever. And sometimes they're a distraction, a waste of precious time and money.
In trying to use predictive data mining to root out terrorists, U.S. government agencies are doing just that--throwing a lot of money and manpower at technical measures that can't possibly achieve the desired outcome. Or so argue the Cato Institute's Jim Harper and IBM's Jeff Jonas in a new paper.
Predictive analytics involves running historical data through mathematical algorithms to identify trends and patterns and predict future outcomes. As we reported in June, 14 of the 199 data mining projects revealed by the Government Accountability Office in 2004 involved analyzing intelligence to detect potential terrorist activities. Eight of those programs involved private-sector data, including personal information provided by aggregators. At the time of our story, the National Security Agency was said to be compiling phone call records of tens of millions of Americans to populate a huge database on which the NSA would perform pattern-based analysis to hunt for terrorists. No one knows how much the feds were or are spending on all of their data mining efforts, but suffice it to say that Cognos, IBM, Teradata, and other vendors and consultants are making more than a few bucks.
Harper and Jonas (who's a hard-core data miner as chief scientist with IBM's Entity Analytic Solutions Group) maintain that data mining is useful in identifying consumer shopping habits and financial fraud. If you're trying to raise response rates for an ad campaign, for instance, data mining might get you up to 5%--still highly inaccurate, but good enough. However, data mining used to identify a few potential terrorists among hundreds of millions of people needs to be far more accurate--and it's not, the authors say. For one thing, "terrorism does not occur with enough frequency to enable the creation of valid predictive models," they say. Such analyses register false positives more than 90% of the time, the authors maintain, a fact that renders predictive analytics for antiterrorism "useless and potentially harmful."
Data mining has proved successful in law enforcement, but on a far more limited scale. In Richmond, Va., police analyze data to determine the probability that a particular crime will occur in a specific area at a given time, influencing where and when commanders deploy officers. "There is no harm in deploying a task force to a certain location based on number crunching," Harper says in an e-mail exchange. "There is harm in making specific individuals into suspects based on their fitting a 'pattern' of terrorism planning."
What's your take? Should the feds be spending our antiterrorism dollars elsewhere? Which other tech initiatives do you consider misguided or over the top? Drop me a note at the e-mail address below.
VP/Editor In Chief
To find out more about Rob Preston, please visit his page.