Big Data

The big data Dead Hand may not be the end of the world

Neither humans nor machines are infallible, but combining the two makes for a far more robust big data system than relying totally on just one.

I've been reading the fascinating book, The Dead Hand by David Hoffman. The book recounts the height of Cold War tensions driving the nuclear arms race between the Soviet Union and United States in the 1980s, and in particular mentions a Soviet system called Dead Hand. The Soviets greatly feared a preemptive US attack that would disable Communist party leadership and internal communications, and effectively prevent the USSR from launching a counterattack. While laughably reminiscent of playground antics 30 years later, mutually assured destruction was a key tenet of nuclear war, and a strategy that gave one side even a slight advantage was perceived as incredibly threatening.

How it was supposed to work

Dead Hand was designed to automate a retaliatory strike if Soviet leadership were disabled, and would automatically launch a series of communications missiles, which would then launch all available nuclear weapons at the United States. While details of the system are still mysterious, apparently the Soviet military was unwilling to enable the system in a fully automated mode, preferring one last human check, from deep within a hardened bunker, before control of a retaliatory strike was handed over to automation.

One interview in the book suggests that no one knew if the soldiers trained to activate Dead Hand would actually perform their duties in a real nuclear war and, for some, this was a positive characteristic of the system. The author notes that some of the designers found it appealing that in the heat of nuclear Armageddon a lone soldier might decide to spare half the world, rather than launching a strike that would effectively nuke the other side.

While it's unlikely any Big Data initiatives we are dealing with have the capacity to end life as we know it, we're often enamored with a similar concept of completely automated processes. For some in marketing, the ultimate expression of Big Data is a department with a CMO, and perhaps an administrator to write checks to the various providers, completely automating marketing functions.

The human touch

Most of us have fallen victim to a mistakenly released or poorly targeted Big Data-driven marketing campaign. One of my favorites was an advertisement for "male enhancement" on a major global newspaper's paid online site. Whatever fraction of a penny that might have resulted from a click on the ad, certainly wasn't worth the subsequent brand damage. Similarly, highly automated stock trading systems have triggered a minor market crash and resulted in a suspension of trading. Even in the highly disciplined scientific fields, data-driven analyses predicted rampant global cooling a few decades ago, while current science points toward the opposite trend.

It's easy to cast this as "machines gone wild" fear mongering, but just as the Soviets added one final human element to their Dead Hand system, it's worth adding a human element to Big Data-driven applications as well. Big Data is based on complex, layered models, each with its own set of assumptions. Even the best models struggle to accommodate a rapidly changing macroeconomic environment or predict disruptive market changes. Neither humans nor machines are infallible, but combining the two makes for a far more robust system than relying totally on one, especially when business strategy, customer perception, and financial resources are at stake.


Patrick Gray works for a global Fortune 500 consulting and IT services company and is the author of Breakthrough IT: Supercharging Organizational Value through Technology as well as the companion e-book The Breakthrough CIO's Companion. He has spent ...

Mark W. Kaelin
Mark W. Kaelin moderator

How automated is you big data analytic system? Should it be more or less automated in your opinion? What is the correct balance?

Editor's Picks