Siegel: Peril, Promise and the Price of Predictive technology

As computers are entrusted to make judgment calls traditionally decided by people, should we worry? Predictions can cause difficulties not only when wrong, but also when they are right. Eric Siegel, founder of PAW: Predictive Analytics World, looks into implications.

WSJ MarketWatch, By Eric Siegel, Mar 22, 2013

There's a surprising twist. While some question whether the prescient machines that drive decisions by way of induction and prediction are trustworthy, an emerging problem is that they often predict too well. Predictive technology is so powerful, it reveals a future often considered private.

Millions of operational decisions in finance, marketing, law enforcement, and health care are now machine-driven - often with improved dexterity - using electronic predictions of human behavior, one person at a time. The technology to do this, predictive analytics, is a booming practice that's taken hold across many industries.

Computerized prediction will never be perfect - like people, prognostic technology often gets it wrong, although in many applications it turns out to be more accurate than people are. But predictive analytics can cause difficulties not only when its predictions are wrong, but when its predictions are right.

Future tension
For example, Hewlett-Packard flags those of its 330,000 employees who are most likely to quit their job - then, a small but growing number of managers review these predictions. When the computer's prediction is correct, might there be qualms about divulging an employee's private intentions to the boss?

As we learned last year, TargetTarget predicts which female customers are pregnant in order to identify sales opportunities with soon-to-be parents. When the technology is correct for a customer, her expectant status is now unwittingly in the hands of strangers, marketers who are not trained in the privacy protocols of the health-care industry. Should the marketing department of companies like Target be entrusted with such information?

[Note: About 75% of KDnuggets readers thought Target had a right to use this information ]


There is no easy solution. Privacy advocates - of which I am one - go too far when they sound alarms that imply predictive analytics ought to be sweepingly indicted. As with many technologies, it can enact both good and evil - like a knife. Outlawing it completely is not viable, and would be akin to forbidding deduction.

... But now is the time to define how we may predictively serve the masses without mistreating individuals. As we strike the balance between caution and cashing in, a new cultural acceptance of machine risk is bound to emerge.

Read more.