Algorithms Need Managers, Too
Most managers’ jobs involve makingpredictions. When HR specialists decide whom to hire, they’re predicting who will be most effective. When marketers choose which distribution channels to use, they’re predicting where a product will sell best. When VCs determine whether to fund a start-up, they’re predicting whether it will succeed. To make these and myriad other business predictions, companies today are turning more and more to computer algorithms, which perform step-by-step analytical operations at incredible speed and scale.
Algorithms make predictions more accurate—but they also create risks of their own, especially if we do not understand them. High-profile examples abound. When Netflix ran a million-dollar competition to develop an algorithm that could identify which movies a given user would like, teams of data scientists joined forces and produced a winner. But it was one that applied to DVDs—and as Netflix’s viewers transitioned to streaming movies, their preferences shifted in ways that didn’t match the algorithm’s predictions.
Another example comes from social media. Today many sites deploy algorithms to decide which ads and links to show users. When these algorithms focus too narrowly on maximizing user click-throughs, sites become choked with low-quality “click-bait” articles. Click-through rates rise, but overall customer satisfaction may plummet.
Problems like these aren’t inevitable. In our work designing and implementing algorithms and identifying new data sources with a range of organizations, we have seen that the source of difficulty often isn’t bugs in the algorithms; it’s bugs in the way we interact with them. To avoid missteps, managers need to understand what algorithms do well—what questions they answer and what questions they do not.
As a growing body of evidence shows, humanizing algorithms makes us more comfortable with them. This can be useful if, for example, you’re designing an automated call function. A real person’s voice is more likely than an electronic voice to get people to listen. The fundamental problem, however, is that people treat algorithms and the machines that run them the same way they’d treat an employee, supervisor, or colleague. But algorithms behave very differently from humans, in two important ways:
In the latest Avengers movie, Tony Stark (also known as Iron Man) creates Ultron, an artificial-intelligence defense system tasked with protecting Earth. But Ultron interprets the task literally, concluding that the best way to save Earth is to destroy all humans. In many ways, Ultron behaves like a typical algorithm: It does exactly what it’s told—and ignores every other consideration. We get into trouble when we don’t manage algorithms carefully.
The social media sites that were suddenly swamped with click-bait fell into a similar trap. Their overall goal was clear: Provide content that would be most appealing and engaging to users. In communicating it to the algorithm, they came up with a set of instructions that seemed like a good proxy—find items that users will click on the most. And it’s not a bad proxy: People typically click on content because it interests them. But making selections solely on the basis of clicks quickly filled sites with superficial and offensive material that hurt their reputation. A human would understand that the sites’ designers meant “Maximize quality as measured by clicks,” not “Maximize clicks even at the expense of quality.” An algorithm, on the other hand, understands only what it is explicitly told.
In Shakespeare’s Julius Caesar, a soothsayer warns Caesar to “beware the ides of March.” The recommendation was perfectly clear: Caesar had better watch out. Yet at the same time it was completely incomprehensible. Watch out for what? Why? Caesar, frustrated with the mysterious message, dismissed the soothsayer, declaring, “He is a dreamer; let us leave him.” Indeed, the ides of March turned out to be a bad day for the ruler.
