Algorithms need managers, too

• Autores: Michael Luca, Jon Kleinberg, Sendhil Mullainathan
• Localización: Harvard business review, ISSN 0017-8012, Vol. 94, Nº 1, 2016
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Dialnet Métricas: 1 Cita
• Resumen

  • Algorithms are powerful predictive tools, but they can run amok when not applied properly. Consider what often happens with social media sites. Today many use algorithms to decide which ads and links to show users. But when these algorithms focus too narrowly on maximizing click-throughs, sites quickly become choked with low-quality content. While clicks rise, customer satisfaction plummets. The glitches, say the authors, are not in the algorithms but in the way we interact with them. Managers need to recognize their two major limitations: First, they’re completely literal; algorithms do exactly what they’re told and disregard every other consideration. While a human would have understood that the sites’ designers wanted to maximize quality as measured by clicks, the algorithms maximized clicks at the expense of quality. Second, algorithms are black boxes. Though they can predict the future with great accuracy, they won’t say what will cause an event or why. They’ll tell you which magazine articles are likely to be shared on Twitter without explaining what motivates people to tweet about them, for instance. To avoid missteps, you need to be explicit about all your goals—hard and soft—when formulating your algorithms. You also must consider the long-term implications of the data the algorithms incorporate to make sure they’re not focusing nearsightedly on short-term outcomes. And choose the right data inputs, being sure to gather a wide breadth of information from a diversity of sources