Covering Disruptive Technology Powering Business in The Digital Age

This article was originally published by blog.wsj.com and can be viewed in full here

As the tech world celebrated next-gen products and services at the Consumer Electronics Show, a federal agency issued a stern warning to companies trying to take advantage of big data: Beware your biases.

In a report released Wednesday on the use of big data by businesses, the Federal Trade Commission asked companies to consider whether their data sets represented the user population, whether their algorithms were insulated against social bias, and whether the predictions they generated were accurate.

Bias can be difficult to weed out of big data programs, as a string of public incidents has revealed. For example, a company that advertises services to customers who apply through social media may be neglecting less tech-savvy populations, the report said. Alternatively,  software that tries to identify promising job applications by factoring in graduation from atop schools may inherit the biases of college admission decisions.

The FTC urged companies to remember a golden rule of logic: Correlation does not equal causation. “Companies should remember that while big data is very good at detecting correlations, it does not explain which correlations are meaningful,” the report said.

Exhibit A: Google Flu Trends, the Alphabet Inc.GOOGL -2.41% division’s technology for predicting flu outbreaks. Google’s algorithm attempted to use search queries on flu-related topics to detect incidence of flu ahead of government estimates. It proved accurate initially but overshot prevalence of the illness in subsequent years, leading to an ongoing debate about whether data mining techniques and new sources of data should be used to make big predictions about public health.

The FTC report goes on to point out that big data has had positive results. People judged a poor credit risk by conventional means have received loans through lenders such as ZestFinance or Affirm that judge creditworthiness based on big-data techniques. Some educational institutions use big data to identify students who are at risk of dropping out, the report notes, and healthcare organizations mine data to predict a patient’s likelihood of hospital readmission or predisposition to disease.

However, the report noted rising concerns that big data can deny opportunities to some people, such as those judged to be a poor credit risk based on factors tangential to their finances or repayment history.

Other critics worry that big data can risk exposing sensitive information. For instance, researchers found that combining public Facebook FB -4.90% likes with basic survey information enabled them to predict Facebook users’ ethnic origin 95% of the time, male Facebook users’ sexual orientation 88% of the time, whether users were Christian or Muslim 82% of the time, their political orientation, and whether they had used alcohol or drugs.

(0)(0)