An Algorithm That Can Predict Human Behavior Better Than Humans (mit.edu)

Quartz describes an MIT study with the surprising conclusion that at least in some circumstances, an algorithm can not only sift numbers faster than humans (after all, that's what computers are best at), but also discern relevant factors within a complex data set more accurately and more quickly than can teams of humans. In a competition involving 905 human teams, a system called the Data Science Machine, designed by MIT master's student Max Kanter and his advisor, Kalyan Veeramachaneni, beat most of the humans for accuracy and speed in three tests of predictive power, including one about "whether a student would drop out during the next ten days, based on student interactions with resources on an online course." Teams might have looked at how late students turned in their problem sets, or whether they spent any time looking at lecture notes. But instead, MIT News reports, the two most important indicators turned out to be how far ahead of a deadline the student began working on their problem set, and how much time the student spent on the course website. ... The Data Science Machine performed well in this competition. It was also successful in two other competitions, one in which participants had to predict whether a crowd-funded project would be considered “exciting” and another if a customer would become a repeat buyer.

Comments

Overblown headline

[binarstu]

The headline of the Quartz article and the Slashdot summary, "An algorithm can predict human behavior better than humans", is, not surprisingly, hugely overblown.

What these researchers actually did was develop a system for automatically taking a massive data set with a huge number of variables, identifying the subset of variables or new combinations of variables that are most likely to be useful for predicting a particular response, and then formulating a predictive model. (This is an extremely simplified summary.) That is really cool, but to present it as some sort of general "algorithm for predicting human behavior" is silly. It's no more an algorithm for predicting human behavior than are automated statistical methods for building a predictive model from a massive dataset.