A New York Times article this week about the struggles within Facebook to get a handle on disinformation after the election illuminates a fascinating idea: surveying users to generate training data for artificial intelligence algorithms. As the Times reported:
The company had surveyed users about whether certain posts they had seen were “good for the world” or “bad for the world.” They found that high-reach posts—posts seen by many users—were more likely to be considered “bad for the world,” a finding that some employees said alarmed them.
So the team trained a machine-learning algorithm to predict posts that users would consider “bad for the world” and demote them in news feeds. In early tests, the new algorithm successfully reduced the visibility of objectionable content. But it also lowered the number of times users opened Facebook, an internal metric known as “sessions” that executives monitor closely.
This metric was known internally as P(Bad for the World)—in statistics parlance, describing the probability, from 0% to 100%, that a user would rate a certain post as bad for the world.