Facebook: an army of virtual robots simulate trolls and fraudsters

A group of Facebook researchers has created a copy of the social network populated only by robots. These virtual users simulate different behaviors, such as trolls and fraudsters, in order to discover the bugs on the site.

Facebook must be constantly on the lookout for fraudsters and other malicious people who take advantage of the slightest flaw in its system. To help it better secure its code and discover unexpected effects of updates, the firm has created aparallel social network populated by robots called ”  Web-Enabled Simulation  ” (WES).

Until now, Facebook has used a purely virtual simulation of the site called Sapienz. With WES, the researchers relied on the actual infrastructure of the site made up of hundreds of millions of lines of code. They have thus formed a network of profiles managed by robots, some programmed to behave like simple users, others to act like crooks or trolls .

Artificial intelligence used to find site bugs

This research is based on machine learning and in particular artificial intelligence developed for games. Researchers are studying the interactions between these virtual users, who use the same processes as real users to like, share and add as friends. Their purpose is to detect malfunctions, for example if a robot suddenly has access to a lot more data after a site update.

Some of these robots operate in read-only mode, and could therefore access the real version of the social network without risk. For the others, the researchers emphasize that they take special care to isolate them from human users so as not to risk unexpected interaction.