Facebook’s new AI-powered Bots to detect harmful behavior among users

Its social media era now, across all generations and regions the use of social media has increased drastically. Although most people use social media to connect with their dear ones, various users have exhibited harmful behaviors like spreading spam, scamming others, bullying, and selling weapons, drugs, etc. The crimes using social media is increasing day by day. These behaviors can cause severe emotional fatigue to the effected victim. Facebook has recently come up with an innovation to identify and prevent this bad behavior from an early stage.

Facebook engineers have come up with new AI-powered bots that can simulate the actions of bad actors by letting them loose in a parallel version of Facebook. This simulator is based on Facebook’s real code base and is named as ‘WW’ and pronounced ‘Dub Dub’. These researchers can study the first study the behavior of these bots and later find ways to stop them. The way these bot works is quite interesting, in real life most scammers work by first forming a friendship group with the people they are targeting. Taking into consideration this trend Facebook engineers modeled the behavior of their bots in such a way that there two sets of bots ‘innocent bots’ and ‘bad bots’. The engineers will further try to stop the behavior of these bots by introducing various constraints like limiting the number of messages sent, messages posted, etc. This will help them learn how the behavior of the bots are affected and further help to redefine them.

The advantage of this simulation is that they working on the real version of Facebook and thereby the results will be realistic. These bots can also uncover potential weaknesses in Facebook’s architecture. While a limitation of this bot is that they cannot simulate complex behaviors.

Harman on speaking about this said that the main focus now is to train the bots about the behaviors that are seen on these platforms. This type of innovation is very important at this point and will help us to find a solution to these bad behaviors rather than catching up.

(This article was originally published in Passionate in Analytics.)

Industry-Academia Connect Platform