The bots are called the Web-Enabled Simulation (WES), uses Artificial Intelligence (AI) to replicate the behavior of human (Facebook users) such as posts like and sending friend requests.
According to Mark Harman, Research Scientist at Facebook AI, human behavior changes and evolves over time and it will be different in different geographies, which makes the job more difficult to predict the ways in which an individual behaves even for a small change. Therefore, WES has overcome the problem.
WES is a very new method for constructing realistic simulations of social networks, Harman wrote in a blog post.
The ML solution will train the bots to simulate the behavior on social media platforms thereby improving the software testing in complicated environments in product areas such as security, safety, and privacy.
Bots are trained in such a way to interact among themselves so that they send and receive messages, comment on the bots’ posts, publish their own posts or send friend requests to the other bots like the real users. Bots cannot interact with real users and the bots’ behavior will impact real users or their (real users) experience. WES has the capability to enable the interactions between thousands or millions of bots.
Harman also added that the company is using both offline and online simulation, by training bots through Machine Learning (ML) to be efficient and effective.
These train bots are deployed on the production codebase by WES. Bots can interact with one another but are isolated from real users.
Facebook said that the company is developing an ability to answer what-if and counterfactual questions using WES with realism.
Facebook has developed a simulated Facebook environment called WW using WES using the same production codebase. The simulation is called WW because it is a smaller version or recreation of the World Wide Web.
As said earlier bots cannot interact with Facebook users, but Facebook is trying to ensure that the simulations are authentic and matches the behavior from human users.