Facebook to Build an AI Assistant Using Minecraft

Facebook has recently announced that it is training an AI smart assistant that it has been working on using Minecraft. Yes, you heard that right, Minecraft. Minecraft is essentially an open world sandbox game which gives players complete freedom and choice over what they want to do in their world. Minecraft has grown to be one of the most popular games ever since its release in May 2009. The reason for Minecraft’s popularities lies heavily on the fact that it encourages players to explore, fight, build - rewarding them as they go along.

Essentially, Facebook’s plan in the long-run is to use Minecraft as a way for its artificially intelligent smart assistant to perform a variety of tasks from a given voice command. Researchers at Facebook and MIT released a paper in July stating how it plans to utilise the popular sandbox game to train its AI.

One of the researchers stated why he thinks that Minecraft is the most suitable environment to train Facebook’s AI assistant. He said that it was mainly due to Minecraft being a sandbox game. The idea of training an AI within a sandbox game like Minecraft means that the AI would be able to multitask, performing a variety of useful tasks rather than perform one task perfectly - like most artificially intelligent programs do in today’s day and age.

AI will be capable of doing a variety of tasks for users such as retrieving specific items or venturing with them.

AI will be capable of doing a variety of tasks for users such as retrieving specific items or venturing with them.

We wish to build an intelligent in-game assistant that can perform whatever players may want it to do. Simple examples of the assistant’s duties could range from building simple or complex structures to entire city scapes, breaking down structures, dancing, catching mobs, etc.
— Research paper published by Facebook and MIT

The way the researchers plan on making it work is by turning the AI assistant into a bot so that it can interact with human players and learn from them by performing multiple tasks. Researchers are hoping that from interacting with human players, the AI will be able to pick up on and understand how to perform these tasks after being prompted multiple times. An example of how this may work is if a player asks the bot to retrieve specific materials so that they can build a structure. Ideally, by continuously performing tasks like this and many others the AI should be able to pick up on what the player is asking for and their intention for asking the bot.

Players who are interested in using Facebook’s smart assistant can download it through GitHub using this link.

Although artificial intelligence has shaped our lives and made a lot of repetitive processes much faster and more efficient for us, it is about time we start looking into developing much more advanced AI programs capable of carrying out a variety of tasks.

Zacharia Sharifresearch