Facebook Plans to Design an AI System That Can Beat the ‘World’s Toughest Game’

June 09, 2021 | 2 minutes read


Facebook has reported that it plans to finance AI scientists to represent the organization at the NeurIPS 2021 NetHack Challenge.

The tournament is equipped to take on NetHack – a popular game created back in the 1980s that is not only one of the oldest games still played, but is also referred to as one of the most difficult games to win of all time. 

Though the aim of the plan is to win the competition, yet Facebook recognizes that the most likely situation would be for the AI framework to set the highest record at the game. By doing this, the organization believes that it can build up NetHack as a feasible artificial intelligence learning framework – similarly as AI research has so far zeroed in on Minecraft and Go, among others.

NetHack isn’t the only game for which AI specialists are adopting a targeted strategy with the aim to enhance intelligence and automation in machines. Sports and games not only require set strategies, but also practice, responsiveness, short reaction time, improvisation, instinct and similar skills.

The human mind has the ability to not only come up with innovative solutions but also brainstorm and create unique strategies to beat contenders. For AI, this is the targeted objective – to have the ability to learn, incorporate and measure human level cognizance – and video games are the best source to reach this goal.

According to a Facebook engineer, “The candidate agents will play a number of games, each with a randomly drawn character role and fantasy race. For a given set of evaluation episodes for an agent, the average number of episodes where the agent completes the game will be computed, along with the median in-game end-of-episode score. Entries will be ranked by average number of wins and, if tied, by median score.”

There is a lot of patience, luck, and knowledge required to win at this game. Since the game is deviously difficult, “luck” is thought of to be the biggest factor that can help you win the game.

The game actually does not expect participants to win – In fact, it has been designed in a way to kill the participants at one point or another, which they do eventually. Each time a player is killed, he has to restart the game all over again.

The design of the game itself makes it an excellent fit for Facebook, which aims that this project, with the use of different methodologies, will aid in improving machine learning and AI solutions. The winners of the game will be announced in December 2021.