Facebook Wants to Create an AI System That Can Beat the ‘Toughest Game in the World’

Facebook is inviting AI researchers for the NeurIPS 2021 NetHack Challenge to beat NetHack – the apparent toughest game in the world.


Facebook has announced that it is looking to fund AI researchers to represent the company at the NeurIPS 2021 NetHack Challenge. The competition is geared to take on NetHack – a notoriously difficult game from the 1980s that not just has its own cult following, but is often regarded as one of the toughest games in the world. The project will seek to beat the game, but Facebook acknowledges that the more probably scenario would be for the AI system to register the highest possible score at the game. By doing this, the company hopes that it can establish NetHack as a viable artificial intelligence learning system – in the same way as AI research has so far focused on Go and Minecraft, among others.

NetHack will not be the first game where AI researchers are taking a targeted approach, in a bid to improve automation and intelligence in machines. Games and sports are arenas that often require not just set strategies, but skills such as improvisation, quick reactions, mid-game adaptations, responsiveness, intuition and more. Such skills have typically always been the forte of the human brain, which has seen the best participants come up with unique, innovative solutions in order to beat competitors. For AI, this is the eventual goal – to be able to learn, synthesise and process human level cognisance – something for which video games are just about perfect.

“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,” said the Facebook Engineering post from earlier today.

NetHack is known for its notorious levels of difficulty, ever since being built in the 1980s. The game actually does not expect participants to win – in fact, it is designed for everyone to die at some point, which is its eventuality. Every time a participant dies, the entire dungeon of a playground is reset and all users begin again. Its very nature makes it a great fit for Facebook, which hopes that its project will help it bring forth AI and machine learning solutions, based on different methodologies. NeurIPS 2021 begins this month itself, and continues until June 15. Winners will be announced in December 2021.

Thanks for reading till the end of this article. For more such informative and exclusive tech content, like our Facebook page