Facebook wants to develop an AI (artificial intelligence) that is capable of beating the hardest game in the world. The social media giant wants to leave no stone unturned in a bid to bring down NetHack.
Different gaming genres including, RPGs, real-time strategy, and building games have served as a fertile training platform for the latest artificial intelligence systems in recent years. Digital competitors have effortlessly been defeating their human opponents.
They have outperformed them in titles such as StarCraft II and Dota 2 to Minecraft and Go.
Now, Facebook is urging the AI community to help the company to bring down NetHack. Aside from asking the community to help take down one of the most difficult titles in gaming history, Facebook is hoping it can help computers learn to imitate instances quicker without using a lot of resources.
Can you overcome the brutality of NetHack?
Even years after its development, NetHack is actively updated today. Every time a player perishes, the entire dungeon resets. Beating the game requires you to combine luck, outside-the-box problem solving, and research skills at NetHack Wiki.
The only way to best the game is to learn from the misadventures of those who came before you. It is equally important to learn to move your character, go up and down the stairs and fight monsters, according to NetHack Wiki.
Facebook working on an AI that can beat NetHack
Facebook is inviting researchers to participate in its NetHack 2021 NeurIPS Challenge. The NetHack Challenge is based on the NLE (NetHack Learning Environment), where teams will go toe-to-toe to build the best agents to play NetHack.
Researchers will focus on designing, training, and releasing AI systems that can develop reliable agents to beat the game or achieve as high a score as possible. Facebook announced this in its blog post on Wednesday.
More importantly, Facebook believes this will help showcase the NetHack Learning Environment as a workable reinforcement learning system. This will enable a wide range of potential AI/ML solutions based on neural, as well as symbolic methodologies.
Moreover, the candidate agents will play several games and each will feature a fantasy race. Facebook will also add some randomly drawn character roles.
Furthermore, the average number of episodes, where the agent finishes the game, will be calculated for evaluation episodes.
Aside from that, the median in-game end-of-episode score will also be considered. For ranking entries, an average number of wins will be computed. And if tied, the median score will be taken into account. The competition will run from June to October 15th and winners will be announced in December.