Facebook announced on June 9 that it wants to challenge NetHack and is looking for an AI to assist them. The social media giant is launching a competition at the NeurlPS 2021 AI conference in Sydney, Australia.
Facebook stated that NetHack, an '80s video game with simple visuals considered among the most difficult games in the world, can enable data scientists to benchmark state-of-the-art AI methods in a complex environment without the need to run experiments on a computer.
Facebook Wants to Beat NetHack
Games have served as a benchmark for artificial intelligence for years, but things started to escalate in 2013, the year that Google's DeepMind showed a system that could play "Breakout," "Pong," "Seaquest," "Space Invaders," "Enduro," "Beamrider," and "Q*bert" at superhuman levels.
The advancements are not just to improve the game design, according to DeepMind cofounder Demis Hassabis.
Rather, they are informing the development of systems that might diagnose illnesses, predict complicated protein structures, and segment CT scans one day, according to Engadget.
Reinforcement learning, a type of AI that can learn strategies to orchestrate massive systems like manufacturing plants, financial portfolios, traffic control systems, and robots, is transitioning from research labs to impactful, real-world applications.
Self-driving car companies like Waymo are using reinforcement learning to develop control systems for their vehicles. Siemens is employing reinforcement learning to calibrate its CNC machines.
Facebook AI researchers Edward Grefenstette, Tim Rocktaschet, and Eric Hambro wrote in a blog post said that the recent advances in reinforcement learning had been fueled by simulation environments such as games like "Dota 2", "StarCraft II," or "Minecraft."
However, this progress requires running thousands of GPUs in parallel for a single experiment while also falling short of leading to methods that can be transferred to more real-world issues outside of the games.
Facebook wants to create environments that are complex, highlighting shortcomings of RL, while also allowing fast simulation at low computation costs, according to Venture Beat.
NetHack Learning Environment
Facebook's plan follows the release of NetHack Learning Environment or NHLE, a research tool that is based on the original NetHack, according to Android Headlines.
NetHack was first released in 1987. It tasks players with descending more than 50 dungeon levels to retrieve a magical amulet. They must use weapons, wands, armors, potions, spellbooks, and other items and fight monsters.
The levels in NetHack are generated, and every game is different, which the Facebook researchers note tests the generalization limits of leading AI.
Grefenstette, Rocktaschel, and Hambro added that winning a game of NetHack requires long-term planning in an incredibly unforgiving environment. As soon as a player's character dies, the game starts from scratch in a new dungeon.
Completing the game as an expert player takes on 50 times more steps than an average "StarCraft II" game, and players' interactions with objects and the environment are extremely complex, so success usually hinges on calling upon the imagination to solve issues in creative ways and consulting external knowledge sources like the official NetHack Guidebook, the NetHack Wiki, and online videos and forum discussions.
This article is owned by Tech Times
Written by Sophie Webster