Editing NetHack (section)

==NetHack Learning Environment==
The [[Facebook]] [[artificial intelligence]] (AI) research team, along with researchers at the [[University of Oxford]], [[New York University]], the [[Imperial College London]], and [[University College London]], developed an open-source platform called the NetHack Learning Environment, designed to teach AI agents to play ''NetHack''. The base environment is able to maneuver the agent and fight its way through dungeons, but the team seeks community help to build an AI on the complexities of ''NetHack''{{'s}} interconnected systems, using implicit knowledge that comes from player-made resources, thus giving a means for programmers to hook into the environment with additional resources.<ref>{{cite web | url = https://venturebeat.com/2020/06/25/facebook-releases-ai-development-tool-based-on-nethack/ | title = Facebook releases AI development tool based on NetHack | first = Kyle | last = Wiggers | date = June 25, 2020 | access-date = June 26, 2020 | work = [[Venture Beat]]}}</ref><ref>{{cite journal | title = The NetHack Learning Environment | first1 = Heinrich | last1 = Küttler | first2 = Nantas | last2 = Nardelli | first3= Alexander H. | last3 = Miller | first4= Roberta | last4= Raileanu | first5= Marco | last5=Selvatici | first6= Edward | last6= Grefenstette | first7= Tim | last7= Rocktäschel | arxiv = 2006.13760 | journal = [[Machine Learning (journal)|Machine Learning]] | date = 2020}}</ref> Facebook's research led the company to pose ''NetHack'' as a grand challenge in AI in June 2021,<ref>{{cite web |title=Launching the NetHack Challenge at NeurIPS 2021 |url=https://ai.facebook.com/blog/launching-the-nethack-challenge-at-neurips-2021/ |access-date=5 September 2022}}</ref> in part due to the game's permadeath and inability to experiment with the environment without creating a reaction. The competition at the 2021 [[Conference on Neural Information Processing Systems]] involved agents of various designs attempting to ascend. None of the agents managed this; the results were ranked by median in-game score, with the highest-ranked agent (Team AutoAscend) using a symbolic (non-machine-learning) design.<ref>{{cite web |title=The NetHack Challenge: Dungeons, Dragons, and Tourists |url=https://nethackchallenge.com/report.html |access-date=5 September 2022}}</ref>