Notes

Minecraft Diamond Challenge Leaves AI Creators Stumped

Minecraft diamond challenge leaves AI creators stumped By Sam Shead Technology reporter



12 December 2019



It takes a few minutes for new Minecraft players to work out how to find the diamonds that are crucial to the game, but training artificial intelligence to do this is more difficult than we expected.



Microsoft is a Minecraft publisher and other companies pushed developers to create AI agents that could find the sought-after treasures.



Most can crack it in their first session.



However, of the more than 660 entries, only one could have performed the challenge.



The results of the MineRL - which is a pronounced mineral-related - competition are scheduled to be announced formally on Saturday at the NeurIPS AI conference in Vancouver, Canada.
MINECRAFT SERVERS


The objective was to determine if the problem could be solved with a minimum of computing power.



Despite not having a winner, one organizer said she was "hugely amazed by" some of the participants.



"The task we asked for is extremely difficult," said Katja Hofmann, a principal researcher at Microsoft Research. "Finding diamonds in Minecraft takes many steps - from cutting trees, making tools, to exploring caves and actually finding a diamond.



"While the submitted agents aren't competent to tackle the issue completely but they have made significant progress and learned how to use many of the tools needed."



Mining diamonds



Minecraft has become wildly popular since it was released in 2011.



The open-world game has been downloaded more than 180 million times, and the title is played by more than 112 million players every month.



Diamond is a very important resource in Minecraft. It can be used to create powerful weapons and armour.



To acquire the precious gem However, the player must follow a number of other steps.



Jules Portelly, a Minecraft player, said it shouldn't take more then 20 minutes to get your first diamonds if you're comfortable with the game.



The players were allowed to use a single graphics processing unit (GPU) and four days of time to train. To put it in perspective, AI systems usually need months or even years of gaming time to master titles like StarCraft II.



A relatively small Minecraft dataset, with 60 million frames of human player data, was made available to entrants to train their systems.



"At the start of every episode they spawned in a procedurally-generated Minecraft world," explained Dr Hofmann.



"So they had to understand the basics of finding resources, making tools and finding a treasure."



Minecraft to run artificial intelligence tests



DeepMind AI achieves Grandmaster status at Starcraft 2



Minecraft players to be helped by an AI assistant



The goal was to have coders create programs that were learned by imitation, a method known as "imitation learning".



This is where you try to make AI agents adopt the best approach to solve a problem by getting them to imitate humans or other software.



This contrasts with "reinforcement learning" in which agents are trained to find the most effective solution through trial and error.



AI for all



Research has demonstrated that reinforcement learning may result in better results.



For instance DeepMind's AlphaGo Zero program trumped one of the hub's earlier initiatives, which employed reinforcement learning and the study of labelled data from human play to help players learn the game of Go.



However, this "pure" approach typically requires more computing power, making it too expensive for researchers outside of large organisations or governments.



William Guss, the main organizer of the competition and a PhD student at Carnegie Mellon University, told the BBC that the purpose of the competition had been to demonstrate that "throwing massive computing at problems isn't necessarily the best way to push the state of the art in a field".



He added: "It works directly against increasing access to these systems for reinforcement learning and leaves the capacity to train agents in complicated environments to companies with vast amounts of computing."



However, the outcome may be a way to highlight the advantages these well-funded companies have.



Minecraft graphics to look more real



19 August 2019



Minecraft player is sacked for five years



29 April 2019


Homepage: https://bonfire.im/