Minecraft Diamond Challenge Leaves AI Creators Stumped

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

12 December 2019

It takes just a few minutes for new Minecraft players to figure out how to dig up the diamonds that are essential to the game, but educating artificial intelligence to do this has proved harder than expected.

Microsoft, an Minecraft publisher and other companies pushed coders to develop AI agents that could locate the sought-after treasures.

The majority of people can complete it in their first attempt.

However, of the more than 660 entries, not one of them was up to the challenge.

The official announcement of the results of the MineRL competition that is pronounced "mineral", will be announced at the NeurIPSAI conference in Vancouver, Canada on Saturday.

The aim had been to see whether the problem could be solved without massive amounts of computing power.
My city

Despite the lack of any winners, one of the organisers stated that she was "hugely impressed" by the participants.

"The task we posed is very hard," said Katja Hofmann who is a principal researcher at Microsoft Research. "Finding a Minecraft diamond requires many steps, from cutting trees to making tools , to searching caves to finally finding a diamond.

"While no agent submitted has solved the issue completely but they have made a lot of progress and made many of the tools needed to accomplish the task."

Mining diamonds

Since its release in 2011, Minecraft has been wildly popular.

The open-world game has been downloaded more than 180 million times and the game is played by more than 112,000,000 players every month.

Diamond is one of the most important resources found in Minecraft because it can be used to build strong armour and powerful weapons.

To get the precious stone However, the player must go through a few other steps.

"If you're already familiar with the game it shouldn't take you more than 20 minutes to collect your first diamonds," Minecraft player Jules Portelly told the BBC.

The entry requirements for entry were that participants could only make use of one graphics processing unit (GPU), and four days of training. To put it in perspective, AI systems usually need months or years of game play to master games such as StarCraft II.

A comparatively small Minecraft dataset, comprising 60 million frames of human player data, was also made available to participants to train their systems.

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

"So they really had to be able to comprehend the concept of finding resources, making tools and finding diamonds."

Minecraft for artificial intelligence experiments

DeepMind AI is awarded Grandmaster status at Starcraft 2

Minecraft players will be assisted by AI assistant

The organizers wanted the programmers to develop programs that could learn by imitation, using a technique called "imitation learning".

This is where you attempt to get AI agents to take the best method of solving a problem by getting them to imitate humans or other software.

It is in contrast to relying entirely on "reinforcement learning", in which an agent is effectively taught to find the best solution via a process of trial and error, but without using the knowledge of previous experiences.

AI for all

Researchers have discovered that reinforcement learning on its own can sometimes deliver superior results.

DeepMind's AlphaGo Zero program, for instance, beat an earlier effort by the research hub that used reinforcement learning and the analysis of data labelled from human behavior to teach the boardgame Go.

However, this "pure" method typically requires more computing power, which makes it prohibitively expensive for researchers outside of large corporations or governments.

William Guss, a Carnegie Mellon University PhD student, was the principal organizer of the competition. He explained that the purpose of the contest was to prove that "throwing massive computation at problems isn’t necessarily the best method to advance the state of the art as a discipline."

He added: "It works directly against increasing access to these reinforcement learning systems and leaves the capability to train agents in complex environments to corporations with swathes of computing."

However, the final outcome could highlight the advantages these organizations with good funding have.

Minecraft graphics that look more real

19 August 2019

Minecraft player loses five years of game

29 April 2019