Minecraft Will Help Build Artificial Intelligence with Human Capacity

Minecraft, open world game that allows various modifications, always yielded many projects outside the box, as this interactive computer (with calculator!). Since Microsoft bought the title for $ 2.5 billion in 2014, the company’s engineers realized they could use it to improve the artificial intelligence.
On Monday (14), AIX Project was born with this premise. Created by Katja Hofmann, a researcher at Microsoft Research Cambridge in the United Kingdom, the experiment makes use of all this complex and versatile world of Minecraft to test and establish research in order to improve the artificial intelligence.
This project is only possible thanks to the collaboration of users; currently, it is in closed beta for a small group of academic researchers. So far, the experiments are made on the local computers of participants, and are likely to observe ordinary users movements in the background to learn over time how they are made.
Still, it’s a different approach to artificial intelligence used in AlphaGo, Google computer that defeated a world champion in a Chinese game, for example. As pointed out by Engadget, the game Go is made of specific tasks, the computer is trained millions of times to play. Microsoft wants to create a kind of “general” intelligence, closer to human.
With Minecraft, they want to train the computer to learn how to do things like climb to the highest point on the map using the same type of resource that a human would when he learns a new task. The system think for himself, which is a challenge for deep learning systems.
“Unlike other computer games, Minecraft provides its users with endless possibilities, from simple tasks like walking around and look for a treasure to complex, how to build a structure with a group of friends,” says Microsoft. Katja argues that the game is ideal for this type of project, because, as the example shown at the beginning of the post, you can build your own games within the game.
Microsoft explains that the artificial intelligence agent start knowing anything about the environment, then understand what is around you and know what is important or not – following the example above, climb a hill is important, but killing a cow not .
“He needs to go through a lot of trial and error, including regularly fall into rivers and lava. And you need to understand – through gradual rewards – when he reached all or part of your goal, “the post on the company blog.
By September, Microsoft plans to launch the AIX project in an open-source platform for the public. At the BBC, Katja explains that AIX will lead to artificial intelligence beyond its capacity, on a human level of knowledge. “But eventually, we can take the project a step further and include tasks that allow IA agents learn to collaborate with humans and help them in a creative way,” he adds.

Intelligence “general”

Why all this was created? Microsoft argues that, despite the efforts we have ever had with any speech recognition and image, beyond the visual interpretation of the environment from algorithms, we can go further.
They want to create a certain general intelligence, which approaches the human thought and the “complex way” we learn and make decisions. It may seem easy for us, but it’s really difficult for computers.
“A computer algorithm can successfully complete a task equal to or better than an ordinary human, but it can not compete as a child is interacting in a series of stimuli – light, smell, touch, sound, discomfort – and learn that if you cry, chances are good that your mother will feed you, “exemplifies the post on the Microsoft blog announcing the project.
Practice these learned information is also easier using the Minecraft. As we have seen in this post, the difficulties that a robot has to move like humans are tremendous. Imagine, then, if Microsoft had to build a robot to do things like Minecraft, like climbing a mountain, and so learn with time. It costs, and the time it would have, are impractical.
And Minecraft helps make this more accessible paradigm. In the game, when seen in more technical scenarios, you can make complex decisions that generate consequences (sounds like real life?), And you can add elements that make the game more difficult as the player gets better.