Minecraft could be the key to creating adaptable AI

Minecraft is a game for humans, but it could also help AI


Minecraft is not only the best-selling video game in history, it could also play a key role in creating adaptable artificial intelligence models capable of performing a variety of tasks like humans do.

Steven James at the University of the Witwatersrand in South Africa and colleagues developed a gold standard test as part Minecraft to measure the general intelligence of AI models. MinePlanner evaluates an AI's ability to ignore unimportant details while solving a complex, multi-step problem.

Many AI training courses “cheat” by giving a model all the data it needs to learn how to do a job and nothing extra, James says. It's a fruitful approach if you want to create software to accomplish a specific task — like predicting the weather or folding proteins — but not if you're trying to create artificial general intelligence, or AGI.

James says future AI models will need to tackle complex problems, and he hopes MinePlanner will guide that research. AI working to solve a problem in the game will see scenery, extraneous objects, and other details that are not necessarily necessary to solve a problem and should be ignored. He will have to study his environment and determine for himself what is necessary or not.

MinePlanner consists of 15 construction problems, each with an easy, medium and hard level, for a total of 45 tasks. To complete each task, the AI ​​may need to take intermediate steps – building a staircase to place blocks at a certain height, for example. This requires the AI ​​to be able to step away from the problem and plan ahead in order to achieve a goal.

In experiments with the state-of-the-art planning AI models ENHSP and Fast Downward, open source programs designed to manage sequential operations in pursuit of an overall goal, neither model was able to solve none of the difficult problems. Fast Downward was only able to solve one of the medium problems and five of the easy problems, while ENHSP performed slightly better by solving all but one of the easy problems and all but two of the medium problems.

“We can't require a human designer to come in and tell the AI ​​exactly what it should and shouldn't care about for every task it might have to solve,” James says. “That’s the problem we’re trying to solve.”

The subjects:

Source link

Scroll to Top