The Potential of Minecraft for Developing Adaptive AI

Mine Craft Not only is it the best-selling video game of all time, it could be the key to creating adaptive artificial intelligence models that can handle a variety of tasks just like humans.

stephen james and colleagues at the University of the Witwatersrand in South Africa developed a benchmark test in which. Mine Craft Measure the general intelligence of your AI model. MinePlanner evaluates AI's ability to ignore unimportant details when solving complex problems in multiple steps.

According to James, much AI training is “cheating” by giving the model all the data it needs to learn how to do a job, and nothing irrelevant. While this is a useful approach if you're writing software to perform a specific task, such as predicting the weather or folding proteins, it's not useful if you're trying to create artificial general intelligence (AGI).

James says that future AI models will need to tackle wicked problems, and he hopes MinePlanner will guide that research. The AI ​​working to solve in-game problems recognizes scenery, extraneous objects, and other details that are not necessarily needed to solve the problem and should be ignored. You need to investigate your surroundings and decide for yourself what is necessary and what is not.

MinePlanner consists of 15 construction problems, each with easy, medium, and hard settings, for a total of 45 tasks. The AI ​​may need to perform intermediate steps to complete each task. For example, building a series of stairs to place blocks at a certain height. This requires AI to narrow down the problem and plan ahead to achieve the goal.

Experiments with state-of-the-art planning AI models ENHSP and Fast Downward, open-source programs designed to process sequential operations in pursuit of an overall goal, show that both models successfully complete difficult problems. I couldn't do it. Fast Downward was only able to complete one medium problem and five easy problems, while ENHSP completed all but one easy problem and all but two medium problems. By completing all of the above tasks, they achieved slightly better results.

“You can't step in and tell a human designer exactly what to care about and what not to care about for every task that an AI needs to solve,” James said. say. “That's the problem we're trying to address.”