Researchers from Facebook and MIT are collaborating on an artificial intelligence assistant for the online video game “Minecraft”. The goal is to create an AI system that can perform a wide variety of tasks, rather than being “superhuman” at a narrow set of activities. Basically, what’s known as artificial general intelligence.
MIT Technology Review reported the research team is creating an AI assistant that can interact with Minecraft players and perform a wide range of tasks upon request. The AI assistant will learn from these interactions, developing new skills that can be applied to different tasks.
The game Minecraft allows players to build using a variety of different blocks in a procedurally generated 3d world. This requires a degree of creativity from players. Other activities in the game include exploration, resource gathering, crafting, and combat. This “infinite variety” that adheres to “simple and predictable rules” is the reason researchers chose Minecraft.
Machine Learning has been used to train AIs to be inhumanly good at other video games. MarI/O, for example, used neural networks and genetic algorithms to show a level of gameplay in Super Mario World that isn’t humanly possible.
Super Mario World has far fewer choices for gameplay than Minecraft. This is important as the ultimate goal of the researchers is building an AI assistant that can help people with everyday tasks outside a gaming environment. Doing so presents a lot of challenges. Even a request like “build a wall 3 blocks tall and 15 blocks long” requires that the AI understand what building is in-game, plus be able to understand numbers and the measures for length and height. The AI assistant will learn these over time.
The researchers posted a whitepaper of their Minecraft AI assistant online and have also made it available to download on GitHub. They see a huge opportunity for the AI to learn with Minecraft and say the AI should eventually evolve to a point where it can ask human players to help it amass more knowledge.
That sound you hear is Stephen Hawking screaming from his grave in the hopes that Elon Musk will be able to stop this AI before it gets too powerful. That is clearly a joke, but Stephen Hawking and Elon Musk are among the loudest voices to have expressed fear of a self-learning artificial general intelligence.