Technology that ChatGPT platform has the potential to do more than just talk. Linxi “Jim” Fanan AI researcher at a chip manufacturer Nvidiaworked with several colleagues to come up with a powerful language model GPT-4—the “brain” behind ChatGPT and more and more apps application and services—in the comfort of a square video game Minecraft.
The Nvidia team, which includes Anima Anandkumar, the company’s director of machine learning and a professor at Caltech, created a Minecraft bot called a traveler use GPT-4 to solve problems in the game. The language model creates goals that help the agent discover the game, and the code improves the bot’s skills in the game over time.
Travel does not play the game as a person, but it can read the state of the game directly through the API. For example, it can see a fishing rod in its inventory and a nearby river, and use GPT-4 to suggest a goal of doing some fishing activities to gain experience. It will then use this goal to ask GPT-4 to generate the code needed for the character to achieve that goal.
Courtesy of NVIDIA
The most novel part of the project is the code that GPT-4 generates to add behaviors to Voyager. If the original proposed code doesn’t run perfectly, Travel will attempt to refine it with an error message, feedback from the game, and a description of the GPT-4-generated code.
Over time, Travel builds a library of code to learn how to create increasingly complex things and explore more games. A chart created by researchers shows how capable it is compared to other Minecraft agents. Travelers get three times more items; explore more than twice as far; and build tools 15 times faster than other AI agents. Fan says this approach could be improved in the future with the addition of a way for the system to incorporate visual information from the game.
While chatbots like ChatGPT have wowed the world with their eloquence and clear knowledge—even if they often make up stories—Voyager shows great potential for language models to perform useful actions on computers. Using language models in this way could perhaps automate many common office tasks, potentially one of the biggest economic impacts of the technology.
Courtesy of NVIDIA
The process that Travel uses with GPT-4 to figure out how to do things in Minecraft can be adapted for a software assistant figuring out how to automate tasks through the operating system on a PC or mobile phone. phone. OpenAI, the startup that created ChatGPT, added “plugin” to bot allowing it to interact with online services like the grocery delivery app Instacart. Microsoft, the company that owns Minecraft, is also train AI programs to play itand recent company announced Windows 11 Copilot, an operating system feature that uses machine learning and APIs to automate certain tasks. It might be a good idea to test this kind of technology inside a game like Minecraft, where faulty code can do relatively little harm.
Of course, video games have long been the testing ground for AI algorithms. AlphaGomachine learning program master the exquisite board game Back to 2016, cut your teeth by playing the simple Atari video game. AlphaGo used a technique called reinforcement learningtrain an algorithm to play a game by giving positive and negative feedback, such as from scores in a game.
This method would be more difficult to guide an agent in an open-ended game like Minecraft, where there’s no score or set of goals, and where the player’s actions may not pay off until much later. Whether or not you believe we should be prepared to stop the existential threat from AI right now, Minecraft seems like a great playground for technology.
