The tech industry has proven time and time again that video games don't have to be for humans alone. Researchers have been training AIs to play games of varying complexities for years now.
Indeed, some machine learning models can defeat some of the world's top Starcraft and DOTA 2 players. However, that's not enough for AI research team OpenAI. The organization is now turning its attention toward the world of MMOs.
Don't worry, though, OpenAI isn't going let loose hundreds of AI agents in games like World of Warcraft. Instead, it's developed a "massively multiagent" title called Neural MMO, which will be "played" exclusively by machines.
For now, Neural MMO is fairly barebones. Neural network agents must compete over various resources throughout the game world, including food, water, and stone. The AIs within Neural MMO can also fight each other using one of three RuneScape-inspired combat styles: Range, Melee, and Mage.
The purpose of Neural MMO is to see how AIs would behave in a digital environment that more closely mimics the real world. To that end, the OpenAI team has learned a few interesting bits of information.
...increasing the number of AI agents in Neural MMO forced each one to explore the map more thoroughly to avoid competition...
For starters, increasing the number of AI agents in Neural MMO forced each one to explore the map more thoroughly to avoid competition - this is surprisingly human behavior that modern MMO players will likely be familiar with.
Furthermore, after only a few minutes of training, agents became more efficient at gathering resources, learning from each other and their own experiences.
Much like traditional MMOs, Neural MMO will continue to receive updates over time, which should force the AIs to adapt to new situations regularly.
It's tough to say how practical Neural MMO is from a research standpoint, but it's certainly a fascinating project. Let us know what you think of the concept in the comments below.