Camping at an orb location is a solid strategy. Players must pick up orbs to win (imagine Pac-Man’s ghost just hovering near the entrance to each corner of the map). It also makes the game less fun. Players no longer experience exciting chases. Instead, the AI can spawn unpredictable ambushes. Trachel and Peyrot say their goal is not to create “superhuman bots. It’s about finding ways to incorporate it into game AI tools.”
For players who crave better AI, it may sound boring. Still, the machine-learning technique that Trachel and Peyrot demonstrated is useful for adjusting difficulty even if the enemies the player faces in the finished game don’t use it. Julian Togeriusco-founder and research director Modl.aihas spent nearly five years testing games with AI. Modl.ai uses bots to look for graphical glitches, spot flaws in world geometry, and sniff out situations that make victory impossible.
“Tell us what kind of failure conditions you’re interested in, and it’ll basically run. Submit a job and it will run according to the scope you want to explore,” says Togelius. “And of course we can put these together and give you a report, where you think things are going wrong here, etc.”
Modl.ai’s test bot uses machine learning to adapt to each game tested, but the current implementation limits these adaptations to specific titles. According to Togelius, the company is working on a prototype to add Deep Learning, which trains the bot’s behavior in multiple games. Once used, Modl.ai’s bots should learn to emulate the behavior of real players and more efficiently spot issues players find.
Game engines need a revolution for true machine learning
When things get tough, machine learning can be both the problem and the solution. But creating fair and fun challenges isn’t the only hurdle facing developers who want to use machine learning in their games. The problem goes deeper and, in fact, we may have to rethink how we build our games.
Performance is one barrier. Machine learning requires a large amount of training data to get valuable results, and that data can only be obtained by playing the game thousands or tens of thousands of times (bots can lighten the load, Trachel and Peyrot are the tactics I used to build the demo). Also, once training data is collected, running the resulting model in real time can be expensive.
“Yes, performance is clearly an issue, especially with large ML models that process frames on every tick of the game clock,” Trachel and Peyrot said in an email. “In our case, to avoid performance issues, we used a small neural network that only guessed at exact moments in the game.” , is a completely different matter.
Togelius says the way modern game engines work exacerbates the problem. Machine learning will inevitably be slower because “game engines weren’t built for this. One of the many reasons we don’t see more interesting modern AI in games is Unreal and Unity and Because all the likes are basically terrible and anti-AI in so many ways.”
Animation is another matter. Most modern game engines expect animation to be strictly defined on a frame-by-frame basis. This works well when the animator knows for sure how her character will behave in the game, but AI controlled by machine learning can behave in ways the animator doesn’t expect. I have.Designers can avoid this A physics-based approach to animationbut this puts an even greater performance burden on the game console or computer hardware and comes with its own development challenges.
In short, the developers are facing a monster of their own making. The game engine is built using behavior trees and predetermined actions to create an AI-controlled world of his NPCs that works well even on poor hardware. But as machine learning gains momentum, we need to rethink these traditional solutions.
“If you talk to machine learning researchers who don’t know game design, they’ll say, ‘Why don’t we take something new and get NPCs that are more realistic and adapt to the way you play?’ is,” says Togelius. “But you can’t just plug this into an existing game. You have to rethink what a game is.”