Imagine playing the latest version of Call of Duty against the computer. But, instead of just learning the patterns of the AI characters, you discover they can actually think for themselves. Suddenly there is no difference between playing the computer and playing against real gamers. Would this make the game better or worse?
The Video Game AI Debate
While almost every sector of tech has been obsessed with harnessing the power of machine learning and AI, video games have mostly been ignoring recent advances. Instead, most developers continue to use AI concepts that were cutting edge 40 years ago. There is a debate that is just starting to happen in the video game industry. Should advanced machine learning techniques be used in video games?
While machine learning and AI seem a natural fit for an industry whose every product relies on NPCs and other forms of AI, most developers resist using more advanced AI and machine learning in their games. The argument of these traditionalists is that AI that is too good would ruin the gaming experience. If you are not able to control the AI it can make it too hard to design levels that flow as part of a narrative. It would be like trying to make a movie where all the actors and even the set started to ad lib.
But, increasingly there are calls from a variety of different groups for video games to make better use of the latest advances in machine learning and AI. These voices include some daring developers, gamers, and even researchers from universities and private industry who believe that video games could do a lot to further machine learning research.
However, until a commercially successful product that makes use of advanced AI and machine learning makes it way onto the market, most developers will not be willing to experiment with something they view as unstable at best, and game breaking at worst.
What’s Happening in Machine Learning
Machine learning is a huge topic with hundreds of different niches. Some of the most popular and well-funded machine learning projects relate to the Internet of Things (IoT) and self-driving cars. Large tech companies like Apple, Google, and Facebook are some of the deep pockets behind these lines of research.
However, many researchers have also been using games as a tool to help teach machines. One of the early successes on this front was the creation of a bot that was skilled enough to regularly beat humans at Texas hold ’em poker. Poker is of special interest to researchers because to win consistently you have to understand statistics, human psychology, and the role of luck. While it easy to teach a bot the statistics of poker, teaching it about how chance and human psychology interact is a major challenge.
It is easy to see how almost every branch of machine learning research could benefit video games. A machine that could drive a car would also have the “skills” to navigate around a virtual battlefield. A bot that can beat people at poker, would make a worthy opponent in a first-person shooter.
Video Games as Teachers
The relationship between machine learning and video games is not just one way. Researchers have discovered that modern realistic video games make excellent teachers for machine learning purposes. The idea is that machines need to “see” a lot of data to know about the physical environment. The old way of doing this was uploading pictures and coding or labeling the pictures to teach the machine. It would explain what a wall is, what a road is, and so forth. However, realistic video games are now being used as a shortcut. The images are realistic enough for the computer to learn about the real world. But unlike pictures, video games have already “labeled” all the objects. This significantly reduces the time it takes to teach a machine about new objects.
As machine learning progresses, bots could be created to adapt to a video game environment. This would give players a more challenging gaming experience.
What Makes a Game Great?
Ultimately, the role of AI and machine learning in video games will come down to what makes a game fun to play? Developers are understandably worried about protecting the gaming experience. After all, for most players an impossible game quickly loses its appeal. Successful games balance challenges with successes. Will machine learning make this balance impossible?
There are also some security concerns about machine learning and video games. What is to keep a bot from going “rogue” and harassing players? Microsoft’s ill-fated chat bot comes to mind as an example of what happens when machine learning meets the real world.
Machine learning, advanced AI, and video games seem destined to be together. VR games are just now on the cusp of becoming mainstream. How much more fun and popular would a VR game be if the enemies you squared off against got better every time you played?
It will be the gamers that decide how big a role machine learning will play in video games. If the reward innovative and daring developers who use powerful machine learning technologies in their games with massive downloads, even the developers who are AI skeptics will be killing themselves to jump on the machine learning bandwagon.
For the video game industry, the best criteria for what makes a game great is still dollars and cents.