How Machine Learning AI Will Revolutionize Video Games



15 June 2020


If you asked video game fans what an idealized piece of interactive entertainment might look like in 10 or even 20 years from now, they might be describing something like the software featured in Orson Scott Card ‘s classic novel, Ender ‘s Game. In his book, which was later adapted to the big screen in 2013, Card envisioned a combat system based on sophisticated artificial intelligence.

The Mind Game, Card’s military-grade simulation, has been specifically developed to assess the psychological fitness of young recruits, and sometimes presents the participants with conditions that are difficult to test their emotional ability against imminent failure. But it also is an endless process that allows players to perform every action that is possible in a virtual world in the real one, creating environment and situations on the fly. Moreover, it addresses the players’ feelings and psychological condition, adapts to, reacts to, and over time evolves to replicate human behavior. The Mind Game builds on the experiences of a player to create entire game environments adapted to the life of Ender, the main character of the novel.

Moving on from Card’s militaristic application to artificial intelligence, The Mind Game is solid ground to begin discussing the future of AI and video games. Anyone who’s played Call of Duty knows that the AI in the popular shooter is not even remotely as smart as Card’s fantasy AI. So why not?

Although Ender’s Game came out in 1985, researchers are only just starting to toy with video game characters that have modern AI controlling them. Recently self-learning AI hasn’t been implemented in video games, mostly due to most games following a “pathfinding” system. Going from point A to point B limits the amount of choices not only the player has, but also the AI. If you think of Pacman, the AI in that game is extremely simple. The ghosts chase you, they don’t learn your movement patterns or practical routes in order to complete a level. The AI never learns, it’s a pathfinding system. OpenAI, a company funded by Elon Musk, created a team of five self-learning bots in order to compete against a team of pro-level players in the popular video game Dota 2 in late 2018. The team of AI lost handily to the two teams of professional humans. This shows that we’re still in the very early stages of adaptive AI in video games.

However there’s a potential for software developers to use these resources and start to build exciting and creative games that use what is now called state-of-the-art AI technology. The effect will be software for developers to use which will simplify the production of innovative games that would have AI alter and adapt to the player. It sounds like fantasy, but it’s closer than we might expect.

Action RPG Dark Souls was one of the pioneers in creating extremely difficult self-learning AI. The bosses in Dark Souls, infamously some of the most rage-inducing entities ever created, anticipated common human mistakes while using unforgiving speed and precision to beat the player repeatedly. However the bosses could be memorized easily after five to ten tries on a single boss. Although the bosses were prepared for human mistakes, they never altered their movement or ability patterns around what the player was doing. It was a start in the right direction for self-learning AI.

One of the best examples to take recently for the evolution of machine-learning AI would be Rockstar’s realistic Red Dead Redemption 2. Players can interact with nearly everything and everyone and won’t hear the same voice lines on repeat due to the fact that the AI will react differently to you because the character’s standing with the town, what clothes you’re wearing, even going as far as the character’s personal hygiene. The game’s scenarios never repeated themselves. The AI created new opportunities for the player, whereas usually the player would have to create new opportunities for themselves.

Another example would be AI Dungeon 2, an open-source text adventure game that uses a GPT-2 text prediction model to generate endless opportunities for the player. The player begins by picking a genre, like fantasy, mystery, etc. After, the AI generates a setting for you to interact with however you want. The AI even understands Star Wars references, as seen in this tweet. 

Perhaps the most exciting part of this view of the future is not only that a piece of software played a constructive role in the artistic building process of games, but also that such technology could produce customized experiences that are constantly changing and never grow outdated.

Leave a Reply

Your email address will not be published. Required fields are marked *