If you asked videogame fans what an interactive piece of entertainment idealized and not yet possible within 10 or even 20 years from now could look like, they could describe something disturbingly similar to the software presented in the science of Orson Scott Card. classic of fi Ender game . In his novel, Card imagined a military-grade simulation anchored by an advanced and inscrutable artificial intelligence.
The Mind Game, as it is called, is designed primarily to assess the psychological state of young recruits, and often presents its players with impossible situations to test their mental strength in the face of inescapable defeat. However, the game is also infinitely procedural, generates environments and situations on the fly, and allows players to perform any action in a virtual world that they could in the real world. Going even further, responds to the emotional and psychological state of its players, adapts and responds to human behavior and evolves over time. At one point, The Mind Game even resorts to the memories of a player to generate complete game worlds adapted to the past of Ender.
Leaving aside the most morbid military applications of Card's fantasy game (and the fact that the software finally develops sensitivity), The Mind Game is a solid starting point for a conversation about the future of video games and artificial intelligence. Why are games and artificial intelligence used both to help create them and to drive the actions of virtual characters, even remotely as sophisticated? And what tools or technologies still require developers to achieve this hypothetical fusion of AI and simulated reality?
These are questions that game designers and researchers are beginning to address as recent advances in the field of AI begin to shift from experimental laboratories to playable products and usable development tools. Up to now, the type of self-learning AI, namely the subset of deep learning of the wider revolution in machine learning, has led to advances in self-driving cars, computer vision and natural language processing that have not been incorporated. to the development of commercial games. . That's despite the fact that some of these advances in AI are due in part to software that has been enhanced through the game of video games, such as the unbeatable DeepMind AlphaGo program and the OpenAI bot Dota 2 that he is now able to overcome. professional level players.
But there is a point on the horizon where game developers could gain access to these tools and began to create immersive and intelligent games that use what is now considered cutting-edge research in artificial intelligence. The result would be development tools that automate the construction of sophisticated games that can change and respond to the comments of the players, and characters in the game that can evolve as they spend more time with them. It sounds like fiction, but it is closer to reality than we think.
To better understand how artificial intelligence might be more interrelated with video games in the future, it is important to know the shared history of the two fields. Since the early days of the medium, game developers have been programming software to simulate being a human being and to help create virtual worlds without a human designer needing to build every inch of those worlds from scratch.
From the software that controls a palette Pong or a palette Pac-Man up to the universe construction algorithms of the space exploration title Elite that was pioneer In the concept of generating game procedures, developers have been using artificial intelligence in a unique and interesting way for decades. By contrast, Alan Turing, a founding father of AI, developed a chess algorithm before there was a computer to run it.
But at a certain point, the requirements and final objectives of game developers were largely satisfied by the kind of artificial intelligence that we would not consider so intelligent today. Consider the difference between, for example, the goombas you face in the original Super Mario Bros. and a particularly difficult nightmare boss in the From Software action RPG Dark Souls ] Or the procedural level design of the 1980 game Rogue and the successful dungeon tracker of 2017 Dead Cells that made extensive use of the same technique to vary its design level every time you play. Under the hood, the delta between those old classics and the newer titles is not as dramatic as it seems.
What it does Dark Souls is so difficult that its bosses can move with implacable speed and precision, and because they are programmed to anticipate common human errors. But most enemy AIs can still be memorized, adapted and surpassed even by an average human player. (Only in very narrow domains, such as chess, artificial intelligence can, in general, force its way to a sure victory.) And even the procedural universes of a game as vast and complex as Hello Games & # 39; No Man's Sky are even created using mathematics and programming set by games like Rogue Elite and later ones.
The lack of significant leaps is due to the fact that the underlying IA that governs the behavior of these virtual entities, and the tools that generate AI procedures, have not undergone radical changes over the years. "Two of the main components of the AI business game are the finite-state road and search engines," explains Julian Togelius, an associate professor in the department of computer science and engineering at New York University who specializes in the intersection of intelligence. artificial and video games. "The search for ways is how to get from point A to point B in a simple way, and it is used in all games all the time." A finite state machine is a construction where [non-playable character] can be in different states and move among them. "
Togelius says that modern games use variations of these techniques, as well as more advanced approaches such as the search for Monte Carlo trees and what are known as decision and behavior trees, which are more sophisticated than that were in the early 80s and 90s. But most developers still operate with the same fundamental concepts and employ them at larger scales and with the benefits of greater processing power. "Of course, artificial intelligence in commercial games is more complex than that, but those are some of the fundamental principles in which you'll see versions from around the world, "he says.
Now, there is a big difference between the kind of AI you could interact with in a commercial videogame and the kind of AI that is designed for play a game in Superhuman levels For example, the application of Most basic chess can easily beat a human being in the classic board game, just as the IBM DeepBlue system outperformed the Russian grandmaster Garry Kasparov in 1997. And that kind of AI research has only accelerated in recent years .
In the laboratory owned by Google DeepMind, the AI research division of Facebook and other AI teams around the world, researchers work hard to teach videogames to play increasingly sophisticated video games. This includes everything from the Chinese board game Go to the classic games of Atari and titles as advanced as Dota 2 of Valve a competitive strategy of strategy of five against five that dominates the circuits of professional games of the world .
The goal is not to develop an AI that creates more interesting, dynamic and realistic game experiences; AI researchers are using games largely as a way to compare the intelligence level of a piece of software and because virtual worlds, with strict rules and rewards systems, are a particularly useful environment for training software. The hope is that by teaching this software to human researchers they can understand how to train machines to perform more complicated tasks in the future.
"First of all, the mission in DeepMind is to build an artificial general intelligence," Oriol Vinyals, co-leader in Google. The StarCraft 2 project owned by AI Lab said earlier this year, referring to the search for an artificial intelligence agent that can perform any mental task that a human being can. "To do this, it is important to evaluate how our agents perform in a wide variety of tasks."
It is precisely this type of AI, and the other advances that were similarly achieved in teaching software about how to recognize objects in photos and translate text into different languages, that game developers have largely avoided. But there's a good reason why most games, even the most recent big-budget titles that use the most sophisticated design tools and technologies, do not employ that kind of advanced artificial intelligence. This is because true self-learning software could make most games unplayable, either because the game played would be too unpredictable or because the AI would behave in a way that could tell a story or create a satisfactory feedback circuit for almost impossible players.
"Game developers tend to prioritize the types of actions we can predict. Although it's very interesting when AI does unpredictable things, it's not necessarily very fun for players, "explains Tanya Short, game designer and co-founder of the independent studio games KitFox Games." So, unless the game is built around the unpredictability of non-player characters, AI does not necessarily fulfill a great function when it is allowed to run alone. "