By
Alessio Jacona *
Artificial intelligence and video games have long been closely linked.
As recalled by a report drawn up by IIDEA, the trade association of the Italian videogame industry, they are both a training ground for machine learning algorithms and fertile ground for their application, for example to improve intelligence and expressiveness. of non-player characters, the realism of the gameplay or even the matchmaking (pairing) between players of equal skill level.
«AI is part of the essence of video games», explains Lorenzo Fantoni, journalist and author of the splendid essay “Living a thousand lives - Family history of video games”.
"The first videogame experiments - he continues - were born precisely to measure the possibilities of an artificial intelligence, and already the ghosts of Pac-Man were an AI", while today there are even video games "that use them to read the player's behavior, so as to be able to challenge and contrast it, but in this sense there is still a long way to go ».
DeepMind, the popular AI company controlled by Google (via Alphabet Inc.), uses older Atari 2600 console games as a test environment for its deep learning algorithms.
Rising to the headlines in the world news also for having created Alphagi, an algorithm capable of playing the game of Go (very popular in Asia) that in 2016 defeated the Korean champion Lee Sedol, the company has also collaborated with the publisher Activision Blizzard: together they made an AI for the Starcraft 2 video game (called “AlphaStar”), which was able to beat 99.8% of human opponents.
But it is when artificial intelligence is used to “augment” the video game that things get even more interesting: «There are cases in which AI is used to create procedural environments», reveals Fantoni.
"A series of scenario elements are fed to the algorithm and then it is artificial intelligence that creates an ever-changing environment".
The example of "The Last of Us 2" is also fascinating, "a title particularly rich in details in which the facial expressions of the characters are not decided by the programmers, but are chosen by an artificial intelligence system that chooses the best emotion based on context, and manages the character's facial muscles to make it happen.
In this way the animators do not decide the reactions a priori, the AI takes care of finding the right one for each context ».
This is the beauty of the so-called “triple-A” video games, whose budgets and processing times equal and sometimes exceed those of the great Hollywood productions.
And which are useful for the development of artificial intelligence for other reasons as well: for example, because to create them the developers have been using film technologies such as motion and facial capture for years, in order to integrate the performances of real actors within the storytelling.
This practice, created to make video game characters more realistic, has in fact created a vast digital archive of movements and expressions that can now be fed to the AI, so that they can learn to recognize them and generate them in turn - without the intervention of the man - to further improve future video games.
A remarkable virtuous circle, put to good use, for example, by researchers from Electronic Arts (producer of the football videogame series "FIFA" among others): thanks to a partnership with researchers from the University of British Columbia in Vancouver, they have applied a technology called reinforcement learning, to instruct the AI in the automatic animation of humanoid characters, thus obtaining the animation of a footballer in an extremely realistic way.
Another fundamental element of triple-a games is the reproduction of huge open worlds (the so-called open worlds) that the player can explore in complete freedom, and which are particularly complex to develop entirely by hand.
Today, however, it is already possible to exploit the LIDAR mapping (i.e. remote sensing by laser pulse) of surfaces and buildings, to help AI understand how to autonomously generate increasingly realistic (and adherent to reality) territories.
For the famous series of video games with a historical setting "Assassin's Creed", the Ubisoft developers have perfected technologies that have made it possible to reconstruct some of the most representative geographical areas of human history: in the last chapter, "Assassin's Creed: Odyssey", for example , the entire Greece of the 5th century BC, with its regions, islands and most representative cities, is reconstructed (and freely explorable by the player, on foot, on horseback or by sea).
Another application of artificial intelligence in video games allows for example to track the gameplay, that is the behavior of players within a video game, to adapt the behaviors of non-player characters (controlled by the AI), and therefore offer appropriate challenges to the their skill level.
Or, as we said at the beginning, to refine matchmaking algorithms in a multiplayer environment.
Here an example comes from the Milanese developer Milestone, specialized in the creation of racing games on two wheels, who developed together with the Orobics company an AI called ANNA (Artificial Neural Network Agent) and inserted it into the “MotoGP 19” game.
Trained at a rate of about 200,000 laps per day and replicated in groups of virtual drivers, ANNA has honed her driving skills on the circuits, improving judgment skills during races and thus making her behavior on the track as much as possible. close to those of a true professional driver.
Also through AI it is possible to guarantee greater safety for players, for example by applying filters that analyze the interactions between users in search of incorrect or violent behavior and offensive or harmful content (Microsoft's Project Artemis goes in this sense), but also to protect fair play by quickly identifying those who prefer to cheat to win, perhaps by exploiting some bugs in the software or specific programs.
In short, since Pac Man, the love story between video games and street artificial intelligence has made a lot of it.
And we are only at the beginning.
* Journalist expert in innovation and curator of the Artificial Intelligence Observatory ANSA.it
