Ask video game characters out loud and see them respond to you in an improvised way, thanks to generative artificial intelligence (AI). This is the promise made by Ubisoft’s Neo NPC or Nvidia ACE, prototypes respectively presented at the Game Developers Conference in San Francisco on March 20 and at the Consumer Electronics Show in Las Vegas in January.
For the moment, no game placed on the market integrates these technologies using neural learning, which consumes resources and energy. But they raise the question: will they knock out the traditional AI which, in video games, defines the functioning of the worlds and the behaviors of non-player characters (NPCs)?
From its birth, AI has established a relationship with gaming. We find traces of this as early as 1956, when the name and principles of this scientific discipline were formulated during workshops at Dartmouth College, a university in New Hampshire. “Among the objectives, there was that of creating a chess player”explains Eric Jacopin, programmer for the Hawkswell studio and former head of the computer laboratory at the Special Military School of Saint-Cyr.
Ghosts and earthworms
However, we had to wait for the first successes of the arcade game for the general public to discover playful AI. The popularity of Pac-Man (1980) would not have been the same without the AI animating the four ghosts who face the gluttonous hero. It is thanks to her that they alternate between different roles – neutral, aggressive or evasive – in the famous Namco title.
Guided by the mathematical model of the automaton with a finite number of states, they move from one to another based on the player’s actions and periods assigned to each behavior. Superimposed on this system is the individual personality of each ghost. For example, red is a tireless hunter while pink sets ambushes.
Revolutionary in the early days of video games, these AIs do not bluff experienced players for long. “We can avoid behaviors because what is modeled is very simplesays Eric Jacopin. Ghosts are a bit like earthworms, which have only three hundred neurons in their brain and only react to simple stimuli. » In 1982, a book was published to learn how to anticipate them: How to Win at Pac-Man (Pocket Books, untranslated).
Ms. Pac-Man (1982), designed by the Americans at General Computer Corporation, then complicates the situation. The machine integrates an element of chance into the changes of state and the routines of the adversaries. “Humans are very good at anticipating things they have seen before. So when the ghosts became more random, the difficulty went up a notch and so did the fun. In AI terms, this is translated as moving from a deterministic problem to a non-deterministic problem.summarizes Tommy Thompson, AI consultant for the video game industry, who popularizes these subjects on the YouTube channel AI and Games.
You have 61.55% of this article left to read. The rest is reserved for subscribers.