Video games generated by artificial intelligence, can they work?

Game News Video games generated by artificial intelligence, can they work?

Games entirely designed from A to Z by an artificial intelligence, the false good idea?

At the beginning of September, a paper published in the columns of Kotaku communicates the release of yet another game on the Steam platform. This Girl Does Not Exist, only available in English, is offered for a handful of euros, barely 3.99. But besides the article, the communication around the game seems to have failed somewhere; at the time of this writing, only three users have yet left ratings on the game page: “Cute Pen Games has a lot of ideas and probably an endless supply of energy to create other productions”, writes Captain Deatumner. “When you think about all the work it took to make this game, it’s an excellent result.t”, congratulates DDZ. Only a certain Fireflies seem disappointed with the experience: “The description makes the game more interesting and in-depth than it actually is.

The objective of This Girl Does Not Exist is to put together puzzle pieces depicting pretty girls as you go on dates with them. A concept to say the least summary that we already find in a slew of other suggestions at the bottom of the page. The illustrations are reminiscent of an old artbook by Chinese author Benjamin. But this is not the particularity of the title: The developer claims that everything from art and story to music and voice acting was generated by some kind of AI. The idea comes from a couple of developers united under the Cute Pen Games banner, rather accustomed to classic porn games. And while she might seem visionary, no one really takes the bait. The duo transferred keys to their project to no less than 250 YouTubers, with almost none deigning to respond. And despite a catalog full of NSFW content, This Girl Does Not Exist is the studio’s biggest commercial failure. Because despite the ingenuity of the process, it is the part of creativity that seems to be missing.

MidJourney, star of AI-generated games

If it is still in its infancy, no one is really surprised by the concept. Content generated by artificial intelligence is already flooding social networks, especially Twitter. DALL-E, MidJourney… these automated image generation tools only need a simple description to create sometimes stunning visuals. It is the latter which serves as the basic tool for the game of Cute Pen Game. This summer, another title tried out the magic of AI: Shoon, a small 2D shooter made in just three days by self-taught “Nao_u”. MidJourney shapes the background, the player’s ship and the enemies for him. Beforehand, the developer builds the ship models by giving the software text prompts related to Star Wars et Armored Core. Of course, several tries are necessary before obtaining something really usable. On the other hand, difficult to convince by beautiful prospects; the environments are static, impossible to distinguish clouds from vegetation, for example. The AI ​​has other limitations of the same ilk: the character sprites do not animate. If the presence of static spaceships can circumvent this problem, the concern still considerably restricts the field of possibilities.

But the process is still capable of pretty feats. In 2018 at the Georgia Institute of Technology, an algorithm designed by PhD student Matthew Guzdial and associate professor Mark Riedl absorbed hours of footage of people busy playing Super Mario Bros., Kirby’s Adventure et Mega Man. The machine learning system then used these sequences as the basis for its own titles. The duo notably developed Death Wallsa small game with sketchy illustrations in which the player tries to outrun a deadly wall which is rapidly approaching him. It’s the mimicry approach to creativity, which isn’t a bad place to start, because humans also learn to be creative by imitating at first.“, will tell Mr. Riedl in the columns of Vice. And beyond simple reproduction, the developed AI is also able to combine the design of several game levels in order to create new ones. But to really find the right recipe, we must first know where to relegate the AI ​​​​in the work group.

We don’t know how humans and AI should work together at the design level. The AI ​​could be an assistant, it could be a fill-in-the-blank contractor, it could be a peer, and – at the extreme – it could take the lead and the human could be the assistant.

In large studios, another priority

In large game studios, we still prefer to focus AI research efforts on very specific aspects of a video game experience, which generally translates into NPCs. In May 2021, Sony CEO Kenichiro Yoshida announced a strong collaboration between the company’s artificial intelligence research division, Sony AI, and PlayStation developers with a view to create intelligent computer-controlled characters. The latter would follow a reinforcement learning process by which an AI learns how to act by trial and error and would therefore become able to imitate human players. “By leveraging reinforcement learning we are developing in-game artificial intelligence agents that can become a player’s adversary or collaboration partner in-game” said Yoshida.

Related Articles:  Pokémon GO Shadow Jungle: Akala Island research challenges and rewards | cattone shiny | tapu lele | fomantis | niantic | Video game

The same year, Electronic Arts affirms its desire to develop tools capable of using machine learning to reproduce facial expressions and body movements from videos and photos. A system that would exclude the use of actors in a mo-cap studio. Even better, it was also planned to use user-generated content so that they could create an avatar by capturing their own silhouette from a smartphone or a webcam. At Microsoft, the Cambridge artificial intelligence research team is leading the Paidia project, which is similarly investigating the use of reinforcement learning.

The AI ​​of the adversaries is thus much easier to code than that of the allies managed by the computer. With the adversary, the relationship is brief. He doesn’t really need to understand what I’m doing, whereas an ally needs to understand what I’m doing, like outflanking enemies or covering me when attacked. This is why we often use orders, which greatly simplify the programming of allied AIs. – Journalist Jean Zeid for Polytechnique Insights.

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.