🧠 Could AI Break Video Games?

Sign Up | Home Page | Meet The Team

In today’s Sunday Special:

Read Time: 7 minutes

📜THE PRELUDE

Before the character spawns, the video game makes a promise: the gravity won’t glitch, the map won’t morph, and the rules won’t rewrite.

This promise rarely crosses the gamer’s mind, but it’s what makes the video game count. Without it, every swing, every strike, and every smash would feel meaningless. Everything’s neutral, and that neutrality matters. Everyone’s on a level playing field, and that levelness holds the traditional concept of a video game together.

Suddenly, something shifts. A loot crate spawns out of nowhere. A wall materializes from thin air. A bridge forms over a bottomless pit. GenAI threatens to rewrite that promise. Soon, gamers will be able to reshape elements of the Game State as it unfolds.

So, how exactly does GenAI transform video games from static worlds into dynamic universes? How does it affect the shared experience of video games that unite the gaming world?

👾THE PHILOSOPHICAL FOUNDATIONS OF PLAY

In 1938, Dutch historian Johan Huizinga published “Homo Ludens,” discussing the importance of Play across culture and society. He proposed a formal definition of Play: a voluntary activity distinct from “ordinary life,” bounded by different time, space, and rules.

Consider popular sports like soccer, football, and basketball. On the surface, it seems super simple: fans cheer, players compete, coaches strategize, and referees call. Still, for everyone involved, the stakes feel like life or death. So, why do we take sports so seriously? Because they grant us something special: the chance to take risks, push limits, and sometimes accomplish the impossible.

According to Huizinga, the principles of Play create a Magic Circle: a space where the normal rules of reality are replaced by the artificial realm of a game world where different logic applies. This game world gives people the chance to experience emotions and express creativity in ways that everyday life doesn’t allow.

In 1961, French intellectual and prolific writer Roger Caillois expanded on Huizinga’s findings in “Man, Play and Games,” arguing that the concept of Play isn’t a single, orderly phenomenon but four dynamics that pull in different directions:

Despite their differences, Huizinga and Caillois both assumed something fundamental: the rules of Play are established and enforced by humans. In AI-generated video games, would the Magic Circle still apply?

🕹CAN WE MAKE VIDEO GAMES WITH AI?

Released in 1985, Super Mario Bros. lets gamers guide the protagonist Mario through the Mushroom Kingdom to defeat the antagonist Bowser and save Princess Toadstool. Along the way, Mario must run, jump, and stomp his way through eight worlds and 32 levels. He can collect power-ups like Fire Flower to defeat enemies like Goombas and Koopa Troopas with bouncing fireballs.

It was extremely popular, selling over 40 million copies worldwide, making Mario a household name and establishing the side-scrolling platformer genre. By the end of 1985, Japan’s highest-selling book was a strategy guide on how to beat the video game.

Virtuals Protocol recently developed “MarioVGG,” a text-to-video diffusion model that continuously generates playable Super Mario Bros. levels. It takes a brief text description of a gameplay action (e.g., “Mario jumping over a Goomba!”) and generates a corresponding interactive video clip.

It relies on a Generative Adversarial Network (GAN), which has two parts:

Over time, the Generator gets close to producing fake data that can fool the Discriminator. This process helps “MarioVGG” generate levels and worlds that resemble the actual video game.

Despite producing new gameplay, “MarioVGG” is just a pilot project. It possesses several critical limitations. First, it assumes the entire Game State can be inferred from gameplay alone. In reality, video games feature hidden mechanics like hitboxes, spawn triggers, and damage thresholds that aren’t visually explicit. Second, it only allows gamers to run, jump, and smash. In reality, video game engines must handle dozens of interactive elements to track player progress. Third, it’s far too slow for real-time gameplay, generating about 1 FPS, much slower than the required 60 FPS for smooth gameplay.

🎮PREDICTABLE WORLDS, INFINITE LEVELS?

Video games are manually written around the following Game Loop:

This Game Loop creates the illusion of an interactive virtual world. In reality, it just emulates the manually written video game software. Microsoft Research recently introduced “WHAM,” which influences the Game Loop by generating unique controller actions (i.e., inputs) and game visuals (i.e., outputs) that update the Game State (i.e., the current “snapshot” of a video game).

“WHAM” brings us closer to AI-powered video games that generate Game States tailored to each player’s decisions, including characters, levels, worlds, and gameplay mechanics.

🔑KEY TAKEAWAY

It turns out video games don’t require static worlds, but they do require shared ones. If AI-powered video games dynamically “help” struggling players and “test” skilled players, winning no longer signals superiority. Players may wonder: Was that gameplay element generated to “help” me or “test” me? As a result, AI-powered video games may seem less fair and therefore less meaningful. As GenAI reshapes gaming, video games stay meaningful only when changes are experienced relatively equally by all players.

📒FINAL NOTE

{{rp_personalized_text}}

Share your unique referral link: {{rp_refer_url}}