ART-ificial Intelligence in Gaming Part 1: Three Introductory Examples
(Super Mario Bros., DOOM, and NVIDIA’s Recreation of Pac-Man)
“It’s a-me, Mario!”
If you don’t know this catchphrase, I don’t know what planet you came from or why you clicked on this blog post, but still feel free to tag along :)
Hopefully, by now, you’ve caught onto why my series is called “ART-ificial Intelligence in Gaming.” If it isn’t clear, “ART” is emphasized because we’re talking about graphics, or the “art” in video games.
As I mentioned in my first blog post, today we’re exploring how AI can be incorporated in video game graphics. I chose three simple examples — focused on virtual world development — to keep this post short, but there are many other games that use AI in different ways. If you scroll down to my conclusion, I will also point you to two bonus examples focusing more on developing a realistic gaming environment.
Example 1: Super Mario Bros.
Super Mario Bros. is a super popular game, but in case you haven’t heard of it before, it is a platform game based on controlling the iconic mustachioed man wearing red overalls to collect coins, defeat enemies, and avoid obstacles to reach the princess. As the player advances to each new level, the game gets more difficult to beat.
Since we’re talking about implementing AI, your mind probably went immediately to an algorithm that can beat the game for us and score the most points possible. Though that is a valid AI to make, why don’t we instead find a way to create an endless amount of even more challenging (but still possible) levels? Is there a way to train a model to generate consistent, unique Super Mario levels that are increasingly difficult but still engaging? The answer is yes. Find out how in my next blog post, but for now, let’s move on…
Interesting, but a little off-topic
It’s odd how people would be interested in a game where you constantly fail by being demolished by enemies or falling in pits. A couple years ago, the YouTube star and former NASA engineer, Mark Rober, presented a TED Talk about the Super Mario Effect.
This is what I call the Super Mario Effect.
Focusing on the Princess and not the pits, to stick with a task and learn more.
Rober conducted an experiment on 50,000 of his YouTube subscribers and split them into two groups for participating in a basic coding challenge. The minor difference between the groups was that one was penalized for doing something incorrectly, while the other received no punishment. He found that the group who was not punished put in an average of 12 tries before they gave up, yielding a 68% success rate; the punished group put in an average of 5 tries before they gave up, yielding a 52% success rate. Video games such as Super Mario Bros. have minimal punishment, which is why people are so determined to advance to the next level and feel the sense of accomplishment. Instead of focusing on all their failures, they highlight the excitement felt when a challenging level is cleared. The Super Mario Effect is actually meant to be applied to real-world thinking, but I’m using it to emphasize why people are engaged with the game and why they would enjoy AI-generated levels. I highly encourage you to listen to the entire TED Talk, but enough of this tangent — back to the examples.
Example 2: DOOM
The second example is one of the classic first-person shooter games from the early 1990s. Similar to Super Mario, DOOM also has unique maps for defeating enemies and navigating through obstacles. But instead of jumping on cute brown mushroom men, the space marine of DOOM shoots various demons and zombies.
Like the previous example, we won’t be focusing on creating an undefeatable AI player. Rather, we’ll learn how to develop an algorithm for creating compelling levels to enhance player experience. I included this example mainly to show that AI can be used in both two-dimensional and three-dimensional games, as well as with a variety of video game styles.
Example 3: NVIDIA’s Recreation of Pac-Man
The third and final example I will be sharing with you is NVIDIA’s recreation of the classic arcade game, Pac-Man. If you don’t know this game, you seriously live under a rock. Essentially, you guide the yellow munching circle around the map to eat all the dots, collecting power-ups and avoiding the colorful ghosts along the way. Each level gets more difficult.
NVIDIA is a computer game company, most well known for their graphics cards, but they also conduct plenty of research in the field of artificial intelligence. Earlier this year, NVIDIA Research revealed their new AI model, GameGAN, that trained on 50,000 PAC-MAN episodes and later recreated a fully functional version of the game without an underlying game engine. There are a ton of articles surrounding the powerful AI model, but to sum up all the hype with informal language: GameGAN is pretty much a baby who blankly stared at hours upon hours of gaming until it figured out how to replicate what it was seeing, without the help of anything or anyone.
CONCLUSION
Those were my three introductory examples of AI in video game graphics, specifically virtual world development. As promised, my two bonus examples for more realistic gaming environments:
- The Legend of Zelda: Breath of the Wild
- Grand Theft Auto V
I won’t go into too much detail about how these utilize artificial intelligence, but the main application is the development of a world with applied laws of physics, creating immersive virtual environments.
My next post will dive into how exactly the AI is trained for the three examples we just explored. It will be the most content-heavy post of the mini series and will focus on generative adversarial networks (GANs, which should look familiar because of Example 3).
See you soon :)
Danielle Trinh is a Student Ambassador in the Inspirit AI Student Ambassadors Program. Inspirit AI is a pre-collegiate enrichment program that exposes curious high school students globally to AI through live online classes. Learn more at https://www.inspiritai.com/.
