Like when any major news story breaks, however, I had one question on my mind: How will this impact video games? If AI is putting the work of writers and digital artists at risk, who’s to say it won’t one day do the same for game developers? Is it possible that AI might one day be able to make video games all on its own? I did some research, and here’s what I found.
Artificial Intelligence in Gaming: An Overview
As many of you may know, AI isn’t anything new when it comes to video games. They have played a pivotal role in giving games the level of interactivity that they have today. The most infamous example of AI involvement would be Finite State Machines (FSMs). These were behavioral models used to control NPC behaviour, allowing them to switch between certain “states” (such as “attack” or “flee”) to react to the player in real time.
The ghosts in Pac-Man showcase FSMs excellently. Outside of AI, other types of software have also done a lot to help with game development, such as procedural generation. This system allowed developers to automate the design of virtual worlds, requiring minimum supervision.
The original Mass Effect had one of the most popular uses of Finite State Machine AI
None of these systems come close to the power or potential of generative AI however. Ultimately, this new technology can create brand new content seemingly from scratch. It’s hard to strictly say how it works, as not all generative AIs function in the same way. At present generative AI is limited to tackling relatively simple content, such as images or songs. But should this technology advance enough, could generative AI eventually mass produce video games, thereby making human developers completely redundant?
Many might write off this line of thinking as something taken straight out of science fiction. But as it turns out, AI developed games already well underway.
“AI Generated” Gaming Experiences
Depending on who you ask, “fully AI generated” may already be on the market. A great example would be the unfortunately low-rated title This Girl Does Not Exist. This Steam-exclusive had an AI generate all of its assets, such as its art, its characters, and even its voice-acting. You can argue though that this isn’t exactly “fully AI generated”. After all, a human was ultimately pulling the strings here. And as revealed in Kotaku’s interview with the game’s creator, it’s clear that a lot of human input was needed to get the AI to do its job correctly.
This is something with which the AI struggles, how to generate images of the same person, yet in different poses / setting… I had to rerun a lot of the commands and try many times until I got out of it a set of pictures which would be the ‘same person.’
-Discord name mrspotatoes, co-developer of This Girl Does Not Exist
All other examples of games supposedly “created by AI” are roughly in the same boat. They all have AI algorithms developing parts of a game with humans overseeing and directing the entire project. You can find examples of these all over the internet. It is important not to underestimate the role that AI is playing here however.
These examples show AI directly handling aspects of game development previously reserved for people. With enough time and research, this technology could improve enough to have AI take over from humans in certain gaming industry studios. But even with this in mind, I wasn’t fully convinced that AI was that much of a threat for game devs. That is, until I read about ANGELINA.
The Curious Case of ANGELINA
A True AI Game Developer
ANGELINA is the brainchild of Mike Cook, an AI researcher, game designer, and lecturer at King’s College London. When Mike took on a PhD, he asked a simple question: Can you develop software that can make video games all on its own? And to boil down many years of work and research from him and his colleagues at the University of Falmout, he found the answer was “Yes”.
As part of his research, their team produced an AI named ANGELINA (short for “A Novel Game-Evolving Lab-Rat I Named ANGELINA”). This software could, quite simply, develop games at the push of a button. Humans need not apply. This digital developer works by using something called “Evolutionary Computation”. To make a game, ANGELINA starts by randomly throwing together a series of common video game assets ( eg. non-player characters, collectibles, power ups, etc.) and play testing it repeatedly.
The software then evaluates the game it just developed, and moves on to making a new game. With this new title, ANGELINA then tries to filter out what was “bad” about its previous attempt, keeping in what it thought was “good”. This cycle repeats with its games getting better and better, a practice quite similar to natural selection.
Though we’re quite different, it turns out AI can evolve in the same way as us
ANGELINA VS Game Developers
Mike has previously entered ANGELINA in game jam competition Ludum Dare to check how it would fair against human game devs. The short answer? Not well. He entered two of ANGELINA’s video games into the competition. The first submission, To The Sect, openly stated that it was made by an AI. For the second title however, Stretch Bouquet Point, Mike lied and passed it off as his own creation.
To The Sect, did pretty okay, seemingly because audiences were impressed by ANGELINA and her capabilities. Gamers gave much more critical feedback of Mike’s second title however. Stretch Bouquet Point was poorly rated, receiving a host of criticism and claiming the 551st place overall. It appeared that without having the novelty of being described as “AI generated”, ANGELINA’s games really didn’t seem to stand out. You can judge the gameplay for yourself by watching Mike play through To The Sect below.
But the ratings here aren’t what’s important. The amazing work of Mike and his team have demonstrated that it truly is possible for an AI to do the full game development process. Mike himself has stated he has no desire to replace game developers however. Instead he is trying to develop software that can work with humans, rather than against them (see his latest development, “Puck”). Though now that we know this technology exists, some would say it’s only a matter of time until it becomes just as, if not more, capable than us.
So is that it? Does it look like the jobs of many in the gaming industry are inevitably doomed? From the point of view of non-experts like myself it would certainly look like it. But those with a lot more experience are far more sceptical.
The Problems With Generative AI Game Developers
The Doubts of Experts
Many experts both in AI and modern game development say that there are a lot of obstacles standing in the way of game-generating AI. Creating simple video games with basic graphics is one thing. But creating immersive experiences that can rival some of the amazing game worlds being produced by humans is a different kettle of fish. Cyberpunk 2077 developer Patrick Mills argues that there is still a large gap between having functional AI game developers and reality.
I have seen some, frankly, ludicrous claims about stuff that’s supposedly just around the corner. I saw people suggesting that AI would be able to build out Night City, for example. I think we’re a ways off from that.
-Patrick Mills, Franchise Content Strategy Lead of CD Projekt Red
This sentiment is echoed by Julian Togelius, an Associate Professor at the Department of Computer Science and Engineering at New York University’s Tandon School of Engineering. Togelius has been authoring papers on the role of AI in game development for years. While he definitely is interested in the role generative AI could play in the art form, he describes the notion of a sudden AI takeover in the gaming industry as “bullshit.” It’s clear that this technology won’t quite match human design skills in gaming for a while yet.
It’s possible that there won’t be much competition between game developers an AI in the near future
Other Logistical Difficulties
Perhaps the biggest blockade for fully functioning AI game-makers however is generative AI itself. This type of software typically works independently, producing unpredictable content with a complexity that humans have little to no control over. For projects requiring more precision, generative AI may never quite fit the bill. As stated by Latent Tech CEO Jorge del Val Santos in an interview with Axios, working with this kind of software always involves a “tradeoff between creative control and speed of creation and immersion”.
At the end of the day, there will always be games that will need a level of developmental accuracy that AI can’t, and may never, be able to deliver. Therefore, to whatever extent, AI developers might always just be a shadow of their human counterparts.
Game development may always look like this…
Conclusion
I still agree with my initial statement. Eventually AI will be able to generate high-quality gaming experiences without the need for any human intervention, likely replacing some of the game-making roles we have today. The technology needed to do this is out there, and will only continue to get better. There are two large caveats with this however. The first is that it is quite possible that there will still always be a need for human developers in some capacity. The second is that nobody knows for sure how far away these advancements are. An AI rendition of Dark Souls might still be a hundred years away.
The fact of the matter is that generative AI is pretty overwhelming. It’s understandable for us to be wary of this kind of thing. But while it’s easy to look at the doom and gloom, it’s important to remember that the true future of AI is unknown. How these innovations will impact humanity is yet to be decided. The best we can do however is to look at these new discoveries optimistically, and use our power within democratic societies to guide their integration with humans as best as we can.
