AI has taken on a central role in a lot of the things we’re doing. But can it really work as well as we think? The answer is yes it can, for AI can autogenerate parts of, or in fact whole games in real time. Of course, the output is still not as complex or of as high quality as those created in a studio over time, but the results are still there. So what can AI generate and how successful is it?
No code game creation
To kick things off, back in 2024, generative AI was used to recreate the game, Doom. However, it used none of the code or graphics from the game. Impressively, the Doom game recreated by the AI is completely playable, with researchers who tested this saying that other AI models could also be used to create games from scratch down the line. The model responsible for the recreated Doom game, GameNGen, was created at Google Research to see if it could. The game can only be played for 20 seconds, but it has all the same features as the original including scores, ammunition levels and map layouts. The illusion falls apart quickly though, as the AI runs out of memory. However, the way this worked is that the initial model was tasked to play the game over and over, while a second one simply replicated it, learning how millions of inputs caused changes in the game. This also didn’t happen in real-time, but over time, so what can AI models do now?
Using AI in the game development process
AI is also being used during game development. This is increasingly for all game types from major video games like The Witcher 3 or Elden Ring, to slots that you can find on the top online casino in UK. Now, instead of just tweaking code or minor basic background tasks, AI is creating the backgrounds, and not only that, it’s writing reactive dialogue for NPCs, changing gameplay within the game as you play and more. In fact, AI is changing the game quite literally, streamlining the gaming process in many different creative ways, reducing game crunch time and reducing massive teams to smaller, more focused ones. At this point, AI is taking on a lot of the heavy lifting associated with game creation, so its no surprise to find that some models can now do this in real time.
Real-time game generation
For some models, it’s now possible to simply input a description or concept art for the game, and the AI will create it. These are the games that, while they do work, often show the lack of finesse in the creation process. Yes, the model can do it, but the games lack quality, depth and complexity. As such, it’s still clear that more time is needed to make a good game, rather than doing it on the fly. However, the possibilities are most definitely there.
This is noticeable in other aspects of AI game generation. For example, many companies are switching to AI powered platforms to automate game development tasks. While this isn’t game creation per se, it does mean that AI is changing a game in real-time as it’s being played. This helps to improve the gaming experience and is already in place across a wide array of different games. There are experimental projects in the works too, such as those from Google’s Arts and Culture lab to explore AI’s potential in game development. And then there are things like Oasis, a sandbox similar to Minecraft that showcases just how well AI can build a game from nothing.
So, while the games aren’t great yet, it is entirely possible for AI to create playable games in real-time.