WHAT IS AUTOMATED GAME DESIGN?
Automated Game Design (AGD) refers to the procedural, algorithmic generation of game content or the complete (re)design of entire games, using random data inputs to create something that is unique but also adheres to certain rules. The discussion about AGD resides mainly in the procedural content generation (PCG) community and is often of a very technical, scientific, or even philosophical nature.
Let’s start our exploration of this subject from traditional game design, a time before there were random number generators (RNGs). In common perception, the creator of a game has no idea what specifically is going to happen in their game once players start to play it, but they know everything that could happen. All possible paths are mapped out and considered, both to ensure smooth gameplay with no dead ends and to maximise dramatic and narrative impact.
In game design in its purest form, without randomisation, there are no blank spots, no terra incognita. The game designer is like the all-seeing God, holding the strings in the background.
RNGESUS AND LADY LUCK
A step away from this omnipotent control are randomisers – basically, a list of things that could happen.
In the Diablo series, players encounter monsters, items, quests, even whole maps that are randomised in some way. In Borderlands and Borderlands 2, there are weapons which are created via randomisation, leading to weird combinations that even the developers were surprised about.
And here is where the discussion becomes slightly philosophical because how much does the designer really relinquish control by adding a few randomised bits and pieces? How much does a designer even need to control the narrative to make the game experience compelling and fulfilling? How much randomisation can you include before the result becomes essentially meaningless?
This debate is in no way finished yet and will be one of the most interesting facets of any future AGD development.
Some games exert so much control that the creator will know exactly what path the player must take to successfully beat the game. Even a single step away from that path would spell failure and require a restart of the level or the entire game. An example would be early shooters like Space Invaders, where every wave is pre-determined, or platformers like Super Mario Bros, which contains almost no random elements, and players will need to intimately learn the game ‘by heart’ to finally beat it.
Other games – often multiplayer games where players compete against each other, like Street Fighter or Mortal Kombat – do not need RNG because the human element provides much greater surprise and uncertainty than any RNG ever could.
ARTIFICIAL INTELLIGENCE (AI) CREATING WHOLE UNIVERSES
Still other games are so vast, no human designer could ever dream of creating them by hand. The world of Elite (1984) consists of eight procedurally generated galaxies – the developers originally wanted to go for 248 galaxies, but the publisher refused – with 256 star systems per galaxy and one planet and space station per system. PCG in Elite included galaxy composition, planetary conditions, and game economy.
In other examples, a single Minecraft world is 30 million blocks in all directions apart from height, each block representing one square metre. So, it’s 8x the size of Earth, with around 2.8 trillion Minecraft worlds that can be generated.
In No Man’s Sky, almost every object is procedurally generated, including galaxies (and their names), layouts and names of star systems and planets, all terrain, structures, and mission locations. The game was developed by indie studio Hello Games, and because they didn’t have the resources to create their game world by hand, they instead created tools to have this done procedurally.
Here we are in the realm of terra incognita. Technically, worlds generated from seeds, like Minecraft, might not be blank spots since they can be replicated, i.e. players can copy existing seeds to replay games in exactly the same worlds. But I think the point is still valid: designers are relinquishing control over their worlds in favour of countless possibilities.
2 OZ. OF CHARACTER, A PINCH OF PLOT, AND SOME SALT …
What about narration, can we have randomised plots, too?
For short stories, books, and movie scripts, there are numerous plot generators to find on the web. A simple click on a button will produce plots such as:
Protagonist: A knight, who is taciturn.
Secondary character: A dragon, who is rather incompetent.
Plot: It’s an urban fantasy story about toppling the establishment. It kicks off inside a dark cave with news of the destruction of the capital.
(Note that: the prophecy in the story will be delivered by a seer with a reputation for exaggeration.) And there's a twist! The world is inspired by 14th-century China.
Who wouldn’t want to watch/read this?
But all jest aside, such plot generators do serve a purpose because they stimulate creativity, push boundaries, and create ideas. They ignite sparks which can then develop into roaring fires.
Games with PCG sometimes do the same. In Dwarf Fortress (2006), PCG is not only used to create the map, but each world is also fully fleshed out with its own history, empires, nations, mythologies, and literature. This already comes very close to PCG storytelling, but the future seems to hold even more in store for this area of AGD.
NEURAL NETWORKS AS CREATIVE MASTERMINDS
OpenAI GPT-3 is a state-of-the-art language model with 175 billion parameters and the largest neural network ever created. It can produce (mostly, and often impressively) coherent text in fields like news articles, interviews and chats, poems, and even boardgames. A lot of these are almost indistinguishable from pieces written by humans, and similarly to the plot generators mentioned above, GPT-3 is already being used in games.
The site AI Dungeon, for example, creates text adventures like the classic ones by Sierra. Another one provides generated background stories for all kinds of roleplaying characters, complete with RPG stats. And the usage of GPT-3 goes even further: with the Workshop app, gamemasters can have it produce dungeons, cities, overland passages, and even creatures to fling at their players.
But not only narration, characters, and plot can be covered by AGD; pure mechanics are up for grabs as well: the board game Yavalath, and its sister game Pentalath, was invented by the computer programme LUDI and is “the first, and still only, computer-generated game to be commercially published”.
AI IN GAME JAMS
The biggest steps towards completely independent AI game design have probably been taken by the Angelina project by video game researcher and developer Michael Cook. What started out as a topic for his PhD in 2010 has become an engine that has successfully participated ‘undercover’ in game development jams.
Its latest incarnation, Angelina 6, has managed to produce a variety of games that are actually fun to play. And it has done so via the same route modern AI frequently takes – observing, testing, and evolving: “Guzdial fed the system two-minute gameplay videos of Super Mario Brothers, Megaman, and Kirby, and let it build a game based on what it had learned. […] Again, the game is remarkably good for what it is – something built by a machine based on watching six minutes of gameplay.” (AI Business)
It doesn’t always have to be about complete game creation, though. Just like those early RNG elements were merely a toolbox to create game parts that were too complex or laborious for humans, it is possible to foresee AGD becoming an integral part of future game design – just as much as a spell checker is part of a writer’s tool set.
Like in other parts of our daily life where AI has become a fundamental part of digital interaction, e.g. chatbots, smart assistants, spam filters, media suggestions, or navigation apps, the influence of AI in games will only grow in the future.
AI is being used to play against us (or even itself) when we’re lacking human opponents (or simply have no desire to play against them) – like the jumps in AI design for games like chess and Go, AI is now able to also compete on an even footing in games like Starcraft. AI also helps with the testing of games – and once it can develop games all on its own, we are finally free to spend our time on something else, like competitive dog grooming, dirt polishing, or duck herding!
AI can create a game from the ground up, test it, and then play against itself, and we are able to tune in from time to time like into a game on a sports channel. Glorious times ahead, indeed!