At the end of May, I gave a presentation on the underlying systems and tools that we used to develop the games Symon and Stranded in Singapore at the Procedural Content Generation workshop during the Foundations of Digital Games Conference. Most of the other presenters were computer scientists, as well as my friends. Thus I had a kind audience for this humanist to present the paper I wrote with Alec Thomson (now available online). Feeling a bit of the outsider in terms of background and methods, I also sensed the cultural differences between their approach and my own. In general, the presentations focused on generating the game (including mine). What I found considerably absent was a discussion of human factors: are these games playable? How does PCG transform how we make games? How does it change how we play them?
The workshop obviously had a technical focus, so when it came to talk about evaluating the systems, the discussion focused on how AI solvers / computer players were used to see if the game generated is consistent. Few of the presenters seemed to have used human players (more sophisticated and accessible AIs which you don’t have to implement) to evaluate their systems. On the other hand, there were presentations that dealt with the systems exclusively, not really dealing with why this approach was better for games apart from the pre-existing arguments efficiency in creating more content with smaller teams.
I guess that the presentations at the PCG workshop were clear examples of the proceduralist stance in game development, since the discussion of players seemed to be out of the question. Many of these presentations are more hypothetical, and implemented as early prototypes, still far from being actual games. I’m not saying that this is bad, we do need this kind of studies and tools. It was also the nature of this seminar, which was grounded on computer science, and the expectation seemed to be focusing on the systems and not players.
Throughout the workshop it became evident that we also need the space between procedural generation of content and evaluating that content through playtesting. After two years of working on games using PCG, the conclusion is that, in our case, we can generate procedurally generated narrative puzzles. It’s a lot of work, but it’s true it’s only half of the work. The other half is making them playable and fun. For that, I have less faith on AI and more on actual humans designing and playing.
I’m advocating the creation of a research space closer to HCI, where we study how procedural generation actually affects game design and gameplay. There is a need to study how the process helps both designer, the design process and the players. We need to see this in games that can go beyond academic experiments, that are played by people who don’t know and probably shouldn’t care that these games are part of research. Reaching out beyond the academic sphere is not easy: there’s Facade and Prom Night, and my own games Symon and Stranded in Singapore. (If there are more, please let me know in the comments!) We cannot feel snug about creating a system and making a game that our friends will play. If we want to make an impact on game development and design, we must take it a step further, we need to evaluate how games using PCG are played by people who are not those who developed the system.
There are already some easy questions that we can start looking into:
- If we think of content as something like puzzles, or level design, how do we provide cues for interaction to players? Think of hints to solve a puzzle, or user feedback about where to go. This is a common problem–Gillian Smith had run into these issues as well during the development of Endless Web. We can certainly design a system to provide these cues and feedback, but the best way to do it would be studying how players interact with the game first.
- What are the aspects of game development that can use procedural generation best? Design? Art? Code? QA? Writing?
- All the designers I can think of working on PCG come from computer science. How can we make procedural content generation accessible to non-programmers, or at least people who don’t have a strong background on CS?
- What mechanics and fictional worlds fit PCG best? I believe PCG is one approach to game development, but not the only one. After working on Stranded in Singapore, one of the conclusions was that it was really hard to design puzzles to be procedurally generated when they were based on the real world. Dreams, on the other hand, seem to be a good match for PCG, as seen in Symon and Endless Web.
These are the immediate questions that come to mind, based on my experience making games. I have a few preliminary answers for some of these, but we need to expand our thinking on what PCG means with relation to games.
The workshop taught me (amongst many other things) that many of the people working on PCG already take playtesting as part of their process. There were also slides that made my blood curdle, which reduced human behaviour to mathematical formulas. One presenter had a formula for “fun” depending on the type of player (who I’m guessing it’s also determined with another mathematical formula). Another presenter called the story “filler” in the context of RPGs, which can be just generated to give you a motivation; when I called him out, he admitted that it may not be the best term. The fact that human feelings and behaviour are reduced to numbers, and that narrative is considered filler, may be symptoms of the subconscious disregard certain computer scientists may have for human behaviour. This is one of the dangers of focusing on the procedures so much: look at the screen for too long and one loses sight of play as a human activity, doesn’t question how our brains and hearts fill the gaps so that we don’t have to really generate all the content, and stops giving enough credit to players. By approaching fun and storytelling as things that are generated mechanically, negating that fun is an awfully vague concept (and non-quantifiable), and that stories are not only about events but about worlds and the people in it, we’re heading towards playing with mathematical formulas empty of human meaning.
I’m probably preaching to the converted–most of my friends working on PCG will talk about their playtests and what they learned from them. This post is calling out a certain type of discourse which, also necessary, also seems to leave out the humanity of games. Rather than complaining (too much) about it, we should see this as an opportunity to opening up a new area of study. The combined study of procedurally generated content and human-computer interaction is waiting to happen.Who’s up for it?
4 thoughts on “Thoughts on Procedural Content Generation”
It may be wildly different from what you’re looking for, but have you ever stumbled upon Dwarf Fortress (http://www.bay12games.com/dwarves/) ? The player base seem to have accepted a lot of the quirkiness that comes from PCG as an essential part of the game, embracing it and making it a kind of very loose way of adding a narrative part to the game (it’s normally a city-building fantasy roguelike). There`s even fiction that derived of it (here : http://www.bravemule.com/).
I am indeed aware of Dwarf Fortress and the derived narratives, and even tried to go through the tutorial. The realm of my research, however, is limited to interactive fiction and point-and-click adventure games. There is a wide variety of approaches to games using procedurally generated narratives, such as Prom Week or Mimesis, and each one of us takes a different genre in order to develop new design paradigms and technologies.
I think that accepting the quirkiness of a PCG system is fine, but not a good goal in general. I think there’s a danger that gamers, being quite tech-savvy and interested in new developments, are actually too accepting of new technologies and too willing to be okay with systems which have flaws.
DF produces some fantastic stories, but its reliance on hardcoded systems (which give rise to the funny and enjoyable anomalies) represents something we should work towards finding solutions too, so we can employ the same techniques in real-world scenarios. It’s a bit like the progression from Symon to Singapore; by making the systems more sophisticated we can attempt settings and scenarios that give less leeway to PCG and are less failure-resistant. Which is great.