Capturing the spirit of open-endedness in an algorithmic process has long been a focus of the artificial life community. While the exact nature of open-endedness research is difficult to articulate precisely, it essentially aims to replicate the chain of endless innovation achieved by biological evolution in nature. The games community, too, has similarly long aimed to devise such creative algorithmic processes . . .
The following is part of a series of short essays exploring the intersection of game development, artificial life, and simulation. In each entry, I examine a single paper from my perspective as a cross-domain developer. As someone who is passionate about these fields, I offer analysis and opinions of the research and discuss how it might be applied through interactive means. Join me on this exciting journey as we delve deeper into the world of games and artificial life.
Note: I will provide a link to the paper in the form of a citation, if it is not behind a paywall.
I. Summary
The abstract states “this paper argues that video games serve as a complementary domain for research on open-endedness. In support of this claim, experiments in this paper evaluate the effects of age-based and spatial destructive events in two game domains . . . ” After providing background, the paper sets up an experiment involving two games: The Game of Life, and SimCity. This experiment is designed to “investigate the impacts of destructive events on reward and complexity of game states achieved by a reinforcement-learning agent . . . ” The gist of the methodology is as follows:
- For each domain, have the ANN play several iterations of the provided game. It is to play each iteration for the same number of steps.
- The control is played without destruction events. Subsequently, games are played with one of three event types, occurring at regular intervals.
- The length of the intervals is parameterized.
- A visual snapshot of the board-state is taken and compressed.
Ultimately, the degree of compression provides a useful metric for describing how complex a given board-state is.
II. Understanding
Without the context, it might be difficult to understand what this has to do with open-endedness. As stated within, large-scale destruction has the potential to jump-start evolution on many scales. They claim there is little research showing this phenomenon in artificial life. In this case, the greater the complexity, the more the board-state could be said to be flourishing.
The key to understanding rests at the very start of the paper, “The term open-ended evolution (OEE) was coined by the ALife community to describe processes in the spirit of natural evolution.” My favorite definition of OEE is therein listed, “A system in which components continue to evolve new forms continuously, rather than grinding to a halt when some sort of ‘optimal’ or stable position is reached.” As one might expect, trying for high fitness can lead to stagnation or adaptation bottlenecks.
The destructive events provide a clean slate, open space and newly untapped niches. As such they can assist in resolving evolutionary roadblocks at many levels of observation.
They write, “An important and interesting question to consider is how the concept of openness in games corresponds to notions of open-endedness in artificial life.” This is an interesting question, as is its opposite. They initially limit this possibility space by framing the goal-oriented predestination present in most games as a detriment to open-endedness, but are quick to note that even OEE needs this to some degree. In the end they ponder “where the boundaries between these disciplines truly lay and whether we can all benefit from each other’s insights.” The remainder of the paper will be left for the curious, as it is available below.
III. Insights
There are several large takeaways for me as I begin down the long road of developing Sap:
- I intend on including a complexity measuring system as a universal metric for SAP, solving one of the existing prototype’s weaknesses – the Bioscore. While it serves its function in card selection, the fact that it is a contrived statistic limits its utility. I’m convinced this new technique will provide a more thorough review of the system’s effectiveness.
- I am on the hunt for novel solutions that can improve the system’s capabilities while developing the simulation at the heart of SAP. The idea of introducing destruction events is a prime example of such a concept. While it may seem simple, it has significant potential for improving the system’s performance. Personally, I value this idea because it is one that I was already familiar with, but had not considered incorporating into the simulation.
- In my efforts to foster exploration and keep the board fresh, I was in search of a solution that could sustain an increase in complexity and prevent the system from becoming stagnant. I believe that developing a system capable of repeatedly disrupting stable states is exactly what I was aiming for. It has the ability to stimulate exploration while preventing players from falling into predictable habits, making the game more engaging and dynamic.
Thank you for reading. I hope this was interesting and not too long. I encourage you to go read the paper. It’s not long and really worth it if you’re developing games with even a shade of open-endedness.