Procedural Content Generation via Reinforcement Learning (PCGRL) offers a method for training controllable level designer agents without the need for human datasets, using metrics that serve as proxies for level quality as rewards. Existing PCGRL research focuses on single generator agents, but are bottlenecked by the need to frequently recalculate heuristics of level quality and the agent's need to navigate around potentially large maps. By framing level generation as a multi-agent problem, we mitigate the efficiency bottleneck of single-agent PCGRL by reducing the number of reward calculations relative to the number of agent actions. We also find that multi-agent level generators are better able to generalize to out-of-distribution map shapes, which we argue is due to the generators' learning more local, modular design policies. We conclude that treating content generation as a distributed, multi-agent task is beneficial for generating functional artifacts at scale.

In late August 1993, a young programmer named Dave Taylor walked into an office block on the Lyndon B Johnson freeway in Mesquite, Texas, to start a new job. The building had a jet black glass exterior and sat utterly incongruent amid acres of car parks, single-storey industrial units and strip malls. Game designer Sandy Petersen called it the Devil's Rubik's Cube. Taylor's new workplace was on the sixth floor in office 615. The carpets, he discovered, were stained with spilled soda, the ceiling tiles yellowed by water leaks from above.

Cermati is a financial technology (fintech) startup based in Indonesia. Cermati simplifies the process of finding and applying for financial product by bringing everything online so people can shop around for financial products online and can apply online without having to physically visit a bank. Our team hailed from Silicon Valley Tech companies such as Google, Microsoft, LinkedIn and Sofi as well as Indonesian startups such as Doku, Touchten. We have graduates from well known universities such as Universitas Indonesia, ITB, Stanford, University of Washington, Cornell and many others. We are building a company with the same culture of openness, transparency, drive and meritocracy as Silicon Valley companies.

Generative AI bots have been taking the internet by storm, allowing anybody with a network connection to put in a prompt and watch a piece of software complete it. The latest trendy bot is ChatGPT, developed by OpenAI, which has generated conversations, scripts, text-based games, and even fully-fledged articles to varying degrees of success. DALL-E 2 and Midjourney are other popular programs that produce images based on text prompts, and there are countless more. There are a lot of ethical questions surrounding the use of AI-based tools in creative work, but in video game development, they're only getting more popular. According to George Jijiashvili, an analyst at Game Developer sibling company Omdia, AI tools will be "the hottest topic in games tech" in game development over the next year, with startups launching to fill the space.

There has been significant research interest in Procedural Level Generation via Machine Learning (PLGML), applying ML techniques to automated level generation. One recent trend is in the direction of learning representations for level design via embeddings, such as tile embeddings. Tile Embeddings are continuous vector representations of game levels unifying their visual, contextual and behavioural information. However, the original tile embedding struggled to generate levels with skewed tile distributions. For instance, Super Mario Bros. (SMB) wherein a majority of tiles represent the background. To remedy this, we present a modified tile embedding representation referred to as Clustering-based Tile Embedding (CTE). Further, we employ clustering to discretize the continuous CTE representation and present a novel two-step level generation to leverage both these representations. We evaluate the performance of our approach in generating levels for seen and unseen games with skewed tile distributions and outperform the original tile embeddings.

Mixed-initiative Procedural Content Generation (PCG) refers to tools or systems in which a human designer works with an algorithm to produce game content. This area of research remains relatively under-explored, with the majority of mixed-initiative PCG level design systems using a common set of search-based PCG algorithms. In this paper, we introduce a mixed-initiative tool employing Exhaustive PCG (EPCG) for puzzle level design to further explore mixed-initiative PCG. We run an online human subject study in which individuals use the tool with an EPCG component turned on or off. Our analysis of the results demonstrates that, although a majority of users did not prefer the tool, it made the level design process significantly easier, and that the tool impacted the subjects' design process. This paper describes the study results and draws lessons for mixed-initiative PCG tool design.

It overlaps with computational creativity as well as procedural content generation, and has roots stretching back long before digital games research had begun in the form we know it today [9]. In [6] Cook and Smith offer a critique of the field, suggesting that the history of AGD research, at the time of writing in 2015, was primarily focused on the generation of rules for games, and limited to goal-oriented games with clear objective functions for winning. They write: This mechanics-first view on games is unnecessarily limiting, stifling the creative potential for AGD and restricting the kinds of games that can be automatically designed to ones that have well-defined, simple rule systems. More than half a decade on from the publication of this work, and most of its points still hold true of AGD research today. This is not in itself a flaw in the research being done - it is still valuable, and the field is progressing and creating many new and exciting systems [1, 11]. Yet there remains a need to expand beyond this, to create the "new kinds of play experience" that Cook and Smith talk about, to expand the horizons of AGD as a research field, and most importantly to expand the scope of how AI interacts with, improves and changes games as a creative medium.

It has been a very long time since the last blog post, but a lot has changed since then. To start it off we have two new team members coming aboard to join the Onslaught Team, Mitch will be undertaking some of the new level design and environmental assets and Ethan will be taking on Our UI as well as some level design. In the process of some new team members, we have also changed the Rendering pipeline that our game is going to using, We have swapped from unity (HDRP) to unity(URP). There is not a lot i can explain about the reasons for this other then the FPS/Rendering time for the HDRP were just to much as we would like to produce this game for consoles in the future, the problem is that the rending was our biggest over head, so by cutting certain aspects in by swamping renders gave us more control on the FPS as well as the amount of stuff we could do. As well as the performance boost it seems that the HDRP is not really set up in the community eyes, so finding tutorials and examples was almost near to impossible.

Modern games have advanced in multiple ways over the past decades. Technologies such as physical based rendering or adaptive tessellation have been deployed to great effect and help make modern games look amazing. As awesome as this may be is one more concealed yet even more important piece of technology that seems to feature less, something that used to sit as a crown jewel at the back of the box of any would-be high-quality video game. Of course, we are referring to video game artificial intelligence, or AI for short. But what is artificial intelligence? How do we know whether it is good or not?

Previous work explored blending levels from existing games to create levels for a new game that mixes properties of the original games. In this paper, we use Variational Autoencoders (VAEs) for improving upon such techniques. VAEs are artificial neural networks that learn and use latent representations of datasets to generate novel outputs. We train a VAE on level data from Super Mario Bros. and Kid Icarus, enabling it to capture the latent space spanning both games. We then use this space to generate level segments that combine properties of levels from both games. Moreover, by applying evolutionary search in the latent space, we evolve level segments satisfying specific constraints. We argue that these affordances make the VAE-based approach especially suitable for co-creative level design and compare its performance with similar generative models like the GAN and the VAE-GAN.