1. 5

  2. 2

    Can you say what some of the main conclusions are? Or is this primarily a survey? E.g.,

    this article identifies game facet orchestration as the central challenge for AI-based game generation

    In particular, we identify the different creative facets of games

    Are these the six facets: audio, visuals, levels, rules, narrative and gameplay? But the intro also suggests that only music (audio?) will be looked at. Or maybe its only meant as linguistic metaphor.

    we propose how orchestration can be facilitated in a top-down or bottom-up fashion,


    we conclude by discussing the open questions and challenges ahead.

    I’m guessing this is

    • Combinatorial Explosion
    • Learning the Mapping Between Facets
    • Orchestration with Human Designers
    • Evaluating Orchestration

    envisioned several high-level orchestration processes along a spectrum between purely hierarchical top-down generation and organic bottom-up generation.

    What are some examples of these processes?

    (I’m thinking this is just my unfamiliarity with the topic but the last two sentences of the abstract are saying almost the same thing.

    I wish abstracts in general gave more information about their conclusion like key pieces of information the authors would have liked to know before starting the project. Stuff that would have helped speed things up a lot.

    1. 4

      I’d describe it as a forward-looking survey. The starting point is that there’s a large existing body of work on procedural content generation, but they’re systems that generate one kind of thing at a time, like level generators, pixel-art makers, or procedural audio. Can you get to full-game generation by just plugging these kinds of generators together in some way? We discuss different architectures for orchestrating these kinds of generators: top-down, bottom-up, pipelined, etc., and survey existing systems that have already done some version of that.

      The six facets we propose are ones you mentioned, yeah. There are many other ways you could slice game design, so this isn’t necessarily the ultimate metaphysical truth about games, that they’re made of exactly six kinds of stuff. But it’s based on looking at what kinds of things existing procedural generators currently generate (level generators are probably the most common, but there’s work in the other five too).

      Yeah, the “orchestrate”, “jam”, etc. terminology is just a musical metaphor. We don’t only focus on game music here, but we use analogies like top down orchestra-style scoring/conducting, where every element of the overall system is given its centrally written part, vs. bottom-up jazz improvisation where they coordinate less hierarchically, etc. I can see how that can get confusing, sorry.

      The survey part of the paper is in the case study section, where we give an overview of nine existing systems that generate more than one kind of thing in tandem (e.g. rules, levels, and music). We translate what each of them does to our language of 6 facets and different orchestration strategies, to try to get an idea of how all the existing stuff relates to each other, and what it doesn’t do yet. The first author (A. Liapis) made some colorful triangle facet-orchestration diagrams for that section that I quite like. They summarize what each system is doing in a kind of pictogram language, showing which facets the system generates and how they interrelate (e.g. some systems have a pipeline, some scrape content off the web, some ask for user input at certain points, etc.).

      edit: I also wrote a short twitter thread earlier today with a more concise version of this explanation, for people who like twitter threads (I know, I know).

      1. 1

        Thanks! The figures in the survey are indeed really nice. (It wasn’t obvious before reading your comment that the arrows was the information about the orchestration process.)