Conceptual Game Expansion

Abstract

Automated game design is the problem of automatically producing games through computational processes. Traditionally these methods have relied on the authoring of search spaces by a designer, defining the space of all possible games for the system to author. In this paper we instead learn representations of existing games and use these to approximate a search space of novel games. In a human subject study we demonstrate that these novel games are indistinguishable from human games for certain measures.

Towards Game Design via Creative Machine Learning (GDCML)

“…Thus, imbuing ML models with combinational creativity techniques such as conceptual blending [91], amalgamation [92], compositional adaptation [93] and conceptual expansion [50] could enable tools to assist in such creative forms of game design and generation. Conceptual expansion has in fact been demonstrated to be able to generate entirely new games that combine the levels and mechanics of existing games [94].…”

Learning latent representations for controllable combinational creativity and game design

“…While prior works have formulated combinational creativity processes via linear combinations [75,74], to our knowledge we are the first to do so as a linear combination of specifically latent conceptual spaces, as learned by deep latent variable models. A conceptual space is the generalized version of the input, generic and blend spaces from the four-space theory of conceptual blending described previously, with each of these being instances of conceptual spaces.…”

“…The work presented in this thesis falls within a subset of more creative PCGML [76] that has emerged in recent years, with a focus on learning, reasoning about and leveraging design spaces spanning a number of games rather than learning distributions of individual games. This has yielded applications such as generating new games by recombining learned game graphs [74], learning affordance-based tile embeddings across multiple games [92], domain adaptation and transfer [169,172] as well as some of the level and game blending work we present in this thesis. More recently, these works, as well as several of the ones we will present as part of this thesis, have been grouped together as methods for PCG via knowledge transformation [154], as they generate content for a new game by extracting and transforming knowledge in the form of content from another.…”

“…Thus, imbuing ML models with combinational creativity techniques such as conceptual blending [51], amalgamation [125], compositional adaptation [192] and conceptual expansion [73] could enable tools to assist in creative forms of game design and generation. Conceptual expansion has in fact been demonstrated to be able to generate entirely new games that combine the levels and mechanics of existing games [74].…”