European Commission logo
English English
CORDIS - EU research results

Autonomous Computational Game Designers - Transforming Exploration via Deep Learning, Novelty Search and Emotive Modelling

Final Report Summary - AUTOGAMEDESIGN (Autonomous Computational Game Designers - Transforming Exploration via Deep Learning, Novelty Search and Emotive Modelling)

• Summary description of the project objectives

AutoGameDesign aims to uniquely study the relationship between advanced algorithmic search and creativity, focusing on a) how exploration can be transformed for the generation of novel and valuable outputs; b) how surprise can be modelled computationally; c) how surprise representations can drive the search for and evaluate creativity, and; d) how this can be transferred to computational creators that generate complete pieces of rich and multifaceted artwork such as computer games.

• Description of the work performed

The project aims to answer the question of whether autonomous creative systems are able to generate valuable, novel and surprising outcomes within games, thereby realising computational game creativity. It explores how computational creators can be equipped with transformed exploratory creativity for the generation of novel and valuable outputs and how surprise can be modelled computationally and drive the search for and evaluate creativity. Advanced machine learning methods have been fused with computational predictors of emotive creativity yielding entirely new ways of searching within the creative search space. The computational creators’ output has been primarily evaluated within the domain of game design being among the most content-intensive and multifaceted domains of human creativity and, undoubtedly, the richest form of human-computer interaction. The project’s aim is the development of autonomous computational creators that will advance the measurable capability of computers to produce results that are self-assessed, and assessed by humans, as useful, original and surprising.

The work performed under the AutoGameDesign project during its four year period focused on investigations of surprise search and constrained divergent search as mechanisms for the generation of novel, surprising yet valuable outcomes. The project also introduced the concept of multifaceted computational game creativity and offered numerous applications and technical demonstrators of computational designers that automatically generate parts of game content such as sounds, visuals and levels.

• Main results

The main outcomes the project achieved include:

Core scientific contributions
• Introduction of computational game creativity and multifaceted orchestration for game generation.
• Introduction of the notion of surprise and the general surprise search algorithm for unconventional discovery (problem solving and generative art).
• Multiple game demonstrators showcasing the creative capacity of developed AutoGameDesign generators


Project outcomes have been included in 34 publications in top-tier journals and prestigious conferences in the areas of computational creativity, evolutionary computation and game artificial intelligence. Further details can be found on the project’s webpage.


Two publications of the AutoGameDesign project won best paper awards. The papers were presented in prestigious conferences in the areas of game artificial intelligence and evolutionary computation: the IEEE Conference on Computational Intelligence and Games and the Applications of Evolutionary Computation conference.

Outreach activities:

Several popular science magazine articles and public events for disseminating results of the project. Further details can be found on the project’s webpage.

• Impact (including the socio-economic impact and the wider societal implications)

The scientific and technological advancements that AutoGameDesign brought are obvious: While generative systems existed since the birth of computational creativity never before surprise has been modelled as a self-evaluation mechanism, the relationship between user experience and creativity has never been investigated and machine learning has never been coupled with novelty and surprise search for that purpose. The collective impact of the proposed methods for computational creativity, game AI and game design clearly advances the state of the art in those fields.

Game industry is a key creative industry, eclipsing the global turnover of the music and film industry with a total size estimated around 76 billion USD annually, projected at over 80 billion by 2016. Digital game and downloadable content sales are growing at a rate of 33% per year in the US and EU while the US market for digital game content alone is over 5.9 billion USD, with countries such as France, UK and Germany accounting for more than 1 billion USD each. At the same time, the growth of content-intensive applications like games, the ever-growing need of manual content creation, the increasing complexity and scope of software applications, and the labour-intensive design and production of content by hand (within and beyond games) inflates both development time and cost. Not only does the generation of novel, useful and surprising content by algorithmic means introduced by AutoGameDesign circumvents this content bottleneck, it also allows for faster design iterations, increases design efforts and pushes the limits of human creativity.

• Project website and presence

All details regarding the project’s progress, the project’s logo and visual identity, core dissemination events, published papers, and milestone demos and videos can be found at the project’s webpage: Further details and regular posts about the project’s findings and activities can also be followed through the project’s Facebook page: