Aesthetics and creativity in evolutionary design

Leave a comment
aesthetics, evolutionary design, GA, research, rhinoscript, writing

Institution: TU Delft

Researchers: Aurelie Hiao, Roxana Palfi, Agata Kycia

Year: 2010

This research project focuses on the issues of subjectivity in evolutionary design processes, exploring its potential and constraints in terms of aesthetics and creativity. It provokes questions about what should be the right balance between the role of a designer and the role of a computer in the design process, in order to reflect designer’s aesthetical preferences on the one hand but also allow for novelty and surprise on the other hand.

Principles of evolution have long been used in order to develop optimization tools, what has become one of the main branches of evolutionary computation. Evolutionary design that aims at optimization, usually deals with well-defined problems- those, that can be codified as objective, more or less complex fitness functions. The challenge appears when the problems, which are to be solved using evolutionary computation methods, are problems of subjective nature, as for instance aesthetics. How can we codify aesthetics? How can we implement subjectivity into automated, computerized design process? Moreover, how can we integrate the designer’s aesthetical preferences into the design process in such a way, that it does not limit the range of possible solutions, but still allows for innovative, surprising results beyond human imagination?

In order to find answers for some of those questions, we did  several experiments that implement different degrees of subjectivity into the design process. Since we were interested in investigating the possible applications of evolutionary algorithms in the field of architecture, the subject of our exploration had to be a 3-dimentional form, so that we could evaluate its spatial qualities, even if very simple.

The initial idea was to compare three different experiments, one being a simple genetic algorithm, the seconed one being a genetic algorithm with human selection, and the third one, the most complex one with the machine-learning process that would learn aesthetics based on the choices that were made. However those initial assumptions have changed a bit during the process and we’ve made only the first 2 experiments.

The first challenge was to define such a rule to generate population of shapes, that would always create closed figures (in order to be able to evaluate their spatial/architectural qualities, to be able to talk about the interior, exterior and volume), but in the same time to allow for variation and differentiation.Starting from a cube, we came up with a solution that would create a shape made out of 12 flat triangular based on 8 vertices placed within that cube.

In order to create next generations, the crossover function and a very basic objective fitness function – the biggest volume, were added. However the key concept of our research, that deals with aesthetics, was implemented into the algorithm by introducing human selection.

In every generation 3 shapes were chosen as a base for further generations. We also experimented with the ratio of the solutions chosed by the computer and by the user (so that the user could choose either 1, 2 out of 3 shapes).Obviously, the more choices were made by the user the more similiar the result was to them. However it has to be taken into account that all the shapes were very basic, what makes it difficult to talk about more complex aesthetic features. Moreover, it is hard to judge the novelty since the codified rules remain also very simple.

(left pic) – 2 choices are made by the computer and 1 by the user, (right pic) – 1 choice made by the computer and 2 by the user

Another important aspect is the tiring factor, inseparable from human selection. If the user has to choose even 2 shapes in every generation, he/she becomes very soon bored and tired, unable to pay the same attention to the process as in the beginning. Therefore automation of the selection process seems to be very useful.If the user has to choose only every 20th generation and the other choices makes for him the computer, the algorithm would run faster and much more effectively.

Rules driving design of this research project remain very simple and therefore are limiting the final outcome. Search fot the optimal balance between what can be done by the computer and what by the designer remain still an area for exploration, which parts of the design process benefit most from being automated, and which from introducing the unique human factor.

More about the project you can read in the paper Aesthetics and creativity in evolutionary design.

For more information in the subject-matter of aesthetics and creativity in evolutionary design, you can access the research paper below: AESTHETICS AND CREATIVITY IN EVOLUTIONARY DESIGN

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s