In lieu of an abstract, here is a brief excerpt of the content:

  • Picbreeder:Collaborative Interactive Evolution of Images
  • Jimmy Secretan, Nicholas Beato, David B. D'Ambrosio, Adelein Rodriguez, Adam Campbell, and Kenneth O. Stanley

Picbreeder [1] is a new website that is open to the public for collaborative interactive evolution of images. A unique feature of Picbreeder is that users can continue evolving other users' images by branching. The continual process of evolving and branching means that images can continue to improve and increase in complexity indefinitely, yielding a proliferation of artistic novelty that requires no explicit artistic talent to produce.

Interactive Evolutionary Computation

Picbreeder borrows ideas from Evolutionary Computation (EC), which allows computers to produce a myriad of digital artifacts, from circuit designs to neural networks, by emulating the process of natural selection. In EC, a population of individuals is evaluated for fitness and mutated and/or mated, to produce the next generation. This cycle continues until evolution produces an individual considered significant. Interactive Evolutionary Computation (IEC), originally explored by Dawkins [2], is a type of EC in which a human evaluates the fitness of individuals (see Takagi [3] for a review). This technique is particularly effective at evolving artifacts that are too subjective for the computer to evaluate itself, including artwork, music, and designs.

Picbreeder uses a specialized evolutionary algorithm called Compositional Pattern Producing Networks-based NeuroEvolution of Augmenting Topologies (CPPN-NEAT) [4], which is extended from the original NEAT algorithm [5]. A property unique to NEAT is its ability to complexify, which means that, as evolution progresses, the individuals become more complex. Therefore, the potential complexity and sophistication of images evolved in Picbreeder is unbounded. The genetic art created by this process thus tends to look organic and often resembles familiar objects (Fig. 1).

The Picbreeder client program presents the user with several grayscale images (Fig. 2a). The user selects which images to reproduce. The program then performs crossover and mutation on the chosen images to produce the next generation. As this process iterates, images increasingly satisfy the users' desires. Interactive evolution finally terminates when the user is satisfied.

Picbreeder Principles

A primary aim of Picbreeder is to allow a large group of unskilled individuals to collaborate to produce interesting output. Good taste, rather than explicit artistic talent, powers the process.

This goal faces several challenges. For example, in some other online, interactive evolution communities (e.g. in [6]), input from a large number of users produces an averaging effect, wherein the votes of individual users are washed out by the masses. In addition, in IEC, producing a significant output can take a long time, in many cases longer than the user is willing to continue the process. If, for example, 1,000 one-minute generations are required to produce an interesting image, it would take a single user almost 17 hours to complete. While the output may be significant, such an investment is typically impractical.

Picbreeder solves these problems by allowing users to branch from the works of other users. Once a user has created an interesting image through the IEC client, he or she then publishes the image, making it visible to the rest of the community. Any other user that discovers this image on the website can subsequently branch from it, thereby continuing evolution with his or her own copy of the image. Because of this mechanism, a chain of successive branches can be evolved by many users through thousands of generations, potentially producing otherwise unreachable images. While branching allows many users to contribute to the final image, each can take artistic liberty in the part of the series that they own. Furthermore, branching can produce a diversity of lineages originating from a single source.

While branching is an effective mechanism for IEC, it cannot work properly unless users can find appropriate images from other users. To make finding good images easy, users can rate the images of other users and thereby promote images that are generally interesting. However, Picbreeder protects nascent images by showing the newest in addition to the most highly rated (Fig. 2b). This policy gives the newest images a chance to be seen and rated by users who might be interested. Finally, because often "beauty is in the eye of the beholder...

pdf

Share