Here at Sentient, evolutionary computation is the science that underlies many of our products, including Sentient Ascend, our AI-powered conversion rate optimization platform. As a sequel to our article on Intro to Evolutionary Algorithms, we thought we would demonstrate other applications of evolutionary computation, especially interactive evolutionary design—the result of using evolutionary computation to create art, music, design, and games.
What is Interactive Evolutionary Design?
Interactive evolutionary design characterizes a kind of algorithm that changes and improves a design based on user feedback. The way it works is that a user interacts with the system and chooses which of the designs he or she likes, and then the system breeds the “winners” together and evolves a new set of designs based on the user’s choices. Eventually, with enough “breeding” of designs, the final design will be personalized to the user’s tastes.
Evolutionary Design in Art
Art is thought of as a solely human enterprise with the assumption that creativity is a solely human trait. Yet, strides in evolutionary design have demonstrated that computers too can be artists. The use of evolutionary design in art can be seen as early as 1986, when Richard Dawkins presented his computer simulation of “biomorphs” bred from genes. These biomorphs have appearances similar to plants and animals, and these appearances emerge entirely from combinations of these basic genes. The term “biomorph” was adopted from the artist Desmond Morris, who used this term to describe the animal-like figures in his late surrealist paintings. To get a taste of the beauty of biomorphs, an interactive simulation can be found on the Mount Improbable website.
An evolutionary tree of biomorphs rendered using Mount Improbable.
Electric Sheep was also a notable leap in the application of evolutionary design to art in the early days of the internet. Users with this program installed on their computers received fractal animation screensavers, dubbed “sheeps,” that they could vote on. The amount of votes a sheep received determined its lifetime in the system, and so users could directly evaluate the subjective fitness of these sheeps. The best sheep then reproduced based on a genetic algorithm, with the hopes of creating a new generation of better sheep—evolved sheep, if you will.
An example of the kinds of fractals that Electric Sheep produces.
Since these earlier days of evolutionary design in art, artists like Karl Sims have taken these principles and created large collections of art generated with evolutionary design. Sims has taken this one step further, bringing evolutionary design to the real world with the Reaction-Diffusion Media Wall at the Museum of Science in Boston.
Even 3D digital art is fueled by evolutionary design. A good example can be seen on the site Endless Forms, which shifts and shapes 3D art forms via evolutionary algorithms based on the user’s interests. The way it works is that users select which shapes and designs they like the most and then request to “breed” them together so it evolves into something new and interesting. Eventually, this process personalizes the structure to the user’s preference. You can even order your designs to be 3D printed and sent to you in real life!
Evolutionary design is even playing a role in evolving music. DarwinTunes for instance was an experiment that examined how consumers could shape the music that they listen to based on their choices. The program itself used an evolutionary algorithm to generate sets of music from simple loops of sounds. Humans then listened to the generated music sets and rated them from 0 to 5. Once a participant had rated 20 pieces of music, the algorithm selected the 10 best songs and “mated” them together to produce 20 new pieces of music. This process then continued for another 100 generations until the music was judged to be pleasing and had hit a point where it could not longer evolve to be even more pleasant than it already was.
This may sound familiar if you’ve read about how Spotify and Pandora are using algorithms to predict what type of songs or albums you like. For DarwinTunes, however, it wasn’t predicting which song to recommend next. It was producing a completely new piece of music made from evolving the best rated songs together. It’s as if your favorite songs could mate and have an even better sounding baby.
Evolutionary algorithms can even be used in video games. Take, for instance, the classic Super Mario World game, which does not use evolutionary algorithms itself, but due to its simplicity, allows players to create and implement neural networks that can be used to beat several levels of the game on their own. YouTube user SethBling took this idea one step further and used evolutionary algorithms to train a neural network to play the game, proving that computers can dominate in the world of gaming. And in a surprising coincidence—SethBling’s algorithm was influenced by a paper of our very own CTO and Head of Research, Risto Miikkulainen!
Another game called Petalz, available on Facebook, uses AI to allow users to select and evolve flower DNA to create a variety of vibrant digital flowers that includes purebreds and crossbreeds. Players can then share their flowers with friends on Facebook and then choose to breed their flowers with that of their friends, creating a kind of digital genetic pool for their flowers.
These games are quite simple and fun, but what about more hardcore games? Quake is an addictive, highly competitive, first-person shooter developed by id Software. While the game contains bots that the player can train against, these bots use naive algorithms instead of AI to battle against the player. Luckily, with the open-source nature of Quake 3, Joost Westra was able to design his own bot that uses evolutionary algorithms to train the neural net that controls his bot. The result was a custom-made bot that vastly outperforms the original Quake 3 bot. Advances in evolutionary algorithms in games are offering players more challenging and exciting opponents to play against.
Evolutionary Design for Website Testing
With the breadth of applications of evolutionary design and evolutionary computation in the arts alone, it is no wonder that these methods have also been adapted to the modern world of business—particularly in the realm of ecommerce and website optimization. Sentient Ascend for instance uses evolutionary algorithms to evolve the best performing websites together to find the winning website version that is optimized for the highest conversions. Websites, like several of the other digital art forms mentioned in this article, have their own set of “genes” or elements that can be mutated, combined, eliminated, or exploited in similar ways as human genomes. Evolutionary algorithms in a sense, are the forces of nature that cause these website genetics to evolve.
For more on the evolution of websites check out our white paper Website Natural Selection.