Example of a simple musical Markov model. Each node represents a musical note and each arrow represents a potential transition between notes. Thicker arrows have a greater probability of occurring. To run the model, select a starting point, then follow a path of arrows, with greater likelihood of following thicker arrows than thinner ones. Image credit: Oliver Bown.
A radio antenna design optimized by an evolutionary algorithm. The algorithm tries a population of different designs, testing their performance in simulation and selecting those more successful for future “breeding.” Over time, designs improve. Image credit: NASA (public domain).
One of Kate Compton’s casual creator systems, a parametrically controlled plant designer. Image credit: Kate Compton.
A series of 3D organic forms evolved by Andy Lomas. Image credit: Andy Lomas.
The E-volver generative artwork by Erwin Driessens and Maria Verstappen. Top left: The interaction context showing how people can selectively breed designs. Other panels: A range of agent behaviors. Image credit: Driessens and Verstappen.
Lehman and Stanley’s original demonstration of the effective power of novelty-based search, even for traditional goal-based search tasks. In both images, the large white dot (bottom left) is the starting point for an agent that is trying to find its way to the small white dot (top left) through the maze, within a given time limit. The black dots represent the final resting place of the agent in many different trials. In the left panel the agent search strategy was evolved using target-based evolutionary search: agents that got nearer the target were deemed fitter and more likely to seed future generations. However, the shape of the maze means that fitness might be rewarded for agents that are close to the target as the crow flies but not in reality. In the right panel, novelty search is used: fitness is rewarded to agents whose resting position is novel—that is, dissimilar to previous agents. The images show that the novelty search strategy is actually more effective at reaching the target. Image credit: Joel Lehman and Kenneth Stanley.
Paul Brown’s artworks involve the automatic recombination of elements using algorithmic processes. The design of modular elements that afford recombination to create rich and complex structures is an important part of this art. Here the work is rendered as a series of cards that a person can tile manually. In digital form, the cards arrangement is often managed by procedural or evolutionary rules. Image credit: Paul Brown.
An example of a parametric design exploring gradual variation in the positions of a hexagon’s vertices. Image credit: Andrew Kudless, Andrew Payne (LIFT Architects).
A visual relative of style transfer: high-level structures (dinosaurs) are filled out with low-level structures (plants). Image credit: Chris Rodley.
Example of images generated by Elgammal’s Creative Adversarial Network (CAN). These are the images ranked highest (most liked) by human subjects.
Petros Vrellis’s work Christ Savior by El Greco (as string-art). The work is created using a relatively simple optimization algorithm that attempts to approximate a source image using the intersection of straight lines, then rendered using threads wound across pegs. Image credit: Petros Vrellis.
Palle Dahlstedt’s MutaSynth program, which can be used to evolve the synthesizer sounds on a Nord Modular G2 synthesizer. Image credit: Palle Dahlstedt.
Ben Houge’s Food Opera is a novel application of generative music in which adaptive compositions are coupled with dishes and coordinated among the diners in a distributed fine dining audio experience. Image credit: Ben Houge.