People tend to see artificial intelligence as a strictly technological scientific process. For those who specialize in the social sciences or humanities, artificial intelligence may seem irrelevant. This is not the case; artificial intelligence would be nothing without humanity and creativity. This is where computational creativity comes in. According to the Computational Creativity Conference Steering Committee, computational creativity is a “multidisciplinary endeavor that is located at the intersection of the fields of artificial intelligence, cognitive psychology, philosophy, and the arts.” Regardless of the discipline in which you work, computational creativity can help you understand the connection between technology and humanity. Creativity does not have a single definition; all fields implement creativity in different ways. Through exploring the power of creativity in technology, you can expand your perspective and reinvent how you approach your own field of study.
Harold Cohen: Much More than a Painter
The late abstract painter Harold Cohen was a professor for almost thirty years in the visual arts department at the University of California. During this time, he began working on Aaron, a computer program able to create art autonomously. It was purely out of curiosity that Cohen taught himself how to program; he believed that, hidden in traditional art practices, were clues to expanding the limits of artwork. Cohen started his research by studying the drawings of children, other artists, and even Native American petroglyphs.
The Culture Behind Coding: Nature of Representation
Before Cohen could create his software program, he knew that he had to understand the anthropological development of art. Cohen studied the history and nature of representation; this helped him understand what the “minimum conditions [are] under which a set of marks functions as an image.” Based on what Cohen learned about symbols and depiction, he taught Aaron how to create shapes, pick what goes in the foreground and background, and even recognize when a piece of art is complete. According to Pamela McCorduck, author of Aaron’s Code: Artificial Intelligence, Meta-Art, Artificial Intelligence, and the Work of Harold Cohen, even though a machine knows “nothing about the world, or the viewer, nor [does] it have anything to ‘communicate,’” if it is programmed to recognize cultural symbols, it can still produce art. These symbols–these representations–are what have the ability to communicate with the viewer and evoke a response. Cohen emphasized that “art… is a meaning generator, not a meaning communicator.” We don’t value art because it can communicate particular meanings. Rather, we delight in the artist’s ability to present us with something inspiring; we generate our own meanings from what an artist awakens in ourselves.
Though Cohen wished to create a piece of technology, a knowledge of code was not enough; he needed to study how images are represented throughout different cultures, and how these images affect humanity’s philosophical beliefs and psychological development. The late philosopher Nelson Goodman believed that all depictions “must be analyzed within the context of culture and learning;” all depictions are symbolic. Goodman agreed with Cohen: art is a meaning generator. A viewer finds meaning in a picture based on “past experience and knowledge”; therefore, “the visual image is not a mere representation of ‘reality’ but a symbolic system.” No two people will view an image in exactly the same way. In order to make Aaron a successful model of creative artificial intelligence, Cohen needed to program Aaron to “do some of the things humans do when they make representations.” Ultimately, Cohen realized that he “might possibly learn more about the nature of representation [through coding] than [he] had even done by painting.”
It is coding that helped Cohen understand that art is a process inseparable from culture. In his exhibition at the La Jolla Museum of Contemporary Art, Cohen’s computer completed one drawing a day for twenty-six days. Cohen commented that this exhibition was about the “whole process… not just the drawings.” He was interested in the computer’s “ability to build a primitive decision-making function… into complex functions bearing strange and remarkable resemblances to human logical processes.” It was the “thinking” that went into the computer’s procedure that was the most important to Cohen.
Artificial Brains: Neural Networks and Machine Learning
In Cohen’s studying of culture, he inevitably turned to the differences between a computer and a human brain. Cohen did not want his viewers to anthropomorphize his machine’s activities. It is important to view Aaron as a computer; this allows the viewer to learn about how the technology operates, rather than grouping artificial intelligence artists with human artists. If we learn to see the computer as a “general-purpose manipulator of symbols,” we can compare its functionality with that of the human brain. As we gain insight into the workings of artificial brains, in turn, we can “formulate an algorithmic perspective on creative behavior in humans.”
The technique called “neural networks” mimics the structure of the human brain; this allows machines to learn independently. If the code is written with an understanding of culture and the workings of the human brain, a computer can learn through experience and association. Computer scientist Pedro Domingos states that an artificial neuron “is to a real neuron a bit like an airplane is to a bird. At a certain level of detail, they’re very different, but… they do the same job, they both fly.” Though an artificial neuron is not made out of the same material as a real neuron, they both have the same primary function: just like a human brain, an artificial brain can learn from experience. This ultimately allows a computer to create its own style when applied to art.
Today, projects such as SpiNNaker use the technology behind artificial brains in order to mimic the structure and behavior of human brains. SpiNNaker, an acronym derived from “Spiking Neural Network Architecture,” is a supercomputer that is neuromorphic; it contains electronic circuits that mimic the neuro-biological structure of the nervous system. This machine can create large, biologically realistic models of spiking neural networks. Rather than apply one single algorithm, as Cohen did when he coded Aaron using Lisp, SpiNNaker is expected to be a platform “on which different algorithms can be tested.” Different kinds of neural networks can be created. In turn, each network will create different types of neurons and connectivity patterns.
Neural networks use what is termed “machine learning”– a subfield of computer science in which the computer can learn without explicit programming–in order to allow artificial intelligence to recognize patterns and objects. The computer can then act based on what algorithm is produced. Deep learning, a branch of machine learning, uses algorithms modeled off of biological neural networks. Like in the human brain, artificial neurons create neural networks that exchange messages. Neurons are assigned different tasks; some interpret basic features, while others interpret overall shapes.
Cohen began his work on Aaron in 1968, more than ten years before machine learning became widespread. Cohen chose to “train” Aaron himself, rather than to use more independent machine learning. Instead of prioritizing an automated process, Cohen opted to tweak– or adjust–Aaron’s AI, given extensive prior data collection and analysis. In order to teach Aaron, Cohen analyzed data himself and created a complex set of rules. These rules ultimately help Aaron replicate specific behaviors. Even though Cohen did not use machine learning as it is known today, he understood how patterns of neural firings can be manipulated in order to produce a certain behavior. Through analyzing data and teaching Aaron, he found his own way to learn about neural networks and the exchange of messages.
Cohen’s “Collaboration With My Other Self”
As Cohen created Aaron, he learned about his own psychology. According to Pamela McCorduck, Cohen once stated: “‘I give the machine its identity. It does what I have in mind.’” The computer’s knowledge is extracted from Cohen’s mind. As Cohen coded his piece of creative artificial intelligence, he learned about his own mind; a different part of him shines through in the actions of Aaron. Indicative of this, Cohen entitled one of his exhibits “Collaboration With My Other Self.” His clever programming created Aaron; in turn, Aaron “communicated” back with him through the artwork, which Cohen finished up by painting. Over the years of what Cohen called a “mutual” relationship with his “other half,” he and Aaron created both realist and abstract paintings. In Cohen’s later years, he pushed Aaron to become even more creative; he wanted Aaron to learn how to “reformulate [his] own ‘mental model of the world.’” Cohen hoped that one day, creative artificial intelligence would be able to take on the full responsibility of building images.
Though Cohen made a name for himself as a trailblazer of computer-generated art, he always seemed to want to push Aaron–and himself– further. The success behind Aaron lies in Cohen’s understanding of creativity and artistry. Creativity is essential in the creation of any technology. If software engineers wish to create successful models of computational creativity, it would be wise for them to acknowledge how powerful technology can become if the engineer has an understanding of humanity and creativity.