Frequently Asked Questions About AI Art
What is Artificial Intelligence (AI) artwork?
Artificial Intelligence artwork refers to art generated with the assistance of AI. Put simply, AI is a field of computer science focused on creating machines that simulate human cognition and learning. AI generated art can range from autonomous creations like the work of Ahmed Elgammal to improvisational partners like Sougwen Chung, both of which you can explore above. AI artists vary widely in their mediums, from the artificial intelligence paintings of Pindar Van Arman to the AI created music of Taryn Southern.
What is Google Deep Dream?
Google Deep Dream is a program invented by Alex Mordvintsev, an engineer and researcher at Google. Deep Dream uses computer vision and a convolutional neural network to find and enhance patterns in images. The results are often psychedelic, dreamlike enhancements to an image that are now an entire subgenre of computer generated art. You can explore work that uses Google Deep Dream from Alex Mordvintsev, Mike Tyka and Daniel Ambrosi in the gallery above.
What is machine learning?
Machine learning refers to computer systems that can learn from a data set to get better and better at a specific task. For example, a data set of images of monkeys can help a machine learn what a monkey looks like. Then when faced with a photo it’s never seen before, the machine can draw upon its understanding of what a monkey looks like – which it extracts as elements called features – and accurately predict whether the picture contains a monkey. Although this is a simple example, its applications are extremely powerful, and the basis of everything from self driving cars to ecommerce product recommendations.
What is reinforcement learning?
This refers to a learning method where machines get smarter through a system of rewards and punishments, just like humans. They usually start off naive, and end up improving over time the more they trigger rewards and punishments.
What is embodied AI?
This refers to artificial intelligence that is able to control some kind of physical form, like a body, robot arm, etc. For example, the robots that paint with Sougwen Chung are considered embodied AI.
What is the Turing Test?
This refers to a test of a machine’s ability to act like a human – such that a person can’t tell the difference between the machine and normal human behavior. It was created by Alan Turing in 1950.
What is data science?
This refers to the field of study that uses math to cull meaning and insight out of data. It uses statistics, data mining, machine learning, and other disciplines to analyze features in a data set, with the goal of providing some kind of useful information. With so much data being recorded today, it is becoming one of the most popular and rapidly growing fields in schools and in businesses. Data science and artificial intelligence are inextricably linked, and AI Art is only possible because of a foundation laid by data science.
What is a neural network?
A neural network is an algorithm that processes information by attempting to mimic processes in the human brain. Neural networks include layers of connected “neurons” or nodes that send information to each other. This is a different approach to computation than traditional computer science algorithms, where programmers explicitly directed what a program should do. Instead, neural networks learn on their own without step-by-step instructions. A common type is a Convolutional Neural Network (CNN), which is often used to identify and analyze images.
What is deep learning?
Deep Learning refers to neural networks with multiple layers of connected neurons. It’s the many layers that make a network “deep” vs. shallow.
What is an algorithm?
This refers to a sequence of step-by-step instructions. It can be as simple as the directions “turn left at the stop sign, go straight 100 feet, then enter the store.” Or it can be as complicated as a million lines of code sifting through the entire web to display relevant search results in Google. These instructions are what tell a computer what to do, how to solve problems, how to make calculations, how to display things on a screen, etc.
What is a Generative Adversarial Network (GAN)?
This refers to two neural networks that “compete” with each other: one tries to create output based on a training set (e.g. based on what it’s “learned”), and one tries to identify when that output deviates too far from the original training set. In this way, the creator network gets better and better at generating look-alikes by trying to fool the discriminator network. Together they drive the output toward increasingly accurate output (matching the style of the training set). This is used by artificial intelligence artists like Robbie Barrat to create artworks based on other existing works, either by other artists, by the artist themself, or on datasets that are not necessarily art (e.g. Google’s open image data set, or public Flickr photos).
What is a Creative Adversarial Network (CAN)?
This refers to a modified version of a GAN, where novelty and randomness is built into the output of creator neural network. This enforces the creation of new and often surprising outputs, attempting to mimic natural human creativity and breaking from existing forms and patterns. The output can also be trained to know what kind of novelty is good or bad (often too much randomness makes the output aesthetically uninteresting) so that its “new ideas” fall generally within the realm of what people will find attractive. Dr. Ahmed Elgammal is one of the leading pioneers in this research area.