Neural Networks & Art: The Roles of Logic and Intuition in Art-Making


This past year brought an influx of AI breakthroughs, thawing an ‘AI winter’ and provoking equally optimistic and antagonistic reactions. Art sits squarely at the epicenter of this disruption, demanding discourse about the impacts of AI on art and clarification of the terms used to distinguish between what is human or machine made.

Nested within this intense struggle to understand and react to AI’s progress lies a unique opportunity to peel back the layers of assumption and observe how AI reveals aspects of human nature. Machines possess the ability to teach us a great deal about our own creative processes, but especially about the undefeated role that intuition plays. Let us tease some of these ideas out a little more, focusing first on understanding how neural networks mimic the human brain.

In 1955 John McCarthy defined AI as “the science and engineering of making intelligent machines” [1]. As AI evolved, several formats emerged including machine learning. Machine learning essentially consists of computers improving “their perception, knowledge, thinking, or actions based on experience or data” [2]. Branching directly from this are neural networks. “Modeled loosely on the human brain, a neural net consists of thousands or even millions of simple processing nodes that are densely interconnected” [3]. By initiating these patterns of learning, a human programmer can enable a machine to function with greater independence and efficiency.

Within the past year or so, one of the greatest breakthroughs came when text-to-image AI became widely accessible. This brought along a mixed bag of results and reactions, especially amongst artists. As the senior editor for AI at MIT Technology Review states, “Artists are caught in the middle of one of the biggest upheavals in a decade. And, just like language models, text-to-image generators can amplify the biased and toxic associations buried in training data scraped from the internet” [4]. The ethical concerns surrounding AI deserve everyone’s attention. However, for this particular discussion I would like to focus on what neural networks have demonstrated about the human brain and how that clarifies the merits of human made art.

Quite simply, neural networks function as reflections of the human mind. This has proven to be, “a valuable tool for neuroscientific research. For instance, particular network layouts or rules for adjusting weights and thresholds have reproduced observed features of human neuroanatomy and cognition, an indication that they capture something about how the brain processes information” [5]. If we break this down a bit further, we can begin to observe these mirrored patterns of thinking between neural networks and humans.

When we are exposed to information, especially at a young age, we learn to discriminate between what is or is not something. These delineations can be quite fuzzy at first, but with time we learn to define things more clearly and even to intuit ideas based on related concepts. As children, most people learn to draw as a method of communicating this knowledge. Our attempts to recreate ideas frequently take the form of symbolic gestures, such as stick figures. But at a very young age the ability to communicate these distinctions is still not particularly sophisticated and so we are content with a series of scribbles that we may proclaim represents a “dog” or “mom.”

As both our motor skills and our ability to perceive advance, we develop the ability to more effectively criticize our own output, ironically sometimes at the expense of our desire to make it. So how do machines relate to this? Well, they are in many ways a microcosm of our own patterns of thinking and development. In the case of a generative adversarial network there is a portion of the system dedicated to producing content and another to discriminating between what is or is not the desired output. A machine with sufficient initial training can teach itself to differentiate between images and then turn around and teach its partner system to produce output until it is able to trick itself into believing that its own output is ‘the real thing.’

This is where neural networks begin to differ from their human counterparts, however. Machines can be faster at brute force learning and iterating, but they are also slower and even incapable of adapting to nuance. All input for a machine must be manually assigned and defined whereas humans are exposed to an overwhelming amount of information all the time and through various senses. Once processed by our neural systems, we store this input within elaborate structures of related information.

We draw on these stores of information through nearly inscrutable links, drawing conclusions that perhaps could never be reached by pure logic. This wholly unpredictable process is based in intuition, driven by instinct and the essential need to act on incomplete knowledge. These impulses are essential for growth and innovation and play a basic, yet essential, role in survival. Machines, as they are now, do not possess the drive to survive or the ability to act on intuition - but we do.

Arguably, this remarkable combination of intuition and self-awareness sits squarely at the heart of creativity. While a neural network can perform highly complex tasks, our limitations and physical needs demand that we constantly rely on incomplete knowledge in a theoretically infinite universe. Art is born of this need to make sense of the world and the invisible threads of self-awareness that compel us to create.


Contributors

Rebecca Mott - An artist and alum of the School of the Art Institute of Chicago (MFA, 2022), Rebecca is fascinated by the intersection of artificial intelligence and painting and explores this tension through her writing and artistic practice. For more information, you can visit her website.

Sources

[1] Manning, Christopher. “Artificial Intelligence Definitions.” Stanford University Human-Centered Artificial Intelligence: September, 2020. https://hai.stanford.edu/sites/default/files/2020-09/AI-Definitions-HAI.pdf?sf133061845=1.

[2] Manning, Christopher. “Artificial Intelligence Definitions.” Stanford University Human-Centered Artificial Intelligence: September, 2020. https://hai.stanford.edu/sites/default/files/2020-09/AI-Definitions-HAI.pdf?sf133061845=1.

[3] Hardesty, Larry. “Explained: Neural Networks.” MIT News, April 14, 2017. https://news.mit.edu/2017/explained-neural-networks-deep-learning-0414.

[4] Heaven, Will Douglas. “AI That Makes Images: 10 Breakthrough Technologies 2023.” MIT Technology Review, January 9, 2023. https://www.technologyreview.com/2023/01/09/1064864/image-making-ai-10-breakthrough-technologies-2023/.

[5] Hardesty, Larry. “Explained: Neural Networks.” MIT News, April 14, 2017. https://news.mit.edu/2017/explained-neural-networks-deep-learning-0414.