Mechanized Aesthetics: The Mass Production of Art and the Transformation of Creative Identity in the Age of Artificial Intelligence

Abstract

Prompt engineering has ushered in a new era of creativity, enabling the emergence of thousands of new artists who utilize artificial intelligence as their primary medium. However, the public often debates the legitimacy of these creators, with some questioning whether they can be considered true artists in the traditional sense. This work aims to delve into the complex and evolving definition of art, examining how human perspectives influence what is classified as art and what falls outside its boundaries. Additionally, it seeks to shed light on the growing opposition to AI as a creative tool, exploring the fears and concerns expressed by many about the potential threats large language models pose to freelance artists and other creative professionals. Through an in-depth analysis, this work will evaluate whether these concerns are justified or stem from misconceptions about the role of AI in the artistic landscape, ultimately aiming to contribute to the broader discourse on the future of art in the age of technology.

Introduction

Over the past two decades, there have been LLMs coming out that allow for users to utilize AI through prompt engineering. Models like OpenAI's Dall-E, Stable Diffusion, and Adobe Firefly all boast impressive results in the modern age. However, is there actually a true consensus over if the images they produce are considered art? Does the LLM know what art truly is past a surface-level definition of the concept? Society has been given definition of art that has been changed and remolded time and time again—evolving from historical works done by renowned artists like Michelangelo and DaVinci to conceptual artwork of a banana taped to a white wall with duct tape that was sold at $6.2 million in November of 2024. Recently, artists have been fighting against copyright infringement done by LLM's utilizing their works for image generation with using tools like Glaze—a tool for digitally altering images in a minor way that appears undetectable by the human eye but distorted and abstract to machine learning. Because of this, the question of whether or not artificial intelligence is an enemy, or a friend, has been a hot topic of debate for many as it seems new startups and LLMs are being produced every day. Do the positives that AI brings truly outweigh the negatives? And is it worth it to continue work on AI if innovation leads to thousands of creative casualties in its wake? These are the questions that society and innovators must ponder as we continue to evolve past what we once believed was our peak. 

Are Humans Necessary for the Creation of Art?

In 1973, an artist named Harold Cohen created a program that was capable of painting and drawing. AARON was created in Stanford's Artificial Intelligence lab, created in FORTRAN. Cohen presented an initial prototype of this system in 1971, leading to the numerous works. AARON, unlike the large language models are we accustomed to today, "stored the knowledge of a human expert reformatted into a complex set of rules intended to simulate human decision-making." Cohen stated his motivation was to test whether or not a machine may be capable of "human art-making behavior."⁷ Though met with critics claiming Cohen was the real author of the works created by AARON, Cohen stood by his creation stating, "AARON drew so much better than [I] did," giving up his practice. Cohen had many moments of reflection on the human relationship to the practice of creating art, stating in his essay "Parallel to Perception: Some Notes on the Problem of Machine-Generated Art," "If a photographer takes a picture, we do not say that the picture has been made by the camera. If, on the other hand, a man writes a chess-playing program for a computer, and then loses to it, we do not consider it unreasonable to say that he has been beaten by the computer."¹ Cohen argued that the behavior exhibited by machines when creating art was not parallel to the behavior that humans exhibit when creating art, stating that actions required feedback dependent on the result.

How Artificial Intelligence Creates Images

These image generation models utilize large amounts of data to synthesize realistic images—with text-to-image in particular posing a significant semantic gap between the text domain and image domain, producing presumed to be accurate image outputs from user-generated prompts. Data for these models is gathered from various places, with ImageNet being one of the largest—consisting of more than 1,034,908 images of the human body and 14,197,122 annotated images total in 2024.³ There is also ShapeNet, complete with over 3,000,000 3D CAD models and 222,000 sorted models in 3,135 categories. ImageNet, which is built upon a structure by WordNet, assigns images to the branches created by WordNet. An example of this would be that mammal is to placental as placental is to carnivore. Carnivore is to canine as canine is to working dog, etc.²

Artificial Art and The Public

In a 2023 study, it was found that there was a negative bias against "Al art" after an image was revealed to them to be created by artificial intelligence. Results revealed that, "although it was difficult for individuals to identify Al-generated artwork, they exhibited an implicit prejudice against Al art." Human made images were detected with a 68% accuracy and Al-made paintings were detected with only a 43% accuracy.⁸ Participants in the study did not consider Al-made paintings, "less beautiful, likable, or pleasing." Also in 2023, a study titled, Defending humankind: Anthropocentric bias in the appreciation of Al art, found that, "humans have a strong bias against Al-made artworks, which are perceived as less creative and induce less sense of awe compared to human-made art and that the bias is stronger among individuals with stronger anthropocentric beliefs."⁶ In a 2024 study analyzing the intersection between personality traits and sentiments toward Al art, it was also found that the more empathetic a person was evaluated to be, the better they were able to detect Al-generated images. Participants in general struggled to consistently distinguish between human-made and generated artworks and, despite this, there was still a general preference for Al-generated artworks over human-made ones.⁶

Conclusion

Artificial Intelligence does not necessarily produce anything new. It combs hundreds of thousands of vectors and images to produce an output that could be perceived as favorable to hundreds of thousands of users. One may argue that this is akin to what humans do—combining their knowledge of aesthetics with their knowledge of the craft they are working on, potentially inspired by their peers or these LLMs themselves. However, much like a dog pressing on a camera shutter, a work of art created by an LLM cannot be copyrighted by that LLM. The output cannot even be copyrighted by its prompter. Art creates and invokes a sense of connection and ownership, and there is a reason that people travel hundreds of miles to gaze upon works in person in museums. Though artificial intelligence could illicit a similar feeling through the work, currently, artificial intelligence requires the human touch, selection, and emotional draw for the label of "art" to ring true. Art is everywhere in our world; it's in the design of the coffee mug you enjoy your morning brew in, it's in the logo of your favorite snack, and in the design of the building you may be viewing this poster in. To be an artist is to recognize the beauty that makes up the world and to be able to make the argument that something is or is not art. Large Language Models, I believe, are not currently at that point. The concept of art is something that remains incredibly human and one that will continue to evolve as we see more advances in the field in the future.

References

1. Cohen, Harold. Parallel to Perception: Some Notes on the Problem of Machine-Generated Art, 2000.

2. Deng, Jia, et al. "ImageNet: A Large-Scale Hierarchical Image Database." 2009 IEEE Conference on Computer Vision and Pattern Recognition, June 2009.

3. Fei-Fei, L, et al. "ImageNet: Constructing a Large-Scale Image Database." Journal of Vision, vol. 9, no. 8, 22 Mar. 2010, pp. 1037-1037.

4. Grassini, Simone, and Mika Koivisto. "Understanding How Personality Traits, Experiences, and Attitudes Shape Negative Blas toward Al-Generated Artworks." Scientific Reports, vol. 14, no. 1, 19 Feb. 2024, p. 4113.

5. Jiang, Shuqiang & Zhu Yaohui & Liu, Chenlong & Song, Xinhang & Xiangyang, Li & Min, Weiqing. (2020). Dataset Bais in Few-shot Image Recognition. 10.48350/arXiv.2008.07960.

6. Millet, Kobe, et al. "Defending Humankind: Anthropocentric Bias in the Appreciation of Al Art." Computers in Human Behavior, vol. 143, no. 0747-5632, Feb. 2023, p. 107707.

7. Schwarz, Gabrielle. "The Prophecies of AARON." Outland, 4 Nov. 2022.

8. Zhou, Yizhen, and Hideaki Kawabata. "Eyes Can Tell: Assessment of Implicit Attitudes toward Al Art." I-Perception, vol. 14, no. 5, 1 Sept. 2023.

For Further Discussion

This serves as an overview of the project and does not include the complete work. To further discuss this project, please email Julia Bock.

Course Overview

CSC 494: Senior Thesis II is the second part of a two-semester series in which students work independently under the guidance of a faculty member on a significant thesis culminating in the development of a senior thesis. The CSC 493/CSC 494 course sequence provides students with an opportunity to synthesize their knowledge of computer science. Students explore the profession of computing by engaging in the professional literature and exploration of professional ethics. Students meet regularly to present and discuss progress. During the second part in the sequence, students complete the thesis proposed in CSC 493.

Explore Our Areas of Interest

We've sorted each of our undergraduate, graduate and doctoral programs into unique Areas of Interest. Explore these categories to discover which programs and delivery methods best align with your educational and career goals.

Explore Computing and Technology at Quinnipiac

Explore STEM Programs at Quinnipiac