Senior Capstone Project

I chose to study AI Versus Human Art for my capstone project. I am very concerned how AI is taking artists by surprise and the art world by storm. When I started SHC, my major was graphic design. However, I flipped Graphic Design to my minor and made Communication Arts my major after learning something that really shocked me during my junior year. I was reading online and learned that the occupational outlook for graphic design over the next 10 years did not look good. Since then, I have learned that a MAJOR reason for this grim outlook for graphic design is AI. And after all the hype about AI that I have witnessed over the last year, I knew exactly what I wanted to do for my Seminar project. Study AI art and learn what artists can do to not just coexist but thrive in a culture dominated by AI.

​

As I began my research, I was inundated by all the ways that AI has impacted art and artists. AI art has disrupted the art world because it has redefined what art is. Art used to be considered a creative expression of a human being or the divine. However, AI has made art the product of algorithms that sift through a large dataset from a computer programmer. Creativity is no longer defined as a person’s imagination or original ideas, but the analyzing of patterns and building predictive models. A verbal prompt becomes an image in seconds. AI art is made faster and cheaper than human art. AI art programs do not credit artists nor compensate them for works produced using their images or style. Copyright laws are behind the curve when it comes to AI. There is no protection for artists. And sadly, many artists are losing their jobs because of AI. A 2023 study from Emory University revealed that commercial artists in illustration, animation, and graphic design are feeling the pain the worst. This pain is not felt by the companies producing AI art. They argue that AI art is good for society because it “DEMOCRATIZES” Art- meaning that everyone can now create an original and award-winning work of art.

 

Chat GPT and Dalle are two very popular AI art and text generators. 1 million users subscribed to Chat GPT in its first five days. It now boasts over 100 million active monthly users. ChatGPT has grown so popular that subscribers often get network server errors. Users can generate essays, recipes, translate languages, tell jokes, and even give advice through ChatGPT. The AI program Dalle generates over 2 million art images a day according to its producer OpenAI. Dalle can also make videos, children’s books and magazines, and Sunday Morning stories. Because of the huge popularity of AI, the Academy of International Business reports that AI will grow 39% per year over the next five years. This will be a market growth of 60 Billion to 422 Billion dollars. At the same time, graphic design jobs are predicted to drop by 4 % according to the Occupational Outlook Handbook.

 

 Because of these sobering statistics, research is needed to help artists survive AI. And when you look at what research is available, the studies out there do not focus on visual art as a medium with a message. When you focus on the message within art, you can dig deeper into how people interpret the message. This study looks at the decoding and distinguishing AI art from Human Art. In other words, how do people code the message that they see or experience when viewing art. And what role do these codes have in distinguishing or telling the difference between AI art and human art. It is my hope that these insights can help artists learn what makes AI art appealing and how to coexist with AI art generators. 

 

 So the purpose of my study was three-fold. First, I wanted to see if patterns exist in how people prefer AI art over human art and vice versa.  Second, I wanted to see if there are any measurable relationships between the context surrounding the art, the artist, or audience and how the art is evaluated. And last, I hoped to learn more about people’s opinions about the quality of AI art compared to human art as well as what they thought about AI’s influence on the art world.

 

My research question for this study was “Are there measurable differences in how people interpret and distinguish human-made from AI art?” And the areas I wanted to focus on were context, preferences, and art-origin. I selected these areas after reading studies on AI art and human art that had these themes in them. And I also focused on these three areas because they related directly to the two theories I selected for my study, which are Decoding Theory and Schema Theory.

 

 A Literature Review on AI art and Human Art showed that previous research has focused on studying opinion about whether AI art is actual art or the difficulty of distinguishing AI art from human art. I could not find any study that looks at how an audience decodes a message that is coded in an image generated by AI or made by a human artist. Instead, decoding studies that are available focus on messages within the media or public broadcasting. This study is different in that it treats visual art as a medium with a message. The title of an artwork is the intended message of an artist. Audience opinion can be studied to see how well an artist’s message is received or not.  

 

 An audience receives or decodes a message in three ways according to Stuart Hall who developed the Encoding-Decoding theory of communication. If the audience fully agrees with the message of the author, the dominant code is selected for the message. If the audience does not agree with the message at all, the oppositional code is chosen. A negotiated code is given to a message if the audience agrees with the author’s message but for a different reason or because of a different point of view.

 

Stuart Hall explained that a message is filtered through context before it is received and coded. Context is what makes the difference in how different people code a message. Because context varies from person to person, different people have different codes for the same message. People who share the same socioeconomic background, beliefs, and culture as the artist may identify with the artist’s message and give it a dominant code, while those who come from a different background may reject the author’s message altogether. Some people may agree with the artist’s overall message, but see the message from a unique point of view because of context.

 

A theory related to Reception Analysis is Schema Theory by Frederic Bartlett. Both theories recognize that context plays a role in the understanding of an artwork. In Schema theory, bits of information stored in our memories shape how we perceive an image and the world around us. Human beings understand the world by constructing models of it in their minds. For example, a three-year old sees a horse for the first time and four schemas are stored in his brain about a horse: Hair, Four Legs, and Tail. The next time the child sees a cow, he says “horse” because of the three schemas already stored. When the child is told that the animal is a cow instead, he will then store additional schemas that help him distinguish between a cow and a horse. This cycle just repeats itself over and over as we get exposed to new stimuli and store more information about what we see and experience.

 

A researcher named Hong applied Schema Theory in 2018 to a study of AI art versus Human art. He argued that schemas affect opinion about AI art as real art, and he said the schema that had the most impact was the knowledge of origin. In other words, knowing that the art was produced by AI would influence some people to think it was less valuable than human art.

 

The research I conducted applied both Hall’s Reception Analysis and Bartlett’s Schema Theory to look at the coding patterns and schemas of different people who were asked to evaluate AI and human art.

 

The research I conducted is considered pure research because I was looking for a better understanding about AI art versus human art to help artists coexist with AI art generators. I designed a nine-question survey that asked participants to evaluate two images on how well they expressed, “Leaping Freedom Heart of Fire.” This message is the title of a drawing that I found by artist Kelly Carroll on Etsy. She created the drawing with oils and acrylics on canvas. I then entered the title of her image into DallE-2 and generated the image shown here. This is a unique image. There is no other one like it. At the beginning of the survey, I did not tell the participants that one of the images was AI. I wanted them to rate the images without origin being a factor.  Only after the participants chose which image was best for the message and they selected reception codes for each image, did I tell them that one of the images was generated by AI. Then I asked them to guess which AI image was the AI art. After they guessed, I distinguished the two images for them. Then I asked if their opinion changed about their best choice after learning the origin of that choice.

 

My sample was current students and staff at Spring Hill College. With CAT standing for Communication or Visual Arts Training, I wanted my sample to have a mix of those who had CAT and those who did not have CAT. And I wanted that ratio to be close to the make-up of Spring Hill College’s student body. There are more students majoring in fields unrelated to communication or visual arts, so my sample reflected that. I had 12 participants who majored in communication or visual arts, while the other 43 were majoring in something else.

 

 The male to female ratio in my study was 55% to 45%. The age range of the 53 participants was 17 to 39. 26% of the participants were black, 60% were white, and 12% were other. 

 

 A surprising result of the study was about education and the ability to distinguish AI from human art. I went into the study thinking that the more education you had in communication or visual arts, the better you would be at identifying the art. This was not the case. A Chi Square test was performed during data analysis and there was a significant relationship between education and ability to distinguish art but the reason for this association was just the opposite of what I had thought. Instead of education helping a person to distinguish the art, it made identification more difficult.

 

60% of those with less than 2 years of communication or visual arts education could distinguish the art. Only 27% of those with 3 or more years could distinguish the art. And not a single person who had four or more years of education could distinguish AI from human art. THIS BLEW ME AWAY! There had to be a reason and I had to find it, and I think I did. More on that later.

 

Five facts became clear from this research. 1. Art type did not change opinion. When participants found out that an image was generated by AI, they didn’t change their mind about it. 2. Students from all backgrounds have an unfavorable view about AI’s impact on art culture. 3. There were no significant differences found in how people decoded the art. 4. More education did not improve the ability to distinguish the art and 5. Schemas influenced the choice of best image for “Leaping Freedom Heart of Fire.”

 

 I was surprised how thematic analysis revealed a pattern that I had missed while surveying the 53 participants. Those who chose AI as the best image were focused on the heart, technique, and precision in their responses. Those who chose the human image as the best one focused on the woman, no boundaries, and imperfection in their responses. I believe this preference could explain why all those who had four or more years of training could not distinguish the two images. They tended to rate the AI image better for technique and then mistook it for human art.

 

Following my study, I asked a psychologist who knew SPSS to evaluate my study. She gave me the following criticisms. 1. More thematic analysis may have revealed additional patterns that SPSS could not, particularly in how people from different backgrounds (like age, race, gender) evaluated the art. 2. My data entry in SPSS and testing was correct, but I failed to display cell proportions in several of my tables to give a better understanding of why the statistical relationships were found.  3. She said my survey provided a good mix of quantitative and qualitative data, and that I didn’t need to remove the open-ended questions. This was a criticism of several participants who said there were too many.

 

 The big take aways from this study are to know your audience’s preferences, connect with your audience’s preferences, and to view AI in a better light. If artists wish to survive, they need to use AI as a tool to improve their skill as a source of inspiration. Artists are inspired by other human artists. Imagine all the possibilities if artists would consider AI as a source for inspiration. 

​

It was a close one, but human art won 29-24!

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​

​