Artificial Creativity: Humans versus AI


What does it mean to be creative?

artificial intelligence


In the realm of human inquiry, one concept has steadfastly captivated our imaginations: creativity. Exploring creativity has proven to be an elusive pursuit. Are creative individuals born with an innate ability, or can creativity be nurtured and cultivated? Can we alter our consciousness to inspire creativity? Can other creatures be “creative,” or is this something relegated only to humans? These fundamental questions continue to ignite important discussions.

In August 2022, Midjourney, an artificial intelligence (AI) image generation program, had its capabilities highlighted when a submission generated by the program won a prestigious annual art competition at the Colorado State Fair. The piece entitled “Théâtre D’opéra Spatial” depicted several futuristic figures peering through a golden portal. Human judges were baffled to learn that the artwork was generated via AI, despite the artist clearly stating that his medium was “via Midjourney.” The circumstances of this submission stirred considerable controversy within the artistic community.1 Could an AI win an art contest? Was it making “art”?

These questions about what it means to be creative have given rise to the emerging field of computational creativity, also known as artificial creativity. Researchers within this field are developing, refining, and researching artificial systems that act as creative agents.2

Defining Creativity

Midjourney is only one of many AI image generators; other examples include DALL-E and DALL-E2. AI image generators work in a process that is similar to large language models (LLMs), like OpenAI’s ChatGPT or Google’s Bard. They utilize their large neural networks to work through a process of association. They first sift large amounts of data into abstractions.They then group these abstractions, ultimately arranging pixels to create images.3

LLMs such as ChatGPT and Bard have also ignited discussions regarding their potential artificial creativity, particularly in their writing capabilities.4 These models have captured public attention and raised questions about the nature of creativity itself. The debate regarding the nature of creativity and its potential for quantification roars on. According to the American Psychological Association's Dictionary of Psychology, creativity is defined as5:

“…the ability to produce or develop original work, theories, techniques, or thoughts. A creative individual typically displays originality, imagination, and expressiveness. Analyses have failed to ascertain why one individual is more creative than another, but creativity does appear to be a very durable trait.”

As we analyze this definition, curious questions present: Is AI making “original” or “imaginative” work, or are these creations simple derivations of other human works? Who does this work belong to? Should we consider these to be genuine works of art or are they cheap parlor tricks?


In an effort to explore and quantify creative capabilities, researchers have traditionally turned to psychological testing. Creativity is often attributed to 2 forms of thought. These include divergent and convergent thinking. Convergent thinking affords single solutions for well-defined problems. Divergent thinking affords for idea generation where selections are vaguer and there may be more than one correct answer.6 One of these psychometric tests, named the Divergent Association Task (DAT), has garnered increasing attention.7

A study published in the Proceedings of the National Academy of Sciences in 2021 demonstrated that naming unrelated words can predict creativity. The DAT measures verbal creativity and divergent thinking by evaluating an individual's ability to think of alternative uses for common objects, find associations between words, and solve analytical problems. Participants are required to enter 10 single English words that differ from one another in all meanings and uses. Only 7 of the 10 words are used, in order to allow for misspelling. Researchers found a positive association between semantic distance and performance on problem solving tasks previously known to predict creativity. This test has been validated with approximately 9000 participants.8

The authors of this article sought to compare the capabilities of LLMs on this test. We employed 3 prominent models—ChatGPT 3.5, ChatGPT4, and Bard—and accessed the DAT form at The test instructions were copied verbatim from the website and pasted into each respective model. The authors answered not to the consent for research, so the responses were not contributed as human submissions.

Table. AI Performance on Divergent Association Task

Table. AI Performance on Divergent Association Task

The intriguing results of how these 3 models performed on the DAT are presented in the Table. ChatGPT 3.5 exhibited a creative performance approaching that of an average human. ChatGPT 4, with its larger neural network and enhanced capabilities, surpassed expectations, skyrocketing into the 90th percentile. These scores signify the growing potential of AI systems to emulate and, in some cases, surpass human levels on these measures.

Despite the cursory nature of this investigation, the results were insightful. It is important to note that this was only 1 trial, and it is unclear how often the LLMs have been asked to perform this task before. The authors chose to only query each LLM once as this is analogous to the instructions for task naïve human participants but, when repeated produced approximately the same results. However, there were stark differences in the responses from the respective LLMs. As we delve deeper into the implications of the findings, the contrasting performance of the LLMs prompts us to consider the multifaceted nature of creativity. Authors from the PNAS study noted that measurements of divergence may reflect factors such as overinclusive thinking.8

It is important to note however, that a low score in humans does not necessarily mean a lack of creativity.8 This task is only a single measure regarding an aspect of verbal creativity. One could easily argue that an LLM with a larger neural network would be more proficient at such a task, due to its access to a larger volume of material. Some may argue that anything created by AI cannot be considered creative, as these models were trained from human works, images, words, and ideas. With the exponential growth of this technology, even if one does not agree that these AIs are creative at present, we would assert that it is difficult to argue that they will not be in the future. With the exponential growth of this technology, this is likely to be here sooner rather than later. It remains unclear how society will be affected.

Maybe ironically, at the same time these discussions about artificial creativity are occurring, there are parallel discussions around concerns that humans have become less creative in our technological age. Scores on the Torrance Test, a test of creativity, have decreased significantly since it was first developed in 1966.9 A recent Nature publication highlighted that human publications and patents have become less disruptive over time—potentially as a proxy for being less creative.10

Concluding Thoughts

Will AI creativity further hamper human creativity? Does the emergence of these so-called creative machines open a new world of possibilities? It is imperative to question the ramifications for human creativity. Will the rise of machines dampen our capacity for originality and innovation, or will it serve as a catalyst for pushing the boundaries of expression?

As we reflected upon the results of this endeavor, we asked ourselves what does it mean to be creative? Will human creativity ever be usurped by artificial creativity? If so, when?

Dr Pratt is a resident physician in the Department of Psychiatry at Yale School of Medicine. Dr Madhavan is a resident physician in the Department of Psychiatry at Yale School of Medicine. Dr Weleff is a public psychiatry fellow in the Department of Psychiatry at Yale School of Medicine.


1. Roose K. An A.I.-generated picture won an art prize. Artists aren’t happy. The New York Times. September 2, 2022. Accessed August 1, 2023.

2. Veale T, Pérez Y Pérez R. Leaps and bounds: an introduction to the field of computational creativity. New Generation Computing. 2020;38(4):551-563.

3. Slack, G. (2023). What DALL-E reveals about human creativity. Stanford HAI. January 17, 2023. Accessed August 1, 2023.

4. Samuel S. What happens when ChatGPT starts to feed on its own writing? Vox. April 10, 2023. Accessed August 1, 2023.

5. American Psychological Association. APA Dictionary of Psychology. Accessed August 1, 2023.

6. Zhang W, Sjoerds Z, Hommel B. Metacontrol of human creativity: the neurocognitive mechanisms of convergent and divergent thinking. Neuroimage. 2020;210:116572.

7. Divergent Association Task: fast creativity test. Accessed August 1, 2023.

8. Olson JA, Nahas J, Chmoulevitch D, et al. Naming unrelated words predicts creativity. Proc Natl Acad Sci U S A. 2021;118(25):e2022340118.

9. Kim KH. The creativity crisis: the decrease in creative thinking scores on the Torrance Tests of Creative Thinking. Creativity Research Journal. 2011;23(4):285-295.

10. Park M, Leahey E, Funk RJ. Papers and patents are becoming less disruptive over time. Nature. 2023;613(7942):138-144.

Related Videos
© 2023 MJH Life Sciences

All rights reserved.