My opinions are subject to change based on new information. This is not an attempt to be confrontational, stubborn or difficult. This is not a criticism of Artificial intelligence (AI) nor is it a criticism of its applications. Whether you are for or against AI use in our daily lives, one cannot deny that the technology is remarkable.
Author: Natalie Rosella Boonzaier
Channel: https://www.youtube.com/@NatalieRosellaBoonzaier
Website: https://rosellaprojects.com/
Instagram: https://www.instagram.com/rosellaprojects/9-minute read
Skip to Part 2 for the Arguments Section.
PREFACE A few months ago, I came across a cool video snippet. It was an AI-generated clip of a Xenomorph/Neomorph that displayed its anatomy and form. I immediately assumed the clip was AI-generated and I didn't like the product, nor its concept, any less. Looking into the work, it was described as AI-generated artwork. And the use of the term "artwork" didn't annoy me as such, but it got me thinking about the topic of AI-generated imagery.
As part of my preparation for this article, I left a poll on my channel. Is AI-generated imagery “art’? Two responses were made available: “Yes, AI-generated imagery is Art” and “No, AI-generated imagery is Not Art”. At the time of writing this, the Poll presents a majority of partakers voting that AI-generated imagery is not art (80% vs 20% n=93).
PART 1: INTRODUCTION
(IMAGES as DATA // IMAGES as ART)
I was at a social engagement the first time I heard about AI-generated imagery. I spoke to a person who mentioned Midjourney and how they had used it to produce a “photo” for their significant other. The outcome sounded cool and I had no problem entertaining their admiration for the software. It was only towards the end of their description of the software and of their image, when the person said (knowing well that I am a photographer) “You should give Midjourney a go” that I felt a gut reaction. Something felt off.
The initial jerk reaction was not positive, and I can imagine that might be how some of the painters initially reacted to the introduction of the camera a few hundred years ago. I’m assuming they were not overwhelmed with positivity. The camera could now capture a portrait or a landscape in a less time-consuming manner, and with potentially less effort. To clarify, the camera obscura has existed in principle since around 400-500 BC, but it was embraced as a potential artistic tool in the 17th century. Fast forward a few centuries and various sub genres of photography are now classified as art. However, not all photographic images are art, and a set of criteria need to be adhered to to classify an image as art…unless you subscribe to the Dadaist ethos. In which case, does it matter?
As an artist who is also a photographer, I am not particularly worried about AI and its potential effect on my ‘existence’ as an artist. I just feel a bit odd about my current feelings toward art - what it is and what it isn’t. Interestingly, researching for this essay has unexpectedly resulted in me feeling less for my photography as an art form. Additionally, I am concerned over my degree of concern - I thought I didn’t care. Yet, here I am writing this.
This essay is written from the perspective of a photographer that identifies as an artist, and lays out my opinions on what art is and why I currently feel that AI-generated imagery does not fall into the category of art. A fair amount of this essay will rely on semantics.
From the Oxford English Dictionary:
"Art: the expression or application of human creative skill and imagination, typically in a visual form such as painting or sculpture, producing works to be appreciated primarily for their beauty or emotional power."Based on the primary definition found in the Oxford Dictionary alone, one could close the case right there.
Conversely, one could argue that a human typing the prompts into the AI software adds the human component - and I will address that point.
But, before I get ahead of myself - What is a “prompt” and what do we mean by “AI-generated imagery”?
An Artificial Intelligence (AI)-Generated image is an image created by an AI model trained to generate visual content. Broadly speaking, an AI model is essentially a computer program that learns from data to generate an output or, in the case of image generators, to generate a new image. With increased training, the program will improve its performance over time [1], and will refine its ability to generate new content.
In the case of image generators, the model can generate an image based on a command set (prompts) that a human types in as text. With said prompts, the trained generative models synthesise new images [2]. Generally, these models are trained on large datasets of images available online. These models are trained to identify colour palettes, texture, brushstroke styles, shapes and geometric patterns by processing images as numerical data. The model can even analyse composition based on how various elements, shapes and vectors, are structured within images. Instead of “understanding” an object ("this is an apple", for example) as a human would, the model mathematically analyses common characteristics of apple images and generates new variations based on learned patterns. For a more in-depth description of current AI models and image generation processes, you can check out the Reference section, particularly [2] and [3].
Importantly, AI does not perceive or interpret images as humans do. While an AI model can generate photorealistic, painterly or artistic outputs, the model lacks emotional and cultural awareness. It doesn't understand contextual and artistic intention. AI-generated images are the result of statistical pattern-matching rather than intentional artistic expression. For a deeper dive into the factors that contribute to the 'uniformity' of AI-generated images, check out the article linked in the Reference section [4].
In the case of an AI-Image Generator, what is essentially happening is that the model generates a “new image” that mimics certain components within the images on which it was trained. And it will do so based on what prompts are typed by a user. For example, “Painted portrait of an elderly woman with long hair sitting back in a large antique wooden chair with warm candle light and a dark background”. I have never prompted an AI Image generator, so I don’t know if this is a “good’ prompt (it's probably a bad one). That wouldn’t matter, however, because if I didn’t want to learn how to write a decent prompt, I could go online and buy prompts to generate portraits in the style of Rembrandt - I kid you not.
But once we set aside the phenomenal feats that this technology is breaking, once we segment the scientific technology from the real-world general use of the image generator, and once we recognise what proportion of the image generation is machine-generated vs the proportion that is human-made, one can acknowledge that the creative process has been, to a great extent, cut out. The creative process and human artistic intention of the original works, on which the technology was trained, is eliminated or ignored (or worse, forgotten). Sadly put, in order to write these programs, the creative process and intentional strategies that drove the original artworks (on which these AI programs are trained) are simply forgotten and not taken into account.
PRIMARY QUESTION: IS AI-Generated Imagery REALLY ART?
As much as I don't mind a grey area, my opinion is NO.
"Can something produced by a non-human entity be considered art?"
Some insist that there are strong arguments on both sides. I disagree. I don't really hold it against anyone for having an opinion that opposes mine in this matter, but at my core, I can't help but feel that the attempt to broaden the definition of art is inappropriate. AI lacks subjective experience. AI does not experience emotion. AI can generate imagery that is art-like, not art, by definition.
PART 2: EXPLORING SOME OF THE ARGUMENTS
(AI Art or AI Imagery?)
Arguments That AI-Generated Imagery Is Art:
a. If the output evokes emotion, tells a story, or has aesthetic value, then it's art.
Response: No, if an output evokes emotion it means that the output was displayed to a sentient being, most probably a human.
If we entertain the idea that the main criteria required is a human reaction to classify a piece as art, then:
- What of the art that never gets seen? The argument that art is primarily about the end product, and not the creator, presents a superficial stance at best. Many artists produce art with no intention of sharing it. If an image is dependent on an audience response or reaction to classify it as art, then what of the art that no human except the creator will observe? Is that human-made piece, therefore, not art?
- What of the art that evokes no emotion? I have found myself to be, at times, indulgent to the notion that a painting is just a painting or an image is just an image - no emotion felt, no story perceived, no aesthetic value identified. I can acknowledge that my opinions reflect such apathy predominantly when I am low on energy and not receptive to feel the art or its intention. However, even in an apathetic state, I can acknowledge artistic intention without feeling one iota of emotion. Is that human-made piece, therefore, not art?
b. AI Image Generators are Tools and Art can be defined by its impact on the audience, alone, and not by artistic intent. If an AI-generated image evokes emotion, sparks thought, or inspires creativity, one could consider it art, regardless of whether a human or an algorithm created it. Therefore, AI is just another tool. If an artist uses tools, like a painter uses a brush, a photographer uses a camera, a digital artist uses software, AI is just another tool.
Response: Yes, AI is a tool, and part of this toolset comes the data that this tool analyses. It is an algorithmic tool that gathers existing data from an imaging library from which it has been (probably illegally) systematically, mathematically, and probably also stochastically trained. The AI model then identifies a pattern to generate an output of data that we can perceive as an image.
If the concept of AI being a tool is invoked as a defence for its product being art, would it not make more sense for the data within the AI's algorithmic model to be considered the tools, and the AI's algorithm itself being the generator (“creator”) of the image, therefore making the human the commissioner of the image? With this, even if one were to consider this to be "art" it is not the prompter's artistic work, it is the prompter’s commission [The Commissioned Art Analogy (5)]. And, as the AI is not human, it cannot be the artist - One should rather state that the AI model is the generator of the synthesised image. With this, I can acknowledge that something is generated. But, I will insist that the product is not art; it simply cannot be art by definition (and probably by law).
I have realised that one of the strongest arguments in favour of AI-generated imagery being art is: Human input, in the form of prompt, is still required. The prompts, modifications, and curations by humans mean there is still creative decision-making involved [6]. See below.
Arguments That AI-Generated Imagery Is Not Art:
I take the final point from above, and consider it to, in fact, be a major pitfall to the argument that AI-generated imagery is art: “Human input is still required. The prompts, modifications, and curations by humans mean there is still creative decision-making involved”. Let’s consider the Commissioned Art Analogy.
a. The Commissioned Art Analogy: If a human artist is commissioned, they bring their own skill, interpretation, and creativity to the commissioned piece. In many instances, the commissioner may have directed the work. In the case of an image produced via an image generator, AI does not interpret one’s vision in a creative, emotional or intentional sense; AI does not "interpret" image data like a human does - AI mathematically generates imagery based on patterns identified during training. If you consider my argument made a few paragraphs above, covering the ‘artistic tools’ analogy, and apply that to the “commissioned art analogy” [5], The AI is the commissioned; the human prompter is the commissioner, not the artist. AI is not a human and it cannot be an artist.
b. Intention and Inspiration: To expand on inputting data, even with one's best intention to produce an AI-generated image that is one's art; I'd be interested to know what proportion of generated final output fits exactly what the prompter envisioned. My guess would be a significant amount of final output might not even be what the prompter initially intended. Yet, if the prompter ends up liking the generated image, the prompter can choose to present it as their own. I consider this to be disingenuous. Where is the line that distinguishes intent and accident by incorrect model prediction? This undefined space between original intent and accidental output drives me to argue that the artistic “intentionality" component of an AI-generated piece is void. AI does not intend to express anything. AI generates images based on probability and learned patterns.
To expand on this - a human can be inspired by existing work, even if the human uses this inspiration and visual reference to prompt the production of an AI-generated image; even if a prompter goes into a project having an understanding of what their own intentions are as an artist, the prompter cannot know exactly what original works, and to what extent, various original works were used for the machine training, and therefore in the generation of the final image. A prompter could end up with an image that was synthesised using information from various works of art that were originally created with an intention completely not understood by the algorithm and in no way related to the prompter’s intent.
When an artist is inspired by another piece of art to the point that they want to create a piece of art that stems from that inspiration, that inspiration drives at least part of the intention of the final art piece. AI-generated imagery draws from millions or even billions of data sets. How can one begin to understand the intentions behind every artwork that trained the generative model? This question alone is far-fetched as it suggests that the AI model even had intention to begin with. A lot of the points made in favour of this being art are fundamentally derived from non sense. And I don’t mean that In a derogatory way. The insinuation that there is artistic intention within an algorithm is not rooted in reality.
c. Ethics: Originality, derivation and Theft: AI models, though trained on human-made art to remix existing styles, do not create images from own-lived experiences. There is no personal expression and, with that, no originality. With no originality, one must raise ethical concerns. I do not know what proportion of publicly available image generators are trained on datasets derived from copyrighted or stolen works scraped from the internet without consent, but I suspect it’s most of them (please don’t quote me on that one). Bluntly put, AI generated imagery is not original, but derivative at best, and theft at worst. For this reason, as far as I know, AI-generated work is not easily copyrightable [I am not confident in selecting one source of information to cite on this one but there are hundreds of articles online covering this topic].
Conclusion:
Whether AI-generated imagery is art depends on your definition of art. If art is about human expression, then AI-generated works don’t fit the criteria. But if art is defined by the final aesthetic result and emotional impact, then AI images can be considered artistic works. I dislike being so reliant on the semantics, but we might be at a point where the semantics are needed and a definition for this new realm of imagery is required. Either this, or the core definition of art needs to be contested and amended.
I've walked through my adult life shifting my perspective on what art is, based solely on my mood. If I’m in a decent energy state, then Yeah! I can see art and feel the intention behind it. If I am in a low-energy state (which is my general state) then I couldn’t care less. A picture is a picture and a photo is a photo and I will feel nothing for any of it, not even my own work. I roll my eyes at myself when I refer to some of my photography as art. So if I can feel such indifference to a piece that fits the definition of Art, then what’s the harm in labelling a soulless machine-produced piece as art?
At the very least, AI-generated imagery occupies a new space. A space that challenges traditional notions of creativity, intention and originality. AI and AI-generated visuals are here to stay. It is now the time to encourage critical thinking as observers and as users of AI. We need to equip ourselves to understand the basics of AI and to learn how to distinguish between what makes up the AI component and where the human component lies; what role each one plays, and to what degree [7].
I at no point have felt intimidated by the idea of AI taking my artistic integrity and taking away a commission from me. I am somewhat indifferent to the idea of AI infiltrating the art space. Because, in due time, we will be so bombarded with the sterility being produced that humans will be craving authentic human work, genuine human expression and meaningful human conversation and experience. The question is not how to avoid it, it is how do we navigate through and benefit from AI as humans, while staying true to our humanity. And I ask this today particularly within the context of art, artistic productivity, the creative process and artistic integrity.
The outcomes of my research have also called to question my feelings towards my own photography that I present as art. Perhaps photography, like other forms of imagery, is something other than art, but just as important (I’d like to thank one of my poll participants for opening my perspective to this question).
If I can leave you with anything, let it be this. You do you. It’s your life so do what you want. I do suggest you give it some thought, however. I don’t think this will be settled any time soon - even with a clear definition for art that already exists. AI has resulted in fascinating technology with implications that most of us cannot predict nor understand. AI-generated images are part of the landscape and they will remain part of the picture, so to speak. There are many people using this technology and producing images that have them feeling satisfaction and fulfilment. And who am I to presume to be in any position to take that feeling of fulfilment away from anybody? But I will repeat myself, I strongly recommend you give some thought to how you present your product and what you claim it to be. Whether it’s AI generated, AI-assisted, traditional or purist. If it makes you come alive, keep doing it. Because the world needs more people that feel that way.
// xx
REFERENCES
https://cloud.google.com/learn/artificial-intelligence-vs-machine-learning
https://towardsdatascience.com/generating-images-using-vaes-gans-and-diffusion-models-48963ddeb2b2
https://www.iterate.ai/blog/bias-in-ai-generated-art-examples-and-recommendations
https://www.reddit.com/r/aiArt/comments/10trydu/ai_art_as_commission_defining_the_relationship/
https://www.tandfonline.com/doi/full/10.1080/10447318.2024.2431761#abstract
A BRIEF SUMMARY OF APPROACHES // To maintain transparency, Chat GPT provided me with a list of three approaches with which to tackle the primary question: Traditional approach, Expansive approach and Collaborative approach. I researched this further and developed my own summary of approaches:
I approach the primary question "Is AI-generated Imagery REALLY Art?" using 1. A semantic or traditional view [8], in which human skill, creativity, subjectivity, intent and imagination is required as a prerequisite. 2. A reductive view [9], in which one can argue that art is solely about the final product and not the creator. This approach suggests that art is defined by its impact on the audience, regardless of whether it was made by an algorithm or a human. 3. A collaborative view [10], in which AI is considered an augmentative tool, which allows artists to generate ideas, styles and compositions that they would have not considered independently.
// The arguments in favour of AI-generated imagery being Art embraced the Reductive and Collaborative Approaches. My arguments Against embraced the Semantic or Traditional Approach.
EXTRA MATERIAL
https://www.youtube.com/watch?v=G2XdZIC3AM8
Supporting literature for [1], [2] and [3]:
- https://medium.com/@praneithranganath10/generative-models-a-deep-dive-into-vaes-gans-and-diffusion-models-00de7eb74ac2
- https://www.techtarget.com/searchenterpriseai/definition/AI-art-artificial-intelligence-art
- https://www.bbc.com/future/article/20241018-ai-art-the-end-of-creativity-or-a-new-movement
Supporting literature for [4]
- https://www.theatlantic.com/technology/archive/2024/08/why-does-all-ai-art-look-same/679488/
- https://www.theatlantic.com/newsletters/archive/2024/08/four-theories-that-explain-ai-arts-default-vibe/679498/
- https://getcoai.com/news/why-does-all-ai-generated-art-tend-to-look-similar/