If ChatGPT wrote a novel, would it be art? In order to answer that question, let’s refer to the definition of art I provided last time, as well as our knowledge of how ChatGPT (and other text/ image generators) actually work. Art is ‘communication with the primary intent of beauty’. Thus, if AI art is art, it will be communication, will have a primary intent, and will be capable of beauty. Due to the overlap between these parts, however, we will be analyzing its intent mostly as part of the other analyses.

Does AI art have beauty? We cannot deny that AI art has some aesthetic merit. AI art can be enjoyable to look at or read; when the origin is divorced from the originated, only dishonesty could deny the appearance of beauty. AI art, however, does lack an element always present in the beauty of other things: the intent of beauty. The other two categories of beauty- natural and man-made- possess both an author with an intent. In the case of man-made art, such as the works of Leonardo da Vinci and Michelangelo and Tolkien, a man possessed that intent; in the case of creation, of nature and mathematics and all else which man finds in observation to be beautiful and not of his making, the beauty rises from the intent of God Himself, as the Author of all the universe.

As will be later shown in greater detail, AI art has no specific intent involved in large swathes of the process. The AI takes a mechanical process, one too complicated to be intuited by the human observer but mechanical all the same (see last week), applies it to a bare minimum of intent- the prompt-, and produces the finished ‘art’. In all this, the prompter has so little input that it is easily understood that the AI, not the prompter, is responsible for the quality of the product, whether it is ugly or pretty, whether it is comprehensible or obvious balderdash, and the AI’s own creator, the collection of people who wrote, ran, and guided the generative programs which underlie the AI algorithm, cannot be said to have had any intention, divorced as their comprehension was from both the process (too complex for their understanding, as it is for ours) and from the origination (the prompt, still in the future of an utter stranger). The AI, meanwhile, has all the comprehension of a steam locomotive; it takes data from station A, the prompt, to station B, the text or image generated, and it minds not at all what it has done.

This raises a worthwhile question: does beauty require intention? Underlying this problem is our definition of ‘beauty’. Again, I addressed this in Part One: beauty is a reflection of the coordination of God’s nature, of His attributes and deeds, of His wholeness. God is perfectly conscious of Himself (as His omniscience is not limited to that which is external to Him); His beauty, therefore, cannot but be intended. Furthermore, consider those things which man has acknowledged for centuries to be beautiful. Nature’s grand vistas are the work of God; the paintings of the northern Renaissance are the handiwork of God’s image; the majesty of the great composers, of Handel and Liszt and Dvorak, is the mind-fruit of man. All these things were created by intent, whether to beauty primarily or to beauty as a part of a greater purpose.¹

Beauty has two handmaids as well: goodness and truth. Goodness and truth, being other parts of the nature of God, are intrinsically bound to it, and it to them. Without goodness or truth, beauty diminishes, for sin and lying alike are detriments to beauty.² Intent, unfortunately for AI, is integral to both of these. To tell the truth (to communicate it) is, as will be shown soon, impossible for an AI; it simply lacks the knowledge of the truth required to tell it. It can mimic the form of truth, but it has no content to give (remember, AI’s store association, not facts; the AI does not know when the Norman conquest of England took place, just that ‘1066’ is usually associated with the binary sequence it saw when you typed, ‘The Norman conquest took place in…”). As for goodness, we all understand that without intention, neither good nor bad exists. Rocks can harm people, but we cannot, in sober minds, call that rock evil. The person who threw it, on the other hand, we can call such, for he had an intent, whereas the rock made no choice at all.

The AI, therefore, can offer a form of beauty: its appearance. AI, however, has no intent; therefore its art cannot communicate goodness and truth, only the appearance thereof. To borrow familiar words, “Any resemblance to any [goodness and truth], living or dead, is entirely coincidental.” This AI art has only an incomplete and flawed beauty, inasmuch as it has beauty, a facsimile of the skin but not of the blood and bones and soul.

On, then, to communication. Communication has three elements: something communicated, an audience, and a communicator. Let’s consider these in order, starting with what is communicated. AI art obviously provides information to the viewer. This fact, however, is only the start, because ‘what is communicated’ is a bridge between the communicator and the audience. If one of these two is removed, the bridge falls, regardless of how solid it itself is, whether it has five fingers, six, or twenty-three-and-a-half. One side, thankfully for AI art, is solid. AI art has an audience, willing or not: whoever sees or reads it.

This leaves the question of the communicator, and here we run into trouble. Who is the communicator? Let’s run through the three candidates: the prompter, the coders, and the AI itself. The prompter provides a kernel to be communicated, certainly, but the reason he’s using the AI generator is specifically because he isn’t taking the time to flesh out the kernel, turn the seed into the tree. Furthermore, his conception of how the seed should sprout, once the AI has received it, is not consulted. The AI takes the provided information, extrapolates it in ways the human cannot truly foresee and can only vaguely anticipate, and spits back out the result. Most of us don’t consider us telling somebody to paint a ‘A wasteland valley flanked by high mountains, with a road running down the middle’³ to be enough to designate us the artist; at most we might want him to give us a nod. Thus, we can’t call the prompter the artist.

What about the coders? They provided a process, but along two measures they did not provide meaning First, they did not anticipate or provide the meaning of the prompt or its results. They created processes and connections (at a remove, given that AIs are guided through generation rather than written); the inclusion of apparent randomness also matters here, as it removes the coders’ ability to claim anticipation of and provision of meaning to the prompt’s result. Second, AIs are generated, not written; this means that their actual process of analysis, once the AI gets large enough, is not human-designed. In fact, in many cases it’s almost impervious to human analysis, in part due to its large size, which can have some… disturbing results.⁴

The AI, then, remains the final possibility, but the AI lacks a consciousness or agency to provide meaning. Saying it provides meaning is like saying that Gutenberg’s wooden printing press provided meaning. Just as the printing press takes an input- paper, ink, pressure- and provides an output- printed text- without every comprehending or needing to comprehend the information on that page, so too the AI takes an input- a pattern of electrical charges into which the text has been encoded- and provides an output- a much larger series of electrical states in a pattern correlated with the original input’s pattern- without any understanding of what it processes. In other words, the AI can’t communicate because it has no capacity to possess communicable meaning.

AI art, therefore, has an audience but not a communicator; its communication, therefore, the bridge between these, is revealed to be an illusion. As for the presence of a ‘primary intent’, the AI has none such, and the prompter’s intent underlies only a small fraction of the generated result, its outline but not its implementation.

Verdict

Under this definition of art, AI art is not true art. It lacks a communicator, and as a result it lacks intent. It’s beauty is stunted and malformed, aesthetic rather than true. Because it lacks a communicator, it cannot communicate; it can merely suggest in the same way a box of dried pasta poured onto the floor might suggest an elaborate plan for world peace. AI art is just way more convincing due to a much higher rate of correlation with the appearance of actual communication. It has at least an audience.

I admit that this verdict is not perfectly watertight. This is an area of fuzzy definitions. If the prompter is allowed to provide a communicator and a communicator’s intent, AI art could be moved into the periphery of ‘art’. Yet this would be a hollow change. AI art is at best dilute art, the homeopathy of the art world, with its beauty, intent, and communicator reaching their summit at ‘technically present’. To call AI art ‘art’, therefore, is like calling a solution of thirty gallons water to one drop lemon juice ‘lemon juice’. With a loose enough definition,⁵ it’s true, but it’s false all the same. Human intervention in the AI process can, however, move AI art towards being actual art by diluting the AI and increasing the role of the human from merely ‘prompter’ to ‘active participant’. Where exactly the line falls is, I think, intuitive and subjective rather than scientifically measurable; while God knows the precisely metrics to find that line, knows where it is, we don’t have the capacity to do so. By the time it reaches the use of Photoshop’s AI tools,⁶ I think, it has become art instead of a facsimile of art, assuming an artistic intent in the use.

God bless.

Footnotes

1 – Whether you consider nature to be created with a primary or subsidiary purpose of beauty is dependent on the nuances and expansiveness of your definition of beauty, whether God’s glorification is an aspect of beauty or a result.

2 – Technically speaking, this argument is circular. My proof is actually contained in that which I assert it proves. Yet I think that the proof offered- that sin and lying hurt beauty- is more intuitive and more accepted than its necessary corollary; it therefore serves as serviceable rhetorical proof.

3 – A prompt borrowed from my own desultory fiddling with the Adobe AI image generator.

4 – Here’s a video that gives some explanation of how AI doesn’t know anything (and that’s a problem).

5 – Definitions are tricky, as words are shared symbols for things they have no or little inherent connection to (see this segment of an article for more on this). Thus, art can be defined a million different ways, and technically speaking we have no ‘correct’ definition, merely definitions that correspond more or less closely to our individual definitions or to the average definition of the language’s users. I argue for my definition on the basis that it provides a delineated category which includes all things non-controversially artistic, allows for analysis of these and related entities, and is relatively succinct, leaving aside the thickness of the word ‘beauty’, which I find unavoidable. I also plead for it because I came up with it and like it.

6 – With the disclaimer that I’ve never successfully used these tools. I refer to editing tools like those that allow the removal of parts of images by replacing them with plausible continuations of their surroundings. I consider such tools much closer to cameras than anything else, on an artistic scale, albeit the more detail and choice is offloaded to the algorithm, the more diluted the art grows. I don’t consider the few images I’ve made via Adobe’s AI generator to be ‘art’.