The aesthetics of AI use
Some thoughts
For a second time in several weeks, I've seen some writing on the idea of what it's like to write in the era of AI. The most recent was The Typo Vibe Shift by The Atlantic, and the first was a Substack Note that I cannot find because the platform’s search is terrible, and most recently this post on how to Substack. The thing I've been thinking about through these is that there is a changing aesthetic of writing, and more people are becoming acutely aware of how aesthetics pervade what we create and consume—how we communicate. Aesthetics can mean different things in different contexts, but here what I mean by it is that there is a type of style. To those that are familiar with trends like the “cool girl aesthetic” or the “performative male aesthetic,” this use is more akin to those.
Regarding aesthetics in the age of AI, there is the distinctly rebellious aesthetic of “I created this without using AI, even if it resulted in a worse product.” In its most blatant form, this is the aesthetic of typos and distinctly human writing, and in its more subtle form it is a particular writer’s voice. It is even the aesthetic that I adopt here. Yet, note that this aesthetic is not right in the strong sense—it is not inherently better than another choice of aesthetic. If I were to use AI to elaborate on some of my key points and polish my writing, it's not clear how that is wrong in abstract, as long as I confirm that the additions reflect my vision. Rather, it might not be the choice I, as the creator, want to make to communicate to my recipient how I see my work. If my purpose were to create a piece of writing that is clear and concise, then it may well be a strong choice to use AI to write in the aesthetic of, say, a public academic. But if my purpose is to share my reflections, then that choice may seem confusing. Regardless, the aesthetic is a choice, and one that is used for communication, just as is any other creative decision.
For the aesthetic of a creative product to matter, its content value must already be sufficient. In an forthcoming work, my coauthors and I discuss how key criteria for a proper delegation of a task to AI are trust and competence. For the sake of relevance, let's examine blogging as an example. Trust here comes into play at a couple different levels. First, the reader must trust that the author was sufficiently involved in the production of the work—if delegation is allowed, it must be transparent through either explicit citation like giving the model a byline or implicit social norms. Second, if the author delegates part of the writing to an AI system, the author must trust the system and its creators to generate high-quality and correct writing. Competence in delegation to AI means that the author is able to and does use their basic rational abilities and domain-relevant knowledge. It would be an inappropriate delegation for me to use AI to write about aerospace incidents, because I don’t have enough domain knowledge to verify the correctness of claims the AI would generate.
Let’s assume all goes well in a delegation: the delegation is transparent, trust relations are preserved, the author has sufficient competence, and the author is not deskilling herself.1 In these cases, it might still be inadvisable to use AI, as an aesthetic decision. This is because even if the author cannot identify the "AI-ness" of their writing, readers likely can, or at the very least, the author is missing an opportunity to share through their voice (and perhaps even missing the opportunity to develop it). But as an aesthetic decision, this is akin to choosing to paint a painting in the style of impressionism, with noticeable brush-strokes, instead of a photo-realist style, with depiction as accurate as possible. Neither decision is more "right" than the other in abstract, for the purpose of producing a painting. They capture different subtexts and produce a different response in the viewer, but both produce the same product: a painting of a subject.
Assuming the competence and trust conditions are met, the products only vary with respect to their aesthetics, not to their truthiness or factual value. Aesthetics might be implicated in other pragmatic considerations, such as whether readers are receptive to the content or take it seriously—and that is the point. The aesthetics matter only insofar as they are relevant in these other considerations, not for an intrinsic value.2
For an object’s aesthetics to matter for an intrinsic reason would mean that the product itself is somehow valued. While we may use language in this way, such as saying that the Mona Lisa is a valuable piece of art, it is not clear that this is really what we mean. If there were no humans and no historical context, the Mona Lisa is only a collection of fiber, oils, and pigments. Instead, the language of intrinsic value is a shorthand for other extrinsic values. An object is valued by an agent. The Mona Lisa is valuable because of its historical relevance in the Renaissance period, and a great novel is valued for the impact it has on its readers, the way it is able to transmit ideas through time and space. But these are still agent-centered impacts, not some abstract value that exists in aesthetic space.
tldr;
Even where AI use is acceptable—it is not deskilling the user, deceiving the audience, and is reasonably correct—there are still other reasons we should be careful about using AI. We create things for a purpose, and we should be hesitant to use AI where the product of AI use is not aligned with those purposes. When writing marketing copy devoid of human experience and personality, delegation to AI may be a strong choice. But when writing applications to academic programs or jobs, delegation may be a bad choice even when the model is prompted with personal experiences because the aesthetic of AI writing raises alarms in the reader’s mind.
Footnotes
-
Vallor, S. Moral Deskilling and Upskilling in a New Machine Age: Reflections on the Ambiguous Future of Character. Philos. Technol. 28, 107–124 (2015). https://doi.org/10.1007/s13347-014-0156-9 ↩
-
It is at this point that I should note that I have no training in the philosophy of aesthetics, and this is an entirely from-priors discussion of aesthetics. ↩