A Mind at Work
The productive chaos of the human creative process will never be replicated by the algorithms of Generative AI.
When Elon Musk predicted that AI will eventually replace all human jobs, it felt like rare acknowledgement from within the tech industry that generative AI may not be universally beneficial. Musk did, however, leave the precariously-employed masses a glimmer of hope when he acknowledged certain creative tasks might remain better suited to humans, even if only in the capacity of hobbies.
Musk is hardly known as a defender of the arts, but his small concession does lend support to those creative types who insist that the output of generative AI tools should not be called art. But even if that opinion is widely shared, it’s not easy to define exactly what makes human art superior to the AI-generated content that imitates it.
Despite their name, digital neural networks are very different from human brains, and the detailed inner workings of most AI algorithms are (so far) impenetrable even to their own creators. The inputs and outputs are fully known: text prompts describe what is desired, and moments later, words, images, videos, or music appear in fully-realized form. But what happens in between? What concepts are associated with specific virtual neurons? What precise points in the original training data are used as sources, and which alternatives are discarded because they do not meet a certain score threshold? With AI, we can speculate, but we don’t really know the answer to any of those questions, as there is not always an easy way to disentangle the complex calculations that produce a given output.
The human creative process, on the other hand, is an open book, and it is arguably this very process of creation – with its inevitable struggles, dead-ends, problems, solutions, and surprises – that distinguishes human artistic creation from artificial imitations.
After I first read about “punching in,” a method some younger rappers are using to write their songs, I wasn’t sure what to make of it. The technique doesn’t actually involve any writing. Pen is never put to paper. Fingers do not tap keyboards. Instead, the rappers compose out loud, recording a single line of lyrics on the fly as it comes to mind. Then they hit a button to re-record the same line, replacing the first take with a slightly better version. The initial attempts often contain placeholder words and nonsense syllables, but with each retake, the line is refined a little more. Once it’s satisfactory, the rapper moves on, punching in subsequent lines one at a time until a track is complete, written and recorded simultaneously.
This technique might alarm an old-school English teacher struggling to stress the importance of formal outlines, theses, and conclusions to her indifferent students. But anyone who tries to write with any intent soon realizes how much of the craft takes place long after a tentative first draft has been typed. “People think that writing is writing, but actually, writing is editing,” said Chris Abani, an award-winning author. “Otherwise you’re just taking notes.” Punching-in might feel new, but on reflection, it’s a form of verbal editing, one without a cohesive first draft but many ephemeral later drafts.
In more traditional songwriting, physical copies of early written lyrics rarely reach the public, and when they do, collectors pay big money for the first drafts of well-known songs. It’s fascinating to see early lyrics written by, say, John Lennon, and recognize familiar lines sitting alongside other awkward clunkers that were discarded before the song was recorded. The monetary value of lyric sheets comes from their physical proximity to a celebrity, but they’re also important because they expose the artist’s creative process. For example, we might see ideas in the first draft of a song disappear, then reappear later in an entirely different song, as happened with the Beatles’ In My Life and Penny Lane, recorded roughly a year apart. It’s worth noting that the Beatles sometimes also used a prototypical punching-in method of songwriting, composing the melodies first using placeholder lyrics – Yesterday famously began life as “Scrambled Eggs”.
This is the creative process at its messiest and most interesting, and it shows us how amorphous ideas often need to bounce around awhile before becoming timeless art. Whether written on paper or punched in, whether composed over hours or years, art rarely goes from conception to completion in a straight line. The human creative process is just that: a process. And process can’t be neatly separated from the final result.
In 2016, my wife and I visited the Met Breuer gallery in New York City to see a show titled Unfinished: Thoughts Left Visible featuring incomplete paintings and drawings by artists such as Edouard Manet and Pablo Picasso. Some of the artworks had been left as mostly raw canvas, marked with only a few lines. Others showed highly-polished sections of paint bumping up against loose background washes and skeletal pencil sketches. Though these paintings were never intended to be exhibited unfinished, I found the exhibit unexpectedly inspiring, particularly as someone who has tried his own hand at painting.
The lack of polished completion peeled away the mythical “magic” of creation that tends to linger around work by famous artists. It exposed the hard work – the planning, the sketching, the layering, the mistakes, the retouching – that goes into a painting, even when the artist is highly accomplished. With their process exposed to viewers, these artists seemed a little more human. Painting like Manet seems impossible, but that simple pencil outline looks doable. And that thinly smeared underpainting, well, maybe I could recreate something like that, if I really tried – and now I’m halfway to painting like Manet, at least in my mind. Rather than intimidatingly perfect, the unfinished work becomes an accidental roadmap for others.
Contemporary artists often share their work on Instagram, and some of their most popular posts are process videos. Watching a few initial dabs of colour gradually come to life as the artist’s hand deftly manipulates the paintbrush is mesmerizing. But few painters routinely showcase their mistakes, or their failed experiments. Surely, they must make mistakes? Even the most skilled artist is not perfect, especially if they are pushing themselves to improve. But like all Instagram influencers, these artists are incentivized to present an artificially-enhanced version of themselves, one that serves their own purpose, which is presumably to gain followers and make sales. This is completely understandable, and cultivating a myth of artistic genius is probably good for business. But for aspiring painters who are not as far along in their craft, the perceived perfection can be disheartening. When genius seems to spring forth from another artist’s brush like magic, and your brush doesn’t have any such magic, what is the point in trying?
The more transparent artists openly share the contents of their sketchbooks. These typically contain years worth of quick studies and rough ideas that may, but may not, turn into full-fledged paintings at a later date. Of course sketchbooks can also be manipulated – pages with embarrassing gaffes can be discreetly sliced out and never spoken of again, ask me how I know – but in general, these sketchbooks are more true to the creative process. They are reminders that the best art does not appear in a single flash of inspiration, but is instead based on ideas that have been nurtured for months or years before finding their proper place, much like those Beatles lyrics.
Maybe a mundane sketch lies forgotten until another somewhat different idea reminds the artist of it; then those two related ideas, when put together, create a new idea that is much more compelling than either was on its own. Those intermediate exploratory steps can’t be skipped, because an artist might not realize what they’re looking for until they land on it. “Success is sometimes the outcome of a whole string of failures,” wrote Vincent Van Gogh, in a letter to his brother. Van Gogh’s letters are known for documenting his sad decline, but they also reveal his dedication to continual self-improvement in his art, and his intuitive understanding that making diligent studies and reworking the same subject repeatedly from different angles was the only way to move his art forward.
In the end, I suspect what we crave from creative works is not perfection; what we really want is to glimpse something a writing teacher described to me as “evidence of a mind at work”. Consciously or not, we need to detect some trace of the human that put together the words, the brush strokes, or the guitar notes that are supposed to move us emotionally. Perfection, after all, can be creepy – consider the effect of the uncanny valley in animation. It is also a barrier. We are rarely moved by art if we can't empathize with its creator, and many of us do not empathize with glossy perfection.
And so we return to AI-generated media, and the reason it can induce discomfort – even revulsion – in those who consider themselves to be creators, or at least enthusiastic appreciators of creative works. There is no conscious mind at work behind a block of text or grid of pixels generated by an AI algorithm. It is a facsimile of creativity, at most an aggregation of the past creativity of others. It is an auto-complete tool that begins with a premise, and then calculates what is statistically likely to come next according to its training data. AI makes no drafts, leaves no demo songs, and fills no sketchbooks. Eliminating the process of creation – of experimenting, revising, reflecting, and improving – shortchanges both the person entering the prompt and the audience who has to live with the result. From this perspective, generative AI might output casually convincing content in the blandest commercial sense, but it won’t create art.
For a human creator, the road from concept to completion may be long and winding. A mediocre first-draft essay about a pleasant vacation may eventually turn into an excellent finished piece about the modern culture of overwork, or a trite song itemizing the disjointed memories of a childhood might turn into a meaningful reflection on the inevitability of change. With humans, each idea, even a failed one, can be a jumping point to the next. But AI travels from concept to completion on an express train, one that speeds through a dark tunnel with no scenery and no intermediate stops. AI gives us exactly what we ask for, immediately, and nothing more. We might think that’s what we want, but it’s probably not.
References
#artistsketchbooks. (n.d.). Instagram. Retrieved June 11, 2024, from https://www.instagram.com/explore/search/keyword/?q=%23artistsketchbooks
An exhibition celebrates unfinished art. (2016, August 28). CBS News. https://www.cbsnews.com/news/an-exhibition-celebrates-unfinished-art/
Barnes, J. (2013, September 13). Distracted Writing with Chris Abani. The Poetry Foundation. https://www.poetryfoundation.org/poetry-news/68814/distracted-writing-with-chris-abani
Coscarelli, J., & Throop, N. (2023, September 20). Why Rappers Stopped Writing: How Hip-Hop Is Made Today. New York Times. https://www.nytimes.com/2023/09/20/arts/music/punching-in-rap-video.html
EconTalk (Director). (2023, April 1). AI is Autocomplete [Video recording]. https://www.youtube.com/watch?v=RvU-uRdZ90I
Helsinger, K. (2019, June 11). Why You Should Create Videos of Your Art Process. Messy Ever After. https://messyeverafter.com/self-employed-journey/why-you-should-create-videos-of-your-art-process/
In My Life. (2008, March 15). The Beatles Bible. https://www.beatlesbible.com/songs/in-my-life/
Kelly, S. M. (2024, May 23). Elon Musk says AI will take all our jobs. CNN. https://www.cnn.com/2024/05/23/tech/elon-musk-ai-your-job/index.html
Kendall, E. (2025, January 9). Uncanny valley | Definition, Origin & Facts. Britannica. https://www.britannica.com/topic/uncanny-valley
Mineo, L. (2023, August 15). Is art generated by artificial intelligence real art? Harvard Gazette. https://news.harvard.edu/gazette/story/2023/08/is-art-generated-by-artificial-intelligence-real-art/
Tangermann, V. (2024, June 3). Sam Altman Admits That OpenAI Doesn’t Actually Understand How Its AI Works. Futurism. https://futurism.com/sam-altman-admits-openai-understand-ai
The hilarious original lyrics Paul McCartney wrote for The Beatles song ‘Yesterday.’ (2021, January 17). Far Out Magazine. https://faroutmagazine.co.uk/paul-mccartney-beatles-yesterday-original-lyrics-scrambled-egss/
Perrigo, B. (2024, May 21). Artificial Intelligence Is a “Black Box.” Maybe Not For Long. TIME. https://time.com/6980210/anthropic-interpretability-ai-safety-research/
Van Gogh, V. (1882, October 1). Letter #270. Van Gogh Museum. https://vangoghmuseum-prod.azurewebsites.net/en/highlights/letters/270
Wabi-Sabi: The Japanese Art of Finding the Beauty in Imperfections. (n.d.). Carnegie Library of Pittsburgh. Retrieved June 11, 2024, from https://www.carnegielibrary.org/staff-picks/wabi-sabi-the-japanese-art-of-finding-the-beauty-in-imperfections/