Can we stop being so binary about the use of AI in creative work?
A few weeks ago The New Yorker published an essay by Ted Chiang titled ‘Why A.I. Isn’t Going To Make Art’.
It is - as you might expect from an award-winning writer - a well-written piece.
However, I think it takes an overly binary and unnecessarily defensive position on the role of AI in relation to artistic endeavours.
For one, Chiang presents AI’s involvement in the creation of art as an all or nothing equation; either a human created something without any assistance from AI or AI created something with little or no assistance from a human.
And he paints these two binary scenarios in direct opposition, anthropomorphising AI (“How good could they get?”) and pitting it against humanity (“Could they get better than humans at writing fiction - or making paintings or movies?”)
In reality, we’re going to see a lot more artistic output where humans have involved AI to varying degrees in the creative process, much as digital devices and processes have become a part of so many of today’s creative processes1.
And that’s ok.
Making use of AI doesn’t in and of itself render something art or not art.
Central to Chiang’s thesis is that “art is something that results from making lots of choices”.
I’m not convinced quantity of choices (or quantity of anything else) is a good measure of what is and isn’t art. Should Picasso’s ‘Visage: Head of a Faun’ (1955), which took him 5 minutes to create, not be considered art because it (presumably) involved fewer choices than ‘Guernica’, which is 11 feet high and 25 feet wide and took around a month?
Chiang posits that text-generating AIs “take an average of choices that other writers have made”. However, that’s not how LLMs work. The probabilistic nature of LLMs and the ability to dial up the randomness/creativity of their output using temperature controls/random seeds means they are able to generate novel text and are unlikely to generate the same text twice in response to the same prompt.
Another scenario Chiang outlines is AIs being instructed “to engage in style mimicry, emulating the choices made by a specific writer”. Yes, that’s possible (and AI companies should arguably be doing more to prevent it), but it’s not inherent in using AI as part of a creative process.
Chiang doesn’t detail any other scenarios and concludes that “In neither case is it creating interesting art”. I find Chiang’s use of the word ‘interesting’ here, er, interesting. Interest being very subjective (one person’s interesting is invariably another person’s boring), Chiang implicitly (inadvertently?) acknowledges that AI can be used to create art but he deems it to be universally uninteresting.
He then moves on to paintings, stating that “Real paintings bear the mark of an enormous number of decisions” and then to photography, claiming that “the artistry lies in the many choices that a photographer makes” and that “when you compare an amateur’s photos to a professional’s, you can see the difference”.
Attempting to draw a hard line between the amateur and professional and to imply that amateur photography is necessarily less artful strikes me as problematic. The difference between amateur and professional is money and whilst money undoubtedly plays a major role in art, I’d like to think it’s not the sole arbiter of artistic merit.
If we momentarily accept Chiang’s premise that art is purely about number of choices made, then I disagree with his assertion that there’s isn’t the “opportunity to make a vast number of choices using a text-to-image generator”. In my experience, generating successful images using AI is rarely a one-shot affair and typically involves multiple rounds of careful word selection and regeneration.
Chiang goes on to argue that if a text-to-image generator were to let you enter tens of thousands of words to control the output then “a person could use such a program and still deserve to be called an artist”. The word ‘deserve’ feels revealing here, suggesting an arbiter of who does and doesn’t qualify for the prized moniker of artist, with volume of choices being the key determinant.
Returning to prose, Chiang finds it even “harder to image a program that, over many sessions, helps you write a good novel” and yet that is exactly how Japanese author Rie Kudan described the assistive role ChatGPT played in helping her write her award-winning novel Tokyo-to Dojo-to (Tokyo Sympathy Tower).
And she doesn’t perceive her use of AI to be compromising her creativity: “I plan to continue to profit from the use of A.I. In the writing of my novels while letting my creativity express itself to the fullest”.
Chiang goes on to generalise that “Generative A.I. appeals to people who think they can express themselves in a medium without actually working in that medium”. Ignoring the many talented writers/artists/photographers/musicians etc. excited by/actively exploring the creative possibilities of generative AI, it’s hard not to detect some disdain in “people who think they can express themselves”.
Lots of people lack the ability to effectively manipulate a paintbrush and/or access to the materials to bring an idea to life via traditional methods. Is enabling them to bring their ideas to life with the assistance of generative AI deserving of disdain? Is making it easier for people with lower literacy levels to communicate their thoughts and ideas in prose?
Where would Chiang draw the line on the assistance artists are allowed to receive before their credentials get revoked? Would consulting a print thesaurus be ok? How about a digital thesaurus? How about an AI thesaurus? Should David Hockney’s use of a camera lucida or an iPad or - Lord help us - AI diminish his status as an artist or disqualify those pieces from entry into the Hockney canon?
Chiang then turns to models being trained on copyrighted material - a critical and pressing issue for sure, but not directly relevant to the essay question of whether AI can help create art (it is possible to train AI models without using unlicensed material).
He then looks to reassure his reader that AI models “aren’t particularly intelligent, because they aren’t efficient at gaining new skills” claiming that “It is currently impossible to write a computer program capable of learning even a simple task in only twenty-four trials, if the programmer is not given information about the task beforehand.” That may have been true a few years ago but is no longer the case.
Chiang concludes by arguing that AI’s lack of intention means it is “not actually using language”. Whilst strictly speaking LLMs do trade in tokens rather than words, I think it’s hard to make the case that they are not using language.
Paintbrushes, cameras and laptops don’t have intention but that doesn’t preclude their involvement in creative endeavours.
For Chiang, art appears to be exclusively about the intention of the creator, leaving no room for the experience of the reader/viewer/listener and what a piece of creative work makes them think and feel, regardless of the number of human choices involved in its creation.
We may choose to weigh the value of creative work differently based on the perceived level of intentionality and/or human endeavour (as we have done for many years), but I don’t believe that assessment should be a binary question of whether AI was involved or not.
For the avoidance of doubt, I’m not arguing that all - or indeed most - work created with the assistance of generative AI should be considered art. Rather, I’m advocating for a less binary and defensive response to the use of AI in creative endeavours.
It’s understandable - and entirely appropriate - that we feel concerned about the impact of the widespread application of AI on our lives and livelihoods.
However, we are not under attack from machines. We are experiencing the whiplash of a rapid evolution in technological capabilities coupled with undue haste and insufficient care from companies/nations locked in an arms race in pursuit of market share/dominance and/or AGI.
Rather than railing against a technology and the people making use of it, I’d prefer we focus on demanding governments hold AI companies to account in ensuring models are ethically and sustainably trained and operated, contributors are consulted and fairly remunerated and output is responsibly moderated and transparently labelled.
Or we could just keep trying to put the genie back in the bottle / telling people not to speak to the genie…
I learned this week - via the excellent Acquired podcast - that even Hermès, that exemplar of handcrafting, makes some use of computers in the production of its silk scarves.