I vividly remember the first time I was confounded by Artificial Intelligence. It was 1997 when a computer developed by IBM called “Deep Blue” defeated chess grandmaster Garry Kasparov. Kasparov was considered one of the greatest chess players in history, and if you only knew one name in chess, it was probably his. What scared me a little about the defeat was not that a computer could beat an expert human at a skill that involved a finite number of possible moves, but that in the process of the game, Kasparov himself was unsettled by a particular play wherin Deep Blue was able to determine that it was safer capturing fewer chess pieces in exchange for more theoretical board influence. This calculation was generally considered, at the time, to be the realm of human reasoning and unreachable by algorithmic evaluation. This critical move in game two of the Kasparov-Deep Blue showdown led the grandmaster to accuse the Deep Blue team of having humans assist the computer. This feeling and my, and many others’, reactions to the defeat are now coming up again for artists in the age* of sophisticated AI tools. It is a feeling that resides in the question, “Am I irrelevant if a machine can do what I do convincingly?”
This article is one part of a multi-chapter piece. This first installment offers few, if any, answers and, as such, is structured as an exercise to help people organize their reflections on their place as creators amid AI’s emergence. I am also hoping to organize some of the criticisms of AI, as I have seen many people struggling to stay focused during public debates on its merits. For example, someone might claim that “AI art is not real art,” and when asked why, they might say, “because it is trained on other artists’ work without compensating them.” These are two different questions because artistic merit would exist whether or not the artists were compensated for being used as training data. Whether the art is considered valid would not depend on the compensation model of the company selling the generative AI product, because art is not judged by the economic backend of its creation. Many other similar logical domain errors come up in debates around AI, so let’s look at some of the key points of reflection for artist considering their relevance in the age of commercially available generative Artificial Intelligence tools.
If anyone can do it, why should I?
I have been following public reactions to generative AI art-creation tools for a while now, and I am seeing the same sentiments come up. The first one I want to address is summarized in a post I saw where a composer posted that they honestly had lost all of their drive to make work anymore at this point due to the ability of untrained hobbyists to create convincing musical work without developing any of the skills that this composer had spent decades and considerable expense developing. Behind this sentiment is the assumption that having certain skills makes you special and set apart from others. The idea that a machine can also create convincing work in your domain makes you feel devalued because, in your eyes, it reduces you to a program, which we generally feel superior to. We feel superior to programs because we believe they are automated and we are not; we humans are original, and our work that reflects individual experience is unique to us, due to how novel and unautomated we are, both to each other and to any mechanized process. When we hear convincing work produced by a machine, we feel our value is under threat from both a machine and an untrained user. Before AI tools were available to consumers, skill in a medium was indicative of commitment, and that commitment alone carried some meaning. Now, many creators fear that their commitment no longer carries this meaning because it is not de facto present when a creative work exists. If you find yourself in this scenario, the critical reflection becomes, “If I am in no way special or unique or deserving of being seen as ‘set apart ‘ from anyone due to my skill, why do I create?” Some people have ready answers to this, and some do not. Some might conclude that if they cannot be seen as unique for their skill, they have no reason at all to continue creating. The degree to which this question disturbs someone is the degree to which their sense of the worth of their work is predicated on external, particularly social, validation.
If anyone can do it, is it worth doing?
In the many debates and conversations I’ve had with people about this, a subset of the above question is the marketability of a human skill that AI can replace. Why be a writer if AI can generate a whole book on a topic in a day? If AI can work convincingly and faster than I can, then why waste my time developing a skill for which there is essentially no market? Similarly, people claim to hate AI because it threatens their financial security, as professionals in their field are replaced by AI technology. Feeling apprehensive about job security is, of course, a valid fear. If you find yourself in this scenario, the critical reflection becomes: “If I can not make money from my work, will I still create and why?” As with each of these questions, people will answer differently and with different certainty.
However, if this is someone’s aversion to continuing creative work, then the reflection could orient around how much a motivator economic return is for them. More importantly, in the face of a potential lack of economic viability of your creative output, would you still create, and why?
Did we decide that authenticity doesn’t matter?
One of the things that makes a creative work unique is the artist’s lived experience with the topic matter. With AI music topping charts and AI visual art winning contests, can we still say that any of a work’s intentions reach the listener or viewer? The reason that the success of AI art poses this question is because an AI work will be aggregated from many different works in its training set, so if someone prompts an AI work into being by using a word like “Joyful”, the “Joy” brought about is an average of Joy from many other artists’ stories. This means it cannot be genuinely mined from the promoter’s sense of Joy. The examples I gave above of AI success in charts and competition are one example, but in general, a lot of AI art is receiving high levels of engagement despite the lack of the prompter pulling the work through their own personal story with their own relationship to their medium. If authenticity on the part of the artist/prompter is no longer a key component as to whether or not people find the work ‘successful’, then what is authenticity worth? If you find yourself in this scenario, the critical reflection becomes, “If my authenticity does not guarantee inroads to efficacy for my audience, why should I be authentic?”
If my style or voice can be emulated, what distinguishes me as unique or useful?
Many of the posts I have seen rejecting AI as a legitimate focus on the idea that AI creation platforms extract data from existing artistic works, potentially illegally, on an unprecedented scale. It is also now within the capability of commercial AI platforms to analyse a specific artist’s output for the purposes of recreating its key characteristics. This makes artists feel taken advantage of, especially when decades of hard work developing one’s style can be easily exploited by a novice creator with little to no skill in the art form. That someone with skill in the artform can also do this doesn’t make matters any better. If the hard work of aggregating one’s influences and inspirations to create a unique style does not guarantee exclusive possession of that style, is it worth pursuing? If this is a relevant question for you in the face of the current state of AI tools, then the line of questioning becomes, “If all of the surface characteristics of my personal style can be emulated, why should I develop it?”
Is effort irrational now?
As all of the previous sections show, many long-time artists are scrutinizing how to relate to AI users who can create high-volume portfolios with little skill in an artistic medium outside of prompting or selecting from a variety of work created, originated, or augmented by prompts. This relates to multiple previous sections of this article, but I want to put a specific light on this question because it is not specific to issues arising from the advent of AI tools, but it has been greatly accelerated by it. AI tools have greatly lowered an already low barrier to entry in the arts, and many AI enthusiasts are quick to point out that a difficult or arduous process has never guaranteed, nor been a prerequisite for, generating effective work. By contrast, traditionally trained artists often argue that the friction involved in bringing a work to life brings out its depth and character. It is, however, undeniable that even human creations with great depth can be created spontaneously with little effort or striving. If machine creations can not be assailed or invalidated based on a lack of effort on the part of the user, for a purely artistic line of questioning, can we really invalidate machine-based creations at all? If effort is not, strictly speaking, a requirement, wouldn’t that necessitate us letting skillless AI artists off the hook for the effort required to learn the medium as well? Isn’t this supported by the fact that people with itinerant command of a medium have generated compelling work in their field? If this is a question that is coming up for you, the relevant line of questioning that may yield insight is, “What do I know about the ways that effort in my work translates into actual value for my audience? If I am not comfortable speculating about that, can I articulate a way that effort creates value of any kind at any stage in my work?”
Are we all algorithmic creators?
One of the most common devices I have seen in the AI apologetics camp is the defense of AI training on existing work, often framed as a comparison of an artist’s personal influences to those of an AI model. This argument claims that the way AI models train on existing work is no different from how human artists learn from and are inspired by work they encounter during their studies. For many artists, the scale of AI training alone causes this analogy to ring a bit false, but many are not sure exactly how. Humans can create plagiarized and trope-laden work just like AI, so the fact that a work is human does not absolve it from many of the criticisms leveled at AI-generated work. For the purposes of this question, it is irrelevant that the AI illegally trains on work to create a product because that is a financial concern. It is a valid concern, but it has implications for finances, not the artistic viability of a product. If the AI model is legally trained on works that the authors licensed to the AI model, this question would still be relevant, as it is about algorithmic vs. human navigation of influences. This question of legality does not inform the structural artistic question. There is a compelling argument that human creators steal from their influences in ways very similar to how AI models are trained to create new works in a given style. If we are essentially advanced aggregators of our many diverse influences, what makes our process of absorbing inspiration different from how AI models are trained, from a purely artistic perspective? If this is a question that you are confronted with in your process or in conversation with others, the revealing line of inquiry might be found in the following reflection. “What do I know about my influences and inspirations’ interaction with my lived human experience and how that bears on my work? In what ways is that interaction similar or different to how an algorithm would recombine those same influences? Can I describe in detail how the processes are different?” For this question, it is distractingly easy to try to find an answer by restating the question: “Because my work is real, from real human experience.” This is a given, and responding this way only restates the circumstances and does not constitute a useful answer. Describe the actual process of human assimilation of influences as distinct from AI model aggregation in as much detail as possible, even if the detail is quasi-poetic, abstract, or relies on analogy.
In the next installment, we will look at some possible answers to these questions. Everyone will have different realizations when thinking through these questions, although I expect that for many in the group who decide that, in the age of AI, there is actually very little reason for them to continue, there will be some consensus. There will probably also be some consensus among those who find meticulous reasons to keep creating.