I oppose the use of AI in any argument to replace an artist or craftsman. Artificial Intelligence is only a tool and cannot create art. It’s the human who uses it that is the artist, not the tool or the medium. Art is a form of communication, and therefore, any Generative Artificial Intelligence cannot produce art of any kind, because it does not communicate. Many people are unfortunately under the impression that these tools are replacements for human effort and will. The world is obsessed with production quantity and speed for profit, but art by necessity cannot be manufactured.

First, a very brief overview on how AI works for the layman. Generative Artificial Intelligence is a highly complex mathematical equation, trained using massive amounts of data, to create a function that matches a trend-line from all of that data. The machines that run AI’s algorithm don’t generate a data point; they input a value into a function and produce a result on the trend-line. Essentially, it's like auto-complete on your phone’s keyboard, which guesses what your next word might be. It could guess the next pixel, the next note in a song, or the next paragraph of text, and it has no understanding whether it is wrong or right.

When these AI programs are constructed, they first gather terabytes of annotated data. The general rule is that 80% of the data is used to train the algorithm and 20% is used to confirm that it works as intended. There’s a fairly famous example of an AI that incorrectly identified a canine in a photo as a wolf, despite being a husky. They discovered the AI was actually looking for ‘snow’ in the image. The AI did not ‘think’, it produced a result, and it was not always a helpful one.

AI itself isn’t bad, but it’s evil-adjacent. The data used to train these massive models must be both varied (without bias) and annotated by human effort. A human must describe images and music before the algorithm can digest them, because otherwise, it has no correlation to the prompts. All of this ‘meta-data’ is how humans group and classify this data, however imperfectly. All current substantial AI models today incorporate works in their algorithm without compensating the artists who generated the originals. In the vast majority of cases, they were not even asked for permissions or licenses. 

The actions of these companies are evil-adjacent because art isn’t just intellectual property; it has cultural meaning and is the livelihood of creatives. The sale and licensing of artworks by artists is how they feed themselves and their families. Professional artists make a living from these licenses and permissions. Without any monetary incentive, the number of artists will dwindle to almost nothing, much like in the Dark Ages. As I write this, Disney and other companies have filed suit against a major AI company because a significant portion of their intellectual property was compromised, and now anyone, including counterfeiters and plagiarists, can produce Disney-esque characters with the touch of a button. We’ve seen it happen before, citing the Dark Ages, when only nobility, governments, and religious institutions could afford to hire or become artists.

Art, at its core, is a form of communication. If an artist were alone and drew in the dirt, that is a communication, either to themselves or to someone who would see that dirt-drawing. While I believe art isn’t complete until it's observed, the artist does observe their artwork after its creation. Thus, an artist constructs a message within themselves, consciously or unconsciously, and then chooses a medium to transmit that message. Practiced artists consider what the message means to the intended audience, and adjust the artwork to communicate more clearly. Ducamp had intended to communicate about what can and cannot be ‘art’ when he produced his artwork ‘Fountain’ by signing his name to a urinal. Here, we see the construction of a message and its transmission to an observer, the people ‘defining’ art in his day, and us now as the people who study that message.

AI cannot produce art, because it cannot have a thought. It can only export a file, convert text to an image or more text. Much like a paintbrush can take in paint and produce color on a canvas. A brush is a tool, and so is AI. New artists will gain skills with this tool and, no doubt, produce artwork with it. A common misconception is that AI will produce new images, even if you send in the same text repeatedly, mimicking original thought and variation. But this is an illusion. What happens in these engines is that there’s a random number added to the prompt, sometimes called a ‘salt’ or ‘seed’, to ensure that results are differentiated. In some of these engines, the random number can be locked in, and repeated prompts generate remarkably similar, if not identical, data. AI does not ‘create’ anything; it merely guesses the next 0 and 1 in the language of computers. It does not perceive at all, let alone perceive as a human would.

Some people who are creating ‘artworks’ using this tool are unfortunately viewing it as a replacement for all of the practice and training it takes to produce great artwork. I believe that well-practiced artists can and will leverage AI into their workflows to eliminate repetitive work or utilize it as a valuable tool within their reach. New and young artists will initially see the products of using this tool and have early success. There’s some benefit to this, as it can motivate improvement. However, these new artists will inevitably have a new or original thought, and will quickly discover that the AI engine does not support creating this type of artwork. For example, if ‘A Rainbow-feathered bird with a city atop each wing’ were placed into a prompt, any images would be unsuitable, and none would match the artist’s first intent. The artist’s original thought need not be completely original. We are humans, and our thoughts tend toward patterns and trends. Still, these human patterns are a myriad of material, mixing and matching in an endless number of permutations that no computer could replicate. The training data of an AI restricts what it can do, by definition. The trend-line of the AI’s mathematical graph will always force the result toward the most common points in its data, given similar prompts. The number of images on the internet involving Shrek, for example, means that AI can generate fairly well-constructed image that involve the word ‘Shrek’ if we desire, but the ‘Bird of Two Cities’ image will be wildly inconsistent if not impossible to generate.

In Summary, AI is a wonderful tool, but it has been developed in bad faith by for-profit companies with stolen human effort. While those companies will have to atone for their actions, it doesn’t put the AI genie back in the bottle. It cannot produce art, no matter how anthropomorphized this auto-complete program is. It is not a replacement for skill or practice, because a paint-brush is not a replacement for skill or practice. As artists and humans, we need to see through the techno-babble at the heart of the hype and realize the truth: AI cannot replace the artist. Not even to produce as much as possible to drive a profit. The value in art, both culturally and in its price tag, is based on the messages from human to human.