Because I follow so many creatives (fine artists, photographers, writers and designers) on social media, I see a trend in the conversation around AI that is steeped in the fear that this technology has been tailored to steal their creativity.

I can truly see why they might think that. The AI stories that dominate social media feeds are all about the most “creative” and visually striking outputs that various models are capable of: Midjourney creating a realistic fashion model that clothing designers use to sell their products, Sora generating nearly instant cinema-quality video with special effects and believable “performances,” and entire sequel novels written using existing works (like the novels of George R. R. Martin) as the source material.

Among artists and non-artists alike there is a fear that our creative outlets are being overtaken by a faceless robotic legion hellbent on replacing not our menial labor, but our most precious and meaningful forms of expression.

Understanding what AI can and cannot do

One of the first tools to dispel fear is to understand the general purpose and limitations of “artificial intelligence.”

A very common example used to explain how popular AI models work (particularly Large Language Models like Chat GPT) is to reference the predictive text programs we have on our phone. As we type or text, this program is looking at our previous data and rapidly offering a series of “guesses” as to which word might come next.

If I type the letter “s” and then the letter “t,” an algorithm begins to predict words based on the most commonly used “st” words in the English language, along with a comparison against my most commonly used “st” words.

It might offer words like “stop,” or “steam,” but (given my personal habits) it might likely offer “Star Wars” or “Stratascale” because those are “st” words I more commonly type. The algorithm is predictive in that it uses a body of information (the most common “st” words typed by all English-speaking users) and compares that against fine tuning data (the most common “st” words that I personally type) in order to offer predictions of the outcome.

LLMs work in a nearly identical way, looking through your prompts as a user to compare what you’ve written against the sum total of internet data it was “trained” on, and comparing that against the wording of your specific prompt to make educated predictions about what the next likely word in the response should be.

This means that an artificial intelligence isn’t being creative per se, but rather attempting to mathematically predict the kind of creativity that you are asking it to reproduce, and therein lies the crucial difference.

Unlimited limitations

When we ask an AI like Chat GPT or Midjourney to make us a creative deliverable, it can only make us something that is a mathematical representation of the best similar creative output it’s ever seen. In the case of creating an image, the model is never going to produce an image better than the best image it’s ever evaluated. It may be able to infinitely create variations (based on the math) of what it has seen before, but it cannot create ex nihilo the way a human mind seems to be capable of.

In truth, human intelligence doesn’t produce creativity “out of nothing” either. There is a component of our cognitive process often referred to as the “subconscious” or “unconscious” that seems to absorb and internalize data that serves as the inspiration for creativity without our even being aware of it.

But where the human subconscious passively stores data it is exposed to, and later connects that data to the process of creative output, AI models are capable of no such passive inference. Every piece of data an AI is exposed to is part of a conscious decision on the part of a human being. There is no “creative spark” without the deliberate effort of a person, and that is both limiting to what an AI is capable of, and reassuring

Malicious mimicry

So if an AI is incapable of creative output without human instruction or intervention, where is the concern for artists an creatives coming from?

Recently a number of artists have begun to discover that the developers of image-generation models have used their work explicitly to train the AI. This means that a user who wants to create a work of art can say something like “create an illustration in the style of Mark Roma” and the AI might generate an image that so fully mimics that style that it is mathematically calculating the elements of my artistic creativity common to my body of work that I’m not even consciously aware of. Where my style of an artist is the sum total of my lived experiences and training, the AI can create and mimic the pattern of those experiences very believably.

This is  understandably an ethical concern in the industry of building AI models. Artificial intelligence engineers should never use data (i.e. images) to train an AI without the express consent of those images’ creators. Unfortunately the “cat is out of the bag” so to speak, in that Chat GPT and similar models have been trained on the sum total of internet data available at the time of their inception. Which means that, without my express consent, an AI may have seen my posted images anyways, and created a mathematical analysis that would allow it to mimic my work if asked.

To further add to these concerns, recent terms of service language from big companies like Meta and Adobe explicitly state that a user’s personal output will be used to train an AI, without explaining what the AI will do with that data (or be allowed to do.)

To my mind, this is clearly something that forms the basis of an issue that needs to be tested in the courts. Just as internet streaming was (and still is) testing the legal limits around how actors and writers are compensated for work that can be endlessly played digitally, AI must have legal limits around how it can be trained and with what data in terms of violating individual and corporate intellectual property laws. This is a discussion of “when” those cases will be tried, not “if.”

I think it’s fair to say there is reason for concern around the ethical and legal use of our creativity to produce similar (but clearly inferior) “knock off” creative output. But given the technical limitations of how AI functions, it’s safe to say that (for the time being at least) this technology isn’t here to steal our creative essence. It cannot have the ability to live some sort of artificial life of leisure and creativity while we slave away at our 9 to 5 daily grind. Every AI output is the result of human effort, and it will take even more human output to police that work and ensure that our expression is safeguarded from theft.