Guephren's Writing

[EDIT: This are simply my thoughts, I'm not an expert on this subject by any stretch of the imagination. I simply wanted to lay down my thoughts somewhere.]

Visual artist, you are a runner– your time training has given you skills, techniques, endurance– and now you are pitted against a race car.

“AI Art” (from here on out called Synthetic Imagery) has made rounds lately as several services have popped up, and people are up in arms, some in favor and some against. The problem with it is not that this technology is inherently negative, but that the way it's currently constructed is unethical, and some people want to turn their eyes away from that.

Some argue that this is the same thing that happened with photography. Painters panicked, they felt that they would be replaced by this new tool but fortunately, their worst fears didn't come to pass. Now it was easier to get a portrait, yet portrait painting never went away, nor did landscapes, or any other number of subjects you can also capture with a camera.

What is the difference between photography and Synthetic Imagery right now? Photography matured into a medium of expression in itself, while Synthetic Imagery currently aims to replace an existing one. Currently, users write prompts with the specific names of artists from which they want to pull a “style”, and the machine provides approximately what the user asks from its latent space.

Where did the machine get that information from in the first place? From the internet, of course! From the hundreds of millions of images on the internet that people have uploaded throughout the years, many of which are private or copyrighted. Taking photos, drawings, and paintings, and repurposing them for something else is an old practice on the internet. It was bad enough when individuals did it (sometimes even after the artist asked not to do it, even when they shouldn't have had to) but the extent of the databases used for ML training is magnitudes larger.

Training in and of itself is not inherently a bad thing– the sourcing is. Let's be honest: Synthetic Imagery can already produce convincing and appealing images, the problem resides in that to be able to synthesize such images the machine relies on training data that no one agreed to. Artists who share their copyrighted images didn't agree to this, and it can be argued that not even works under Creative Commons rules count as those were written before Synthetic Imagery existed.

Weirdly enough, people don't want to admit to the problematic origin of the training data. It is comfortable: all you need now is a few words to get more or less what you wanted. Yet, Ethically Sourcing materials is already something that some companies strive to do– from chocolate to minerals, there are ways to mitigate the damage to both the environment and communities. The same should be done with Synthetic Imagery.

The saddest part is when people turn to “adapt or die”, or worse, “if you didn't want your art stolen you shouldn't have uploaded it to the internet”. The latter puts the onus on the artists themselves to protect their art because the internet can't help itself, which is just asking for creative people to no longer share that which brought emotions to themselves and others. The former asks you to race against a car.

As an artist –be it a painter, draftsman, musician, sculptor, writer, etc.– you work hard to increase your skills. It takes time and effort, and much like a runner, you train to run better and further. Synthetic Imagery is a tool, and in this scenario, it can be seen as a race car. The runner and the car can compete in a race, but it's unfair. Of course, the race car wins, but saying that you are better at running because you used a car is not true. You used a car. You used prompts. You won the race. You got an image. It has its merits but it's simply not the same.

As a tool, Machine Learning (and Synthetic Imagery) is an amazing technological achievement that can be useful. There is even an untapped market in making and providing original, high-quality material, specifically made to train models. This would give work to people and make Synthetic Imagery ethical… at the high cost of sourcing material in this manner and waiting the time that would be needed to create it.

The question is: will the ones behind Machine Learning accept these ethical limitations, or will the lure of unchecked growth, novelty, and “progress”, enable them to dismiss the voices of whom they are building their technology on top of and barrel forward?

As of April 9th, 2023 #MachineLearning #AI #ML