Editors note: recently I was browsing through the usual Facebook groups when I came across this post by Joseph Turmes. It’s too well-written, too unique perspective, to stay buried in a random FB thread forever. I knew right away that Jospeh needed a bigger platform to get his nuanced point of view out to the world. We decided to publish it here anyway…
Artificial Intelligence in Heavy Metal
by Joe Turmes
“Anything that is in the world when you’re born is normal and ordinary and is just a natural part of the way the world works. Anything that’s invented between when you’re fifteen and thirty-five is new and exciting and revolutionary and you can probably get a career in it. Anything invented after you’re thirty-five is against the natural order of things.”
-Douglas Adams
Recently, we have seen a lot of controversy around the use of Artificial Intelligence in heavy metal, with the new High on Fire video, the Deicide album cover, and even Unleash The Archers attempting to ethically source AI models for their new video.
I want to spend some time talking about the state of AI today and where I think we are going. I am not attempting to discredit anyone’s thoughts or opinions on the matter, and I think the debate is very important. There are serious risks to consider, however I think effectively mitigating the risks requires better understanding of the issues at play.
I think I have an interesting perspective because I have worked in the tech field for nearly 25 years, and I am currently an engineer for a large semiconductor manufacturer who is likely going to make a lot of money on the AI boom. I am also an artist. I am a musician that plays in a band, as well as a photographer and videographer. My time making art is just as important to me as my career.
Generally there is not a whole lot of overlap between these two communities, and as such I think there is a lot of misunderstanding, especially when a new piece of tech comes out that has potential for a huge impact in the art world. Personally, I feel like my experience in both the tech and the art world has been very beneficial for both my art and my career. Exposure to the creative side has enabled me to approach engineering problems in very unconventional ways, and my engineering experience also impacts how I approach making art.
Two Communities in Conflict
I think one of the big differences between these communities is how their work is perceived; some see their work as a product, while others see it as a process. While not always cut and dry, engineering culture is often focused on the final product; anything that can be automated or streamlined to accelerate getting to the final product is a very good thing. If anything, the process that an Engineer values is being clever about making their process more efficient. Artists on the other hand, tend to value the process of making art, not only as a means to create their product, but by internalizing the process such that the process itself is deeply ingrained into the experience for the artist, greatly influencing the final product.
I saw a video recently of Rick Rubin, stating “We don’t make these things for an outcome. We are making the best thing we can make. It is a devotional practice, and whatever happens happens.” Streamlining and automating this process disconnects the artist from the process required to make meaningful art. Artists want to spend more time making their art, not less.
Of course, this is all only relatable to the artists’ chosen discipline. While there are great examples of bands making their art a truly multimedia experience with well thought out album covers, music videos, liners and inserts like Tool, Yob, Bongripper and Mastodon, there are many more bands that are singularly focused on the music itself and are happy to automate or streamline every other aspect of being in a band.
A Core Misunderstanding
At its core, AI offers artists and engineers alike opportunities to streamline their work; however many people in both communities fundamentally misunderstand what AI can and cant do today. I have seen both vastly overestimate what AI can accomplish, at times with surprisingly poor results. While AI solves certain problems quicker, it requires a lot more attention to detail in the output.
For example, if you tell AI to write code for you, the resulting script may at a glance produce acceptable results. The engineer however has removed themselves from the development process, and is no longer intimately aware of all the contents in the script or may not understand why AI chose certain methods over others.
This disconnects the engineer from the process, and while the initial results may look good, in depth review of the output becomes much more important to ensure that the script is actually doing what you want. A lot of engineers are going to get tripped up by this in the coming months as they implement AI into their work flows. I predict we will see news reports of AI generated mistakes getting missed and causing serious problems.
In the news, we hear of companies axing entire teams of graphic artists, coders, copywriters with the expectation that AI will be able to fully take over those positions. I predict that those companies, lured with the promise of AI cheaply replacing labor, are going to be regretting those decisions as their quality flounders, while companies that intelligently augment their current staff with AI will flourish.
For some artists, the assumption sometimes becomes that all one has to do is ask a general question or submit a couple sentences and an AI engine will do all of the work for you, completely removing the need for the artist entirely, except to provide the initial prompt. This is also not the case.
Much like in the early 2000s, where “Google Fu” became an important skill that separated some skilled workers from others, AI Prompt Engineering is going to be the same way. Those that learn how to work the prompts in the various models will have greater levels of success in generating an output that matches the user’s intent. Currently, there is no level of Prompt Engineering that can generate every possible artistic output, which I will go into detail about shortly.
Diffuse Derivatives
Another aspect of AI that is going to impact both artists and engineers is the fact that AI is inherently derivative. It is excellent at compiling known facts and previously solved problems and presenting the user when asked, and for art, it is limited by the art that it is trained on. AI is always backwards looking, and is poor at extrapolating future trends and finding truly novel solutions.
This also means that how it generates art is always going to be strongly guided by the models that are available to it, and likely weighted towards what is more common, meaning creating truly interesting and niche content will always be more challenging. I have found this to be true when I was learning how to use Stable Diffusion.
Stable Diffusion is an artistic AI generator that is locally run and generated based on model profiles you load, that are created by other users. Because of this, the output generated follows a market dynamic; where there is demand, models are created, and users generate content. This means currently, there are lots of models for anime characters and photorealistic women, but if you want something specific, like say animal biology or a niche art style, you are not going to have a lot of success.
That being said, being a derivative art form is not necessarily good or bad, as there are plenty of examples of derivative art being beloved (Converge’s Jane Doe album cover being a prime example).
The derivative nature of AI has of course been equated to theft by many artists. This is an important debate to be had, but in order to classify AI as stealing would require a complete rewrite of Fair Use laws, which in my opinion, would be disastrous. On a conceptual level, it would be hard to hold AI to a standard that is consistent to the one we hold humans to. Humans are already allowed to chop up, modify, satirize, parody and sample other works of art and be protected under Fair Use, regardless if the original artist approves.
Additionally, style and technique are generally aspects of art that are not covered by copyright, and is generally not treated as theft (Imagine if Meshuggah had been able to copyright the entire genre of Djent). Humans do this at a granularity that is a small fraction as complex as an AI model. If an AI is trained on millions of pieces of art, how do you determine what share of an AI art piece is consisted of any single previous piece of art?
What this means is that for most bands seeking to use AI as an affordable means of producing album art and music videos, the art itself is going to be perceived as lazy, on top of the negative perception of the uncanny valley affect that AI produces.
Two Paths of AI
Going forward, however, I predict AI will follow two additional paths that have been established by previous tools. The first path is one of less notability; the AI tools will become more subtle, such that they become an assistive tool for artists to achieve their vision without their hallmarks that make them distasteful. HDR photography and early CGI both have followed this path; overused, ugly and off-putting in their origins, but nearly unnoticed now when in the hands of capable artists.
The second path is one of achieving cultural significance. There will come a day when a groundbreaking artist creates a piece of AI art that becomes significant enough to give the entire field credibility. Think of the next generation’s Banksy or Andy Warhol. This artist will likely be invested in the artistic process as well, not only with Prompt Engineering but custom designing their own models as well.
Artists that were around before the emergence of this artist will grind their teeth about their changing world and wonder what is wrong with the kids, while artists that come after will be inspired, join the cause and expand the legitimacy of AI as an art form. Bands that used AI art for album covers and videos will have been considered lazy before, but after it will be considered a matter of pride to enlist the efforts of well known AI artists.
-Jospeh Turmes
One thought on “Artificial Intelligence in Heavy Metal”