I loved Halo 3 as a kid. I especially liked Forge Mode where you would play as a little flying ball and build levels using game assets. You could alter the world geometry and also the game rules. This led to a lot of incredible games, such as Grffball which inspired hit games later like Rocket League. One day, Halo 4 came out and all my friends thought Halo 4 forge mode was the bee’s knees, but I never felt the same way. In Halo 3, physics was an inevitable force to contend with. There was a money system to prevent the asset count from going ballistic which really constrained your options. Your structures had to be really structurally sound or else they’d collapse. A floating platform was a masterwork of glitching the physics system for hours and cancelling out forces leveraging overflows and floating point errors. It was a sight to behold. In Halo 4, you could just turn off physics for any object, clip it through walls and float it anywhere.
When I see AI art, it feels like Halo 4 forge maps to me. The frontier for skill has shifted, and the costly signals of competence have migrated elsewhere. Its like Minecraft builds in creative mode.
In Halo 3, a colossal build that defies gravity itself was a feat and an indication that you were in the presence of real mastery. Come the early days of Halo 4, these structures are what clogged the sewers at the bottom of Halo Waypoint’s barrel. The frontier for mastery shifted, and we didn’t yet know where to. It took some time for new masters to emerge. These masters would redefine what a great forge map is and push the potential for gameplay to a whole new level, but it would never look quite like the pinnacle of what people would do back in Halo 3.
💡 In a world where there is more content than we could ever hope to consume such as the one in which we live, we need a useful way to discern the gems from the rubbish. We need a way to search for mastery and what we are missing with AI is a distinct playbook of what mastery looks like. No one mourns the day shallow depth of field went from an indication of high budget and high taste to a gauche excess of the Leica necklace gang.
I am not sure what the signal for mastery will be in the age of AI art, but I think I might have a guess.
After attending many ComfyUI meetups in NYC, I got to hear from some of the coolest AI artists in the biz about their workflows. One who’s ideas stuck with me is Calvin Herbst . His work using real world footage from this library to curate the looks of AI films and images really spoke to me. He knows the language of the lens and of the emulsion and forces the model to conform to his ideas in ways that words never could. Can you succinctly describe to me how the look of a Sigma 35mm/1.4 differs from that of the Zeiss CP.2? Or describe the look of Ektachrome in a way that an AI model could replicate? Can current models understand that the background should read 5 stops below the subject in this set of shots, while being a stop above in this other set? Calvin does this through LORAs which lock in a specific look and yield incredible results.
So perhaps the skill of the DP remains the same, and the masterful choices we use to shape light are the same, and the new costly (in the time, as well as the money sense) skills is how precisely you can coax the model towards your vision. The difference between a good enough result and the result in your mind.
And remember, some of the most incredible Minecraft builds are made in vanilla survival mode.
I used to think that the piece with a banana taped to a wall was an affront to art, and that they’d better replace all those modern art galleries with something like those old German paintings where you could count the hairs on the head of a guy in the corner window of a building. I don’t feel that way anymore; the banana is one of my favorite works of art now. What changed?
I think I felt this aversion to art-without-craft because I spent a great deal of time in my life perfecting forms of craftsmanship, be that in the discipline of photography or creative coding, or visual design. I probably just didn’t that like the banana taped to the wall was in a fancy gallery while I’d poured a great deal of effort into works which never were. I think I was missing the point.
I’ve always admired contrarian or subversive people, actions, things, and yet I failed to identify the subversion in works like the banana. At times, I’d thought that if I think something is stupid, I ought to be able to parody it; to make some art which highlights why exactly it’s stupid. When thinking about how I might do this for the banana on the wall, I realized that the banana was that subversive and contrarian work which is making fun of the “poorly crafted” and money laundered aspect of modern art. It wasn’t about the fact that no one else could accomplish such a work of craftsmanship, but that it’s very installation and existence was a form of meta-art.
With this principle in mind, what is left for me to sneer at as lesser art, or non-art?
So we’ve established some kind of a defense of art without craftsmanship, but these concepts still remain ill-defined. I don’t think I’ll be able to define art, but I can definitely define craftsmanship and it’s relation to art.
Craftsmanship is the tangible quality of work a person puts into a process which yields some kind of an artifact. This could be learning to paint, code, photograph, or really anything at all. The common thread is the perfection of the craftsman’s ability to render their medium into precisely what they have in mind. I contend that a craftsman’s skill is measured by their ability to transform a conception of a work in their mind into their medium.
Does this mean that a great artist must necessarily be a great craftsman? I don’t think so. If we define craft as the process of creation, what’s left is the concepts themselves. It’s those ideas which are the domain of the artist. If an artist has a vision which they wish to bring to life, they are very likely to become a skilled craftsman to be able to achieve that vision. However, they might not have to. Humans are lazy, and through great creativity we often find easier ways of going about hard problems. Perhaps an artist is anyone who has something to express, and goes about expressing it.
💡 Maybe this is why so much creative work can be draining. When you sign up to be an artist, so much of the reason you do it is to express your own vision. It’s really hard to get someone to pay you to do that. So, often creatives turn to expressing other people’s artistic ideas; and trading on their ability to translate often abstract artistic ideas into more tangible works.
AI. AI based artistic tools specifically. These can take many forms. [comfyui pic], [nano banana pic]. They may be as simple as generating an image given text, or as intricate as a big web of entangled steps as complicated as a sophisticated visual effects pipeline.
Within the framework of art and craft outlined here, which part could this generative AI based tooling affect?
Craft This is the easy question. There is zero doubt that the world of craft, especially in regards to digital art, is already significantly disrupted by current generative AI tools. It’s a lot harder to get photoshop work now that Gemini Image can do a good enough job at a minimal marginal cost.
If you know what you want to create and can articulate it, you might not need to hire someone or learn the manual process yourself anymore.
Art, The Translation
Not yet, but maybe.
I wouldn’t consider the business of translating abstract intent to a concrete description of an artistic creation to be particularly insulated from AI disruption. LLMs have demonstrated excellent ability to generate outputs which appeal to humans [link to article about AI sycophancy], and their ability to transform semantic content from one space to another has also proven strong. Do the abilities to predict how people will perceive something, and to translate from one semantic space to another seem useful to have when figuring out how to achieve a piece of artistic expression?
Art, The Ideas
Does it matter what “ideas” an LLM or otherwise an AI “has”?
There’s a question of what extent one should personify an AI model. If we don’t personify it, we could say that any artistic idea expressed by something like chat GPT is a reflection of some artistic idea expressed in its training data. To get this idea, we need to somehow sample the outputs of this LLM. The choice of how to conduct this sampling, however, is itself a valid artistic expression. The choice of a highly neutral and unbiased sample, too, is an artistic one. Perhaps we want to know what chatGPT “really thinks” and we analyze it’s inner workings. This process itself could be considered an artistic study of the model’s encoded biases.
I contend that no way in which you choose to sample a generative model, even a highly automated one would be able to escape the defintition of an artistic thought.
What if you’re an AI researcher on the brink of unleashing an AGI with a human-like ability to form ideas? I think art made by this AGI would have to be interpreted similarly to how a devout religious scholar interprets human art. If we believe in a divine creator, what attribution does this creator get for the art his creations create? Do the works of the creator’s creations have merit equal to works of the creator? I’m no religious scholar, but I am confident that it’s a similar kind of question.
I am not as confident in this AGI situation happening as I am with the above. And if it does, it doesn’t directly follow that the artistic works of this AGI would do anything to diminish the value of human art. Creating a new work of art doesn’t diminish the value or other art. It’s not a zero sum game. Perhaps this AGI wouldn’t be much of a craftsman. Maybe it’d even choose to hire humans for that.
💡 But what about being paid for creative work? Is the kind of work an AI could do work you really want to be doing? To me, doing something one could use AI to do instead would feel rather menial. Like doing math by hand when you can just use a calculator. Just use the damn calculator.
Do you think that’s a bad comparison because art isn’t like math; that there’s no right answer? I agree, but I also contend that if you’re doing the kind of work which isn’t more or less a process of executing on instructions of a client to a “correct” (read: client approved) deliverable, you’ve got nothing to worry about.