Raw LLM Responses

From someone who actually understands neural nets, deep learning, and transformer block architecture (which means I outpace about 99% of “AI Artists”) the term doesn’t even hold from the AI/mechanistic side of the argument. 

AI artists “create” images. “Create” is doing a very specific kind of rhetorical smuggling. When AI bros say “I created this image”, they’re usually trying to import the whole human-art package without even knowing it. To be an artist and create something, that has historically meant authorship (expression-of-mind), skill (developed capacities-over-time), control (intentionality), responsibility (authorship + intentionality; can explain decisions and reproduce conditions), and originality (novel representation, not an interpolation of a training distribution). 

In most “AI artist” usage, create silently collapses into the much thinner claim of: “I caused an output to appear by giving instruction.”

They have to mean that, because if they actually think the model is creating an image in the same kind of way, let’s say, a painter creates a painting? Mechanistically, let’s see what they’re committing themselves to.:

Essentially, the minimum architecture they’re describing is a system that would need stable long term learning, or plasticity without catastrophic forgetting. Something like persistent intention across revisions and time. Internal goal maintenance, error correction, and self directed iteration. All things that would be additions to current cutting edge architecture like learned world/model representations + a planning/critique loop + an execution/rendering module. Take this to absolutely any AI researcher on Earth and they will tell you “Yup, that’s the minimal architecture to instantiate an AI that could reasonably be considered to have created a work of art on its own.” 

What I described is not “a diffusion model responding to conditioning” (GenAI/CNN architecture), it’s the current holy grail of AI research. Something thousands of PhDs, the combined compute capable of simulating a black hole accretion disk in real time, and trillions of dollars haven't been able to figure out yet. If this AI artist framing were literally true, it would be one of the biggest empirical facts in cognitive science and AI engineering. 

So either, they’re accidentally making a MASSIVE claim about cognition and agency, or
they’re using “artist” and “create” in a causal sense (press button —> results), while borrowing the prestige of the authorial sense.