Ben Bloch, “A lush forest, fall colors, green waters, do not include any sky,” 2024, Acrylic on Canvas, 40.5 x 41.5 in.

There are two populations avidly discussing Artificial Intelligence, or AI, of late. One is the group that created it and promotes it while anticipating soon becoming rich, or at least finally making some money. The other is the rest of us, who have heard a lot about it and have, for better or worse, built an image in our minds of what AI is and how it works. A problem arises when the two groups try to talk to each other. Engineers are notoriously bad at explaining their highly technical work to laypersons, who may use computers without much understanding. Listen to the former try to explain a term like “output modulation” to the latter, saying “that’s when you ask the computer to modulate the output” and you have a typical, less-than-helpful exchange. This is about where Ben Bloch, a landscape artist who divides his time between Utah and Montana, steps in to show how AI works as an artistic tool. Make Me a Landscape, Do Not Include Any Sky is his step-by-step demonstration of precisely how it works.

Three things are needed to clarify how AI works in practice. One is a body of functional examples that a computer can scrape. This rather abrasive term refers to the computer’s uploading of a working archive of examples of the desired medium, such as stories, videos, or, here, landscape paintings. This is also where the problems begin, since no one wants to see their work a) taken without compensation and b) made into an example to be exploited by uncounted competitors. Here Bloch gets around that problem by scraping some of his own works, a book of which is the first item on display close to Finch Lane’s entrance. Two of these paintings are in the same corner of the gallery. In the accompanying texts, he describes them as “created from direct experience, photographs, and memory.” A dozen more works follow in clockwise fashion around the room. They were also painted by Bloch, but based on images produced by a computer that was “trained” on the artist’s pre-existing works. Before getting to them, though, a video presents the steps that the computer and the artist may be said to collaborate on. Without knowing any of that up front, the well-informed Utah audience would likely conclude that everything in the room is the work of a single process, so much do the AI babies resemble their metaphorical parents.

A central term for understanding and using AI, and one that sheds a surprising light on traditional art as well, is “prompt.” In computer speak, a prompt is a command, often couched in the form of a request, that tells the machine what the operator wants it to do. Bloch has clarified its role in the process by using his prompts as titles for the images—properly called models—the computer then produced. “A high mountain lake with mountains in the background reflecting the morning sky,” for example. At this point, it may become clear how the prompt, which initiates the task, is sometimes like the title of a finished work. Such a title may convey the inspiration or intention that found form in the final work, and by which the artist then tells the audience what the work means to contain or convey. In the video, Bloch first shows the prompt page, then the resulting image, or model, followed by the time-lapse, condensed stages of the actual painting; and then concludes with the model and the final work together.

Installation view of Make Me a Landscape, Do Not Include Any Sky, with time-lapse videos demonstrating the progression from text prompt to AI model to finished painting.

Here we see side-by-side images of a ruffed grouse in winter. The image on the left is an image generated by an AI model trained on uploaded images of the artist’s work and prompted by text input from the artist. On the right is Ben Bloch’s painted interpretation of the AI-generated image.

Although the paintings and their AI-assisted sources initially look alike, with a little study, differences do appear. Ben Bloch knows his work far better than I, and when I spoke to him, he corrected my errant comments about different handling of details. Keep in mind that he assisted the computer in turning its designs into finished work—that is, he created a painting from the AI-generated image. Yet there remain diffuse differences between the well-defined originals and the less certain AI versions. Artists may spend years learning to know a deliberate gesture from a tentative, or even exploratory passage, and how to employ the former instead of the latter. No computer is going to understand this.

There is today a newly-minted field of expertise in detecting the difference between genuine efforts and artificial ones, whether they are created by Large Language Models or diffusion models. The first, usually used with text, is linear and predictive—determining what word will statistically follow the last. Meanwhile, image generators begin with visual noise and iteratively refine it until they arrive at an image. In either case, and as Ben Bloch generously demonstrates, when it comes to AI, what you get is essentially what you’ve already seen.

Ben Bloch, “A vibrant colorful sky at sunset,” 2025, Acrylic and charcoal on canvas, 42.5 x 42.5 in.

Ben Bloch: Make Me a Landscape, Do Not Include Any Sky, Finch Lane Gallery, Salt Lake City, through February 20. Opening Reception & Salt Lake Gallery Stroll: Fri, Jan 16, 6-9pm.

Geoff Wichert

Geoff Wichert objects to the term critic. He would rather be thought of as a advocate on behalf of those he writes about.