When looking at a piece of art, we often assume that the artist is at the helm of their own creative operation – that the artwork is a product of their total control. While that can be true, there are exceptions, such as art that relies on chance, or a form of external interrogation; in the case of Eli Muro’s latest series, Interpretations, artistic control is sacrificed to the machine.

For the six works in the series, Muro started by compiling a large dataset of art texts including exhibition reviews and essays from various publications; they then loaded this information onto an unsupervised language-generating system that, when given specific parameters, generates coherent paragraphs of text. This system is so intelligent that it can, in many ways, outperform the human by utilizing language beyond the given dataset; in other words, this system can think for itself – about art, anyway. Using this system, Muro generated fully-articulate, randomized art descriptions of yet-to-be-made work; Interpretations brings six of these descriptions into the three-dimensional realm.

Following the complex, ridiculous, and occasionally offensive descriptive sentences, Muro made works as requested by the machine. The work is strange and uncanny. It looks familiar and yet, completely otherworldly. Is this what the machine desires? If artificial intelligence and autonomous machines are the way of tomorrow, can Muro’s series be considered a futuristic archive?

General Concepts:

Theres alot of talk about AI taking jobs and while thats a real legit threat there are some areas that it will be tough for machines to take over. Recently an AI generated work sold for half a million dollars, raising questions about the value of a human artists and where the limitations of AI art actually are.

I think generally AI art is pretty boring. It seems more of a demo of what a computer can do. Things like, "look at how realistic this looks" or "look at how detailed this is" and it took the computer a couple seconds to create. These things are of less interest to me than things like "What kind of ideas can an AI come up with? " or " How can I collaborate with a machine?"

This AI actually is using the work of many artists and art writers as well as it is trained from actual peoples writing. This raises questions of authorship, bias, value and labour. Even though it is crude at best it is effective even though it is trained on a small amount of data. One could easily imagine a more succinct AI capable of replacing a human.

Outputs to Create

Model outputs


jargon_dataset_big.txt