Artificial Intelligence Art 2019

Last fall, we saw the first auction of AI generated art: Portrait of Edmond de Belamy. It sold at Christie’s for over $430,000, almost 45x it’s estimated valuation. Artificial intelligence (AI) is a broad branch of computer science covering many different types of technologies. However, in the context of creating art, the technology AI artists are typically referring to is GAN, generative adversarial networks. Developing and balancing this type of machine learning system is notoriously difficult, but once achieved, they can produce an infinite number of high quality creative works.

With several AI exhibits and sales since, here are a few art highlights that use the same GAN technology.

Memories of Passersby I — Mario Klingemann (2019)

This installation, which sold for £40,000 at Sotherby’s London back in March, features two screens with ever-evolving AI-generated portraits, one with a more feminine style and one with a more masculine style. It’s interesting how these GANs conceptualize the human portrait, especially the subtle differences it renders between the male and female portraits. Watch the artist’s video here.

nimiia cétiï — Jenna Sutela (2018)

Nimiia cétiï is different from most AI-generated art being created today in that it incorporates AI-generated speech and an AI-generated ‘martian’ handwritten language. The audio-visual installation’s handwriting movement is derived from the movement patterns of bacteria in a petri dish and its speech derived from recordings taken by the artists. The work is a surreal and eerily beautiful example of un-human creation; machines, fueled by organic and human-generated data, creating art. Sutela released an accompanying book: nimiia ïzinibimi of the generated language earlier this year. Watch the artist’s video here.

Machine Hallucination — Refik Anadol (2019)

As the first installation in the newly opened Artechouse, New York City, Refik Anadol immerses viewers with several AI-generated works of art. The GAN was trained on 100 million publicly sourced images of New York City. The result is a representation of a dynamic city architecture that grows, morphs and evolves, much like the city itself. Watch the artist video here.

What are generative adversarial networks (GANs)?

They are a sub-speciality of machine learning where two ‘deep learning’ neural network algorithms are trained to understand the concepts in the example data they have been fed. Training examples can range thousands to millions of images of a particular category, such as 19th century portraiture, video, text or audio. The first algorithm learns concepts from these training inputs and re-constructs unique works based on the elements and aspects of the training data, typically elements that it has seen most often. This generated work is then given feedback by a second algorithm which has learned independently it’s own concept of an acceptable output. The two networks work together until the creator/artist feels the outputs meet their quality standard.

Originally published on artluxe.co on December 10th 2019