The Evolution of the Remix: Mechanical Reproduction to Machine Learning
One of the first times I heard the word “remix” spoken was in the 1988 rap song “Serious” with Steady B and featuring KRS-One who says,
Believe it or not. This is a remix. THIS IS A REMIX…
Rap songs like this one ushered in what is commonly referred to as the “golden era of hip hop,” which covered the late 1980s to the early-mid 1990s, particularly from artists and producers living on the East Coast. As an art student in Brooklyn, NY during that time I was really into collage, another form of remixing using paint, glue and paper and later expanding to include the use of computer software such as Adobe Photoshop.
Note: I majored in visual art in high school and moved over to computer graphics (art) in college and grad school.
Romare Bearden made a practice of copying, redrawing, and reworking or remixing his images. He used Photostat machines and film projectors to make large-scale photo reproductions. Inspired by Bearden’s work, I started experimenting with collage around the time hip hip was entering it’s golden age. At the time, we (creators) were overwhelmed with what Samuel R. Delany called a “river of stuff” (see my previous post). Additionally, the ‘art world’ had rejected modernism’s rigid rules and shifted to an ‘anything goes’ approach to making art.
There are several characteristics which lend art to being postmodern; these include remixing (formerly bricolage), the use of text as the central artistic element, collage, simplification, appropriation (and reappropriation), performance art (improvisation), the recycling of past styles and themes in a modern-day context. Most important was the removal of barriers between high and low art.
In his 2008 book Remix, Lawrence Lessig described the emergence of a subculture that allows and encourages the creation of derivative works by combining existing materials. Remix culture was a response to the Industrial information economy that strengthened exclusive copyrights and weakened creativity and innovation in the public domain. One-third of my book Techno-Vernacular Creativity & Innovation is dedicated to remixing from the perspective of groups that are often (historically) overlooked. I was exploring conceptual and computational remixing through the development and use of block-based software.
By the 2010s another form of remixing was on the horizon, powered by artificial intelligence or AI. Artist-scholar Sofian Audry notes the “profound consequence” of AI-powered software being able to “photoshop” content. The is made possible by machine (deep) learning or generative AI, which is a subset of ML. Audry writes,
[M]achine learning algorithms are able to automatically transform generative processes into (models) that can be stored, copied, and modified. These processes can then be used to generate new, unforeseen media content such as images and sounds… — Sofian Audry, Art in the Age of Machine Learning
Audry’s description of remixing AI-generated content can be compared and contrasted with remixing in hip hop and in the type of art Romare Bearden was creating in the 20th century, especially using machines to assist him in copying, redrawing, and reworking images. The machines of Bearden’s time has become the Dall Es and Midjourneys of our time. There’s no right or wrong way to go about remixing; sometimes an artist/producer will just add something new on top of existing content, and in other cases they might change the arrangement and structure altogether.
For example, I scanned one of my drawings from the late 1980s (see above) and used Midjourney’s/describe command to upload the image and generate four possible prompts based on that image. I chose one of the prompts and ran the generator, along with my original image for reference. What resulted is a remix of the original image. I can also apply the modifier words in the prompt to generate new images.
I can remove the bridge (and original image reference) entirely and reuse the modifier words in the prompt to generate entirely new images. I just replaced the subject but kept everything else. This kind of remixing happens at the prompt level (prompt engineering).
We’ve come a long way from using Photostat machines! I can now add on to the upscaled image (above) with more ideas (see below). What remains are the aesthetic elements from my original drawing (Brooklyn Bridge), plus any new subject matter or composition descriptors I add into the prompt. I can also use AI-powered tools such as Deep Dream Generator and Adobe Photoshop (yes, PS is now AI) to edit the new images.
In summary: I started with an analog drawing I made when I was 18 yrs old, then I used it to create a portrait in Midjourney and, as a final step, I remixed the portrait to add a random basket of puppies (robots, dolls, etc.). It’s a different way of working, for sure, but this is what can happen in the multidimensional latent space of machine learning or generative AI.
The ability provided by deep learning systems to automate the creation of a content-generation program is unprecedented in human history at least to such a scale and with such flexibility.
I think for me, as someone who went through the ‘prescribed’ art route (high school, undergrad and grad school), the use of AI-powered tools to make things is a brave but less risky step compared to new learners. I sat for hours in art studios and computer labs learning how to use traditional tools and practice techniques such as collage. These experiences gave me a vocabulary (of adjectives) that I now apply to the AI-remixing process through prompt engineering.
For me, GenAI is just another remix.