Humans-in-Loop for Generative Art
The article on interactive AI systems and video on experimental creative writing both explored the fascinating world of music, art, and literature in the context of AI and algorithms. Previously, I had encountered some work involving AI generated music and art, but this was the first time I had encountered generative literature. The conversation on robot musicians was espeically interesting to me because it caused me the reflect more deeply on the uses and purposes of these technologies. Just because it is possible to generate art or writing, it does not mean they inherently have meaning. The human-in-loop design thus becomes especially important.
When enjoying art or music, there are two levels I typically enjoy it from. One is from the surface level of purely asking myself if I like how the painting looks or if I like how the song sounds. The other is taking into account the context and the humanness of the work. For example, knowing the history of the Impressionist movement and Monet’s life when looking at Impression, Sunrise leads to me to better appreciate the work; knowing the backstory of “All Too Well (10 Minute Version) (Taylor’s Version) (From the Vault)” leads me to enjoy listening to it just a little more. Purely AI generated works lose out in this aspect of the human connection. Instead of connecting to the human behind the work, I might instead appreciate the beauty of the algorithm that generated the art, which is not necessarily a better or worse but just a different type of enjoyment.
The technology for AI to generate works without human intervention doesn’t seem to be mature enough yet. Some generated works aren’t sensible out-of-the-box without some adjustments and others just lead to some hilarious results like this example of OpenAI's Jukebox on “Country Roads.” Currently, some human interaction can greatly improve these generated works. For example, in the video on experimental creative writing, the user can toggle the amount of balance desired between the two excerpts to create the final mixed output. The generative art by @idflood with geometric cities and blocks are some of my favorites, and though it is generated, the final output is still tweaked and adjusted by a human to look just right.
Geneartive art by @idflood
Someone once told me to “never bet against technology,” and that has proven to be true in so many cases with modern technological advancements. It’s possible that one day, I may only listen to generated music by some virtual robot musician for enjoyment. However, for now, humans-in-loop is critical – the human touch is still an important factor for these generated works.