Radiance Blog

In a collaboration that spanned five decades and three countries, artists Komar and Melamid explored such deep and disparate concepts as elephant painting; preserving Soviet-era Socialist Realist monuments; and an opera featuring George Washington, Lenin, and Marcel Duchamp.

And in two projects in the 1990s, they created art with surveys!

First, they commissioned telephone surveys in fourteen countries about the different aspects of paintings that people liked and disliked. Using that information, they then created “Most Wanted” and “Least Wanted” paintings for each country. There are some remarkable similarities across countries, with most preferring realistic landscape scenes, with culturally unique touches (like in China, Kenya, and the USA). Most people across the different countries disliked abstract art, with the major exception of Holland whose citizens loved abstract art (and bicycles, but that’s a whole other story).

Komar and Melamid then used a (less-representative) survey of 500 visitors to their website to find out what qualities of music people like and dislike. They asked separate questions about features like themes, instruments, genres, and pacing. They then took their results and (with composer Dave Soldier) created People’s Choice Music: two songs, one which has all of the qualities a majority of their respondents liked and another that has all the qualities that respondents didn’t like!

Strangely, I find I like the “bad” song better than the good song! Despite the fact that the bad song is 25 minutes long (and features children singing about Labor Day) it seems more interesting to me than the good song, which sounds like every bad ballad from the 1980s. Does this make me an outlier? Maybe. The artists wanted to make a statement about what art would look like if it was explicitly designed by polling (similar to how many believe our politics are mediated through polls, which is a question for a different day). But it also reveals the limitations of creating a product or service by testing each attribute individually rather than testing them all together (which is why conjoint modeling is such an exciting technique). Sometimes you can’t get a good picture (even of an elephant) by testing different features one by one—you have to look at the whole scene together.