A/B Testing Art – Why Mondrian Was a Great Optimizer | Hacker News Reader

A/B Testing Art – Why Mondrian Was a Great Optimizer

blog.eyequant.com

36 pointsfab1an7 comments

If you're in New York, the Met currently has an exhibit on Matisse's process of making multiple versions of the same painting: classic A/B tester :)

http://www.metmuseum.org/exhibitions/listings/2012/matisse?u...

Artists Komar and Melamid created a 1995 project called "Most Wanted and Least Wanted Paintings" paintings (and, in another project, songs) based on their surveys of artistic preferences vary from different countries.

Their paintings "reflect the artists' interpretation of a professional market research survey about aesthetic preferences and taste in painting. Intending to discover what a true "people's art" would look like, the artists ... expanded their market research to more than a dozen countries around the globe and in turn, created Most Wanted and Least Wanted paintings for each country."

Curiously, Holland's most and least wanted preferences are nearly the reverse of the other countries!

* The paintings: http://awp.diaart.org/km/painting.html (The United States' Most Wanted Painting's thumbnail is broken, but the linked picture works.)

* The survey data: http://awp.diaart.org/km/surveyresults.html

Nice read, thanks! It made me more hungry to learn about artistic processes.

I have to say though, as a plug to a service that helps improve your website presentation and flow, in the context of that blog page, to me it's a big turn off. They talk about a master of visual flow, and then present it in that visual mess of a blog. A link to the .com can't even be found around the top left and instead must be dug up from a cloud of menus on the right, basic visual hierarchy point (point that actually makes a short appearance in the actual article) that so many company blogs get wrong but you'd expect web design experts to get right. I'm not saying their tech isn't good, I have no idea, but I find it hard not to think that if it was good, they'd have figured out how to use it to improve their own promotion site.

I re-read myself and got uncomfortable with how negative my comment sounds. Even though it sort of ended up sounding like I wrote it to be dismissive of other's work, I didn't, I thought my experience would be useful to others as a constructive point. Next time I should express myself with more tact.

I'm a bit conflicted about this post. It's very interesting, but it misuses the term A/B testing.

More later, if I get time.

Here's more:

One of the key insights behind A/B testing is the realisation that it's only what happens with real customers that counts. It doesn't matter what you like, the CEO likes, or your dog likes -- if your customers don't like your latest design/copy/whatever it doesn't count for sh*t. This is the opposite to what the artists mentioned did -- mainly relying on their own judgement before displaying a finished work to the public. The fact they were good at consistently guessing what would be appreciated (let's leave aside defining the metric of success for art) is a large part of what makes them a good artist.

This is a moderately pedantic point, and shouldn't detract from what is to me the most interesting point of the article -- that we can correlate "good" art with objective measurements. A bunch of questions immediately arise:

Does this result generalise beyond Mondrian's work? His work is particularly well suited to the type of experiment described. I imagine fiddling with a hyper-realistic painting in a meaningful way would be harder. Are there consistent similarities in gaze patterns between good art and bad art?

Can we generate a useful predictive model of eye movements? That is, given an image can we predict where the eye will go?

Does this generalise to other domains, like effective landing pages? Could eye movements be used to predict purchasing behaviour?

The latter two seem like the core assumptions that EyeQuant is built on. More depth here would be good.

Thanks Noel, EyeQuant co-founder here - (not the author, though)

Yes, the concept of an A/B test is used fairly liberally in our post and should probably not be interpreted in the exact same way for someone looking to run an actual A/B test on their website - though one hopefully gets the general idea. The article is just trying to point out that some of the most well known pieces of art were the result of an iterative optimization process that can be somewhat "reverse engineered" by creating alternative versions and measuring people's psychophysical response to those. And if this helps some companies out there to embrace both creativity and analytics to make their own websites better, all the better!

Reply on news.ycombinator.com