Aesthetics and Acquine...

The Acquine (Aesthetic Quality Interface Engine) Rating Site allows you to input imagery and see how a computer algorithm rates it in general. How does it do this? What qualities does it hold as important? I put a few sets of images through it to try and suss out what it thinks is 'beautiful' and what it considers bad art.

RATING:  86.9

I started with the cover I created for my book, Barbarians of the Aftermath (cause I'm conceited that way). It scored a respectable rating, as you can see, so I'm inferring that there are a lot of good things about this image from the computer's standpoint. It doesn't specify what exactly it likes, so I went ahead and broke the image down a bit to analyze its methods of discernment.

First I cropped the photo to get rid of the text:

RATING: 58.5

Wow. That's a bit of a drop. Apparently, the words and possibly the black space seemed to have added a great deal to the composition score. Or maybe it doesn't like the background parchment colors. I'll isolate the mask:

RATING: 90.4

Yep. that parchment apparently displeased the AI mightily. But what about background contrast? Let's take out the black:

RATING: 95.2

I think we've topped out here. Contrast between the image and the background is apparently hugely important to the AI's perception of Aesthetic quality. I also put in this image with drop shadow and with the mask alone, scoring 80.4 and 75.9 respectively, so it appears that contrast is king and multiple objects score higher than single ones.

I went ahead and did a few more images, this time from Ian Miller, one of my favorite artists. Let's see how he compares.

RATING: 10.3

RATING: 25.1

 RATING: 4.39

Ouch! Ian scores very poorly indeed in the AI's estimation. Clearly, the complexity of the images has a lot of influence on the rating system, which I believe goes to show that the human appreciation of art is vastly more 'spiritual' than mechanical. Ian Miller is one of those artists you love or hate. I love the intricacies in his work, the myriad small details worked into every line and in the placement of images that create juxtapositions that stimulate the mind in challenging ways. It is the epitome of art for me.

In addition, there are a lot of experiential references in his work, a number of common reference points that I share with him that no computer can count or estimate. His art reminds me of a time when I first got into wargaming as a hobby, when weird fantasy art defined the era I grew up in. It isn't just computer AIs either. Many younger gamers dislike his work for the style, but I believe that a lot of this is because they not only do not share an experiential base with us, but they also have grown up in a different decade where anime and superhero art is their 'art.' 

The same can be said of industrial art, where urban artists appreciate the lines and forms and colors of steal and concrete and rust, whereas I find them repulsive due to the unpleasant rearing I had in the urban ghetto. Again, the computer can't recognize this basic fact, the experiential factor of human art appreciation.

I want to confirm my analysis with one last set of images.

 RATING: 12.3

RATING: 9.8

I added these two pictures to A. try and further suss out some composition parameters, and B. to further illustrate why a computer is rubbish at making Aesthetic judgements.

Clearly, the picture on the bottom has been touched up to brighten the colors and increase the contrast from the original, yet the AI says the original is the better of the two, even though neither rates very highly in general. Why? What is it that the AI likes about the former? The previous examples obviously place some value on high contrast and separation, yet, all other things being equal in these two images composition wise, it is clearly the low contrast image that wins out here. I'm stumped on that one.

But more importantly, this image reflects a number of aesthetic issues that an AI just will not take into account. There is a baby in the image. He is kissing his mother. These things appeal towards women specifically, but even men can appreciate the beauty of the mother son relationship and the fact that this particular mother is a hottie. Ok, I'm biased on that last bit, as that's my wife, but isn't bias a huge part of aesthetic appreciation as well?

In summary, I'm ambivalent about the usefulness of this tool. Outside of measuring the Golden Mean, which is shown, in general, to be aesthetically pleasing, there are too many variables in personal taste to reduce the concept of aesthetic appreciation into some basic formula or mathematical equation. Beauty is in the eye of the the beholder in more ways than one...