I studied photography and sculpture, and initially the only thing that made any difference to whether I enjoyed a piece of art was interpretation by someone who's seen a lot of art. There were only two or three people who could do this effectively for me. Eventually, after talking to them a lot and seeing a lot of art, and thinking about it deeply, you start to develop your own tastes, and realise what kinds of things speak to you.

This project seems under the misapprehension that there's some kind of "perennial philosophy" of art that can be uncovered by asking enough people about randomly generated images. For one thing, this is absolutely not what I (and, I would argue, pretty much anyone who's spent time thinking about it) consider to be "art". Also, the experience of art is not just pretty pictures, but all of the past experiences that you bring with you when looking at a piece of art. Even people who like what I would consider to be twee pictures do so because of what they're bringing to the aesthetic experience.

If statistically significant results come out of this, I'd argue that any information it generates is going to be extremely shallow. It's like a kind of statistical fetish.

I don't see what this could possibly tell us about art. The generated images are both too random, but also have absolutely no character or variation.

There are so many brilliant sources of understanding on what makes art appealing, and so many images of art on the internet. I don't get why the site would use these drawn-in-ten-seconds-in-mspaint images for voting.

Probably a good place to promote https://electricsheep.org/ which is a "crowdsourced evolving art" which has been running for easily 20 years.

The "sheep" (https://electricsheep.org/#/sheep) show what this idea can evolve into.

I don't think the algorithms produce good art. Aside from that, I think changing the rating system from an emotion-based one to something more neutral like numbers would be better. Art that makes you feel "bad" can be good art.

I like the idea but the way the pictures are generated seems way too obvious. It's clear that there is a small set of "patterns" with randomized parameters. So it becomes like "do people like the water drops in the corner or in the center? red or black?". Might be more interesting if the generated images were a little more sophisticated and varied.

"In 1994 the artists Komar & Melamid commissioned a public-research polling firm to conduct The People’s Choice, the first poll on artistic taste in the United States. Individuals were asked approximately 100 questions on a variety of subjects, ranging from their consumer tastes and recreational activities to their knowledge of famous artists, and their preference for angles, curves, brushstrokes, and particular colors, sizes, content, and style in painting. As Russian émigrés, the artists were intrigued by the idea of the consumer-research poll as an outgrowth of American democracy. At the same time, their interest in democracy led the artists to ask what a genuine people’s art would look like. What is a democratic and populist painting?"

https://curatorsintl.org/exhibitions/the-peoples-choice

Maybe its part of the research methodology but I can't help but feel they cut the "art" a bit short here.

What if the generated art was more something like this?

http://color-wander.surge.sh/

Low-contrast images will always be unanimously down-voted (for good reason: us humans can't even see the art clearly). Wonder why the algorithm doesn't use a sane heuristic for picking color palettes.

There might be a way to create algorithmic art that has meaning. But looking at these images, I am not feeling much. They look too much like other graphics I have seen already.

These seem very repetitive, same pattern. Gets boring very quick

If anyone is interested in understanding more about art and human perception of art, then this book is as good a starting place as you'll find: https://en.wikipedia.org/wiki/The_Story_of_Art

As the book says, ″There really is no such thing as Art. There are only artists.″

Every image it generates is one of 8 “pieces”, where each “piece” is an algorithm that has some randomness but a predetermined basic design. [1]

My impression is that the instantiations of each “piece” tend to all look the same and have similar aesthetic value. Personally I like Nested Squares, 45 Degree Paths, and Bezier Curves when it chooses rainbow colors. On the other hand, Overlapping Drops is fundamentally ugly (looks like a 5-year-old using a paint program’s stamp tool) and Patterned Lines is usually ugly (too many lines, too much contrast, no antialiasing, reminds me of the Windows pipes screensaver). But I feel like both the good and bad come mostly from the human who wrote the program, not the random generation itself.

[1] https://artvote.net/about

In this case, I strongly prefer comparing two graphics and deciding which one is better (and creating Elo-like score).

For me, it is hard to judge these pieces of art in the vacuum.

To those saying that this experiment is useless and provides zero information: Check out the results page.

https://artvote.net/results

There are clear tendencies there regarding colours, contrast and complexity, and they do provide some (limited) insight into what viewers expect in art.

Simplistic. Starts with lowest common denominator (monochrome nothingburger) and progresses from there too slowly.

This is too random to be meaningful for "art". But it might be able to build some stats on design layout choices... and if it does so, comparing its results to the general design philosophies built over the years could be an interesting exercise.

Am I the only one who thinks the input data is feeding the generative art engine of someone's generator for an upcoming NFT collection?

Where is the art?

this should use vqgan and clip