Translation: Criticism of 'new' generative art

As for how to evaluate generative art (not market, not visual aesthetics, but more meta), I think it is a very difficult thing. When I was sorting out the generative art data files, there was always a lack of relevant documents in this area. Casey Reas, the founder of Processing, wrote a book that was translated into Chinese, and then he also established his own generative art NFT curation platform Feral File, but in addition There is no public information in Chinese. After all, this is an emerging field, and few art critics have entered (or my ability to collect literature is limited), and critics often start from relevant practitioners (this may require confirmation from STS Technology and Social Studies).

So I started crawling through the critiques of various creators from Twitter to see if I could find the most current professional opinions. When it comes to the sharpest artist, none other than Karsten Schmidt. He is nicknamed Toxi by the community, because he is tit-for-tat with collectors, which has caused the collector community to deal with it. He slipped all the way from the top of the market favorites and vowed to leave the generative art NFT market (although he's coming back soon!).

Because it is very beautiful, I translated the tweet string he published in January this year, and added a few remarks, hoping that readers in the Chinese world will taste it together. Let's take a look at how Toxi criticizes generative art (see comments for the original tweet):

I want to criticize the generative art and related comments that I have missed in recent years. The recent generative art I call "new" generative art. Kolah's (editor's note: the generative art media that Toxi critiqued) commented not on generative art at most levels, but more broadly on any traditional or static visual art, such as color, texture, style, composition, context, story. How to evaluate the aesthetics of generative art, there are more important things than the visual appearance.

From a technical point of view, reading code can analyze generative art, such as difficulty, consumption performance, simplicity, and legibility, but this only involves the analysis of "code", not the "generation" level. The problem is that most people including me, especially now, simply refer to "generating via a computer using any means" as "generating". On the contrary, in the past, we have also distinguished parameterization technology, stepwise technology and generative technology, and then used "computer computing" as a general term.

When we criticize this kind of work, it is very important to at least be aware of and try to understand the difference between the above concepts. In most more complex projects or works, these concepts often form conceptual layers rather than separate principles. I won't explain the difference between procedural, parameter, generative here (Wikipedia is your friend!), but I think any serious criticism should be Look at these concepts as a framework before moving on to visual judging.

As mentioned earlier, most computer art and computer design with a certain level of sophistication include the above levels. For example, all works on Fxhash have parameters (code: parameter, or fearure feature), these parameters will get a special cluster from a unique hash value (seed code) (code: this cluster is a Zhang Generative Art NFT). Parameters define a scope, and the more parameters there are, the more scopes (sometimes referred to as Searching Spaces, although for the vast majority of works, the artist does not search through these scopes). The range of these parameters can be very large, one parameter is one dimension, and ten parameters are ten-dimensional space!

Putting a lot of parameters in a work doesn't directly make it a "better" or "complex" work. In fact this may reveal that you are a rookie, since too large a range will be harder to deal with, even with machine learning techniques. Less is more or something...probably better.

Rather than using afterthoughts, using storytelling to rationalize the creative process afterwards, to understand the parameters chosen by the artist, and the roles and interactions between them, can provide a deeper understanding of the artist’s (or designer’s) thinking process. Of course, if the artist is not using his brain, it is easy to be discovered. But everyone seemed to prefer listening to stories.

What does it mean that the parameters are too complex to handle? This will be a pure state, free from artificial, artistic prejudice, generative art is free to explore any dimension, its only limitation is time. However, most "generative artists" artificially input and constrain the system, such as injecting some "subjective" form into the system. Technically, these interferences are biases or hotspots in the parameter range... . (Editor's note: Hotspots are controllable random works that are more favored by the market or subjectively favored by artists, which is contradictory enough.)

Some hot spots occur naturally, but in many cases, the creator must explore randomly or systematically to discover pearls (edit: commonly known as dice works), and then induce by using rules, exclusions, and conditions, such as adjusting the parameters between Interrelationships (Editor's note: Iterating by Humans). Creating/curating these artificial glimpses requires skill and patience.

The reason why curating parameters is important for most (not all) works is that a system without constraints, especially one without human tweaking, creates a lot of aesthetic noise/garbage (i.e. "non-artist certification" output).

In fact, criticizing works of art using computing as a medium is itself a problem worthy of attention. The artist's intention is divided into two directions. Is it to use only the advantages of automation to create a processed product that conforms to human aesthetics, or to use the advantages of parametric design to generate an automated aesthetic with N-dimensional possibilities, which is both primitive and chaotic? Another question is, do we focus on complexity, treating these parametric generative works of art as entities with "latent energy" and celebrating them from the perspective of the viewer/collector? Or do you value the artist's artificial filtering/direction of this energy, maximizing what the artist considers a good work through the process of deletion/selection?

A related problem is the recent popularity of "rarity" and output (i.e. number of editions), especially NFT art.

A few years ago, before NFTs existed, the choice, scope, detail, level of subtlety (!) of parameters was relatively free and often individually curated by the artist, with the goal of conveying some personal perspective on the parameter system. I have been exploring computational design, computational technology (not necessarily art) since the early 1990s, and my first digital printing project without human intervention was in 2007 with a book cover for Faber & Faber. Similar projects were made in subsequent years, each with a single and strong goal. And recent NFT generative art seems to push artists toward a larger, more diverse goal, often driven by NFT collectors.

In a new platform like Fxhash without any human curation, there is a random seed code system. In order to present a clear and unique artistic concept, I think we must maximize, cover the entire range of parameters, and minimize the output of human bias.

This conundrum and conflict of interest is made even more apparent by our discussion of edition size: collectors prefer small editions to maximize price, while demanding that each edition be "unique". On the other hand, we have a mathematical reality that even if the seed codes on which the entire design system depends are evenly distributed, a considerable number of samples (i.e. minting NFTs) are usually required to explore and guarantee various parameter combinations.

This conundrum results in that instead of spending time on interesting computations or generative processes, artists are now forced to spend more time adjusting to curation problems, since casting is an automatic system. Whereas previously they could manually tweak the hash for diversity in a fraction of the time.

I must say, that's why we see so many derivatives in the current wave of "generative art" (editor's note: imitation disks), imitated aesthetics, imitated objects. Too many people are exploring the aesthetics and foundations (often very simple programs) generated by the same computational process, only to end up with just different artificial styles/manual curation conditions.

Back to how to critique generative art (or critique the platforms that publish it). I would like to see more descriptions of parametric qualities and parametric relationships in the work. None of this is well documented right now for audiences to really appreciate, and the platforms don't care too much. The 1024 word description of the work restricts the artist to delve into any details (Editor's note: fxhash is limited to 1024 words for published works). To be honest, after four publications (Editor's note: Toxi is about to publish a fourth), I wonder if many people really care about these arts other than speculative value.

It is important to understand whether the techniques used in a work are procedural (eg templated) or generative (eg based on some rules, or exhibit some unexpected state, aesthetics), because only by knowing, Ability is a fundamental element (or at least it should be) of critiques of works.

Generative processes like Cellular Automata or Diffusion-limited Aggregation have certain inherent qualities and limitations that can be easily spotted by a trained human. When evaluating works completed using these techniques, the focus of the commentary should be on the context, how these techniques are used by the artist in the work, or abused into the work.

The biggest problem right now is that collectors and reviewers lack the ability to read computer science or algorithms. I see this question every day: the current collecting community often doesn't know or care about the context of these artworks, even thinking that the same technical style didn't exist before Fidenza or similar.

The rise of NFTs will slowly put an end to open source art, as many people (myself included) are reluctant to provide source code for other people's possible derivative works (fake disks) for profit... . Again, I'm going to accuse collectors of a general lack of computer literacy, because these derivative works (fake disks) aren't even identified. Perhaps when the hype and fanaticism ebbs again, a new, more established model will reemerge…. There's a lot more to say on this subject (I will!), but let's save that for later.

(Finish)

Original tweet: https://twitter.com/toxi/status/1481954101437116419

Toxi work: fxhash — toxi profile