How can generative and data art projects be classified?
Reflections on the difference between various types of art
Since I’m compiling a collection of data arts from around the world, I enjoy exploring their features and analyzing similarities and differences. It’s also fascinating to discuss approaches and practices with other people interested in data art!
During our recent discussion on data art and gen art (generative art), with Ivan Dianov and Adam Arutyunov, who study the adjacent sphere of generative art and teach it at the Higher School of Economics (HSE), we discussed many interesting topics, including how to classify and categorize projects of this kind, comparing them based on various criteria.
Generative design is an iterative design process that generates outputs that meet specified constraints to varying degrees. In a second phase, designers can then provide feedback to the generator that explores the feasible region by selecting preferred outputs or changing input parameters for future iterations. (wiki)
Not the clearest definition in my opinion, so I can put it simply — it’s digital art where algorithms play the main role in the creation process.
And:
Information art, which is also known as informatism or data art, is an art form that is inspired by and principally incorporates data, computer science, information technology, artificial intelligence, and related data-driven fields. (wiki)
Data art is a form of data visualization where the design and aesthetics of the project are important, even if readability of the data is sacrificed. It’s a genre that creates art pieces with data embedded within. Data art projects can be physical (sculptures, embroidery), digital (created with code or vector graphics), or analog (pencils, paints).
The line between generative art and automated data art is very thin. Therefore, in this text, I will focus on art projects created using code, without considering analog or material arts and manually created data arts in vector editors.
Let’s try to identify the characteristics by which we can classify generative art and data art projects
Presence of data inside
(not service ones, like color palette codes, but data and knowledge about a certain subject area). One of the main differences between generative art and data art is the presence of data inside.
However, data can also be present in generative art in one form or another, but there is always a question of their readability; in generative art, data is usually heavily abstracted or distorted, making it very difficult to read. In data art, the presence of data is mandatory, while in generative art, it is not.
The example here will be the project made by Kirell Benzi “The Dark Side and The Light”. I adore Kirell’s projects for their complexity and richness! Despite the absence of legends and labels, in this complex network project, there is data — information about Star Wars characters and their alliances. It’s easy to imagine a similar generative art project where only a beautiful complex network structure is depicted.
An example from generative art — from @_nonfigurative_. When looking at this beautiful project, it creates an impression that there is some meaning to it, some patterns can be guessed. But this is generative art. And there are no data there (it seems).
Determinism
Is it always the case that entering the same conditions into the program yields the same result? Or do we get different results each time we run it? In some projects, algorithms have so much freedom that they can generate thousands of results from the same material.
However, there are projects that produce the same result from the same initial data. Of course, the second type of projects is easier to interpret and often allows one to see the underlying data. Determinism is generally higher in data art projects.
Example: “Floats” — a project by Maxim Gryaznov, Ivan Dianov and friends. It’s generative art that draws a float based on the entered text. Since the text is converted into a numerical sequence, which is then visualized, the same text will always result in the same float. But there are projects where randomly generated numbers are among the variables, adding uncertainty and complex diverse patterns.
Level of control by humans over the project outcome, strictness of rules
The presence of generation in the project — does the machine have the freedom to generate objects or does it only work according to a strict algorithm from humans, using strict rules. This, in fact, is another difference between generative art and automated data art, in generative art, there must be some freedom for the machine to generate something within the conditions.
And this result was not initially specified by humans in a system of strict rules. The rules can be frame-based. For example, the programmer specifies the boundaries and structure, while the machine chooses the shapes and colors.
Also, we can talk about the number of variables given to the computer for generation. In the project, there can be many variables — some of them are given for machine control, some are controlled by humans, some are left to the discretion of the external environment (questions in an interactive test, data generation based on ongoing processes).
An example would be the wonderful data-art project Quarantine Portrait — it creates a unique portrait for you based on a test on the website, but the difference between these portraits is minimal — only a few elements differ, such as the color of circles, the number of dots.
In other words, the entire project is created by a human, and only a small part of it is filled with data, and it does so within strict rules — depending on the selected test answers. I love the idea!
Data-art projects usually have a high level of human control over them, if not absolute.
The degree of automation of the project
(what part was done by humans, and what part was done by the program). This characteristic resembles the previous one but is more commonly applicable to data-art projects and only to a certain extent to generative ones.
Was the project completed manually? Were the data collected manually or automatically? Is the project manually updated and replenished? Were the elements for generation created manually? Was the project drawn manually?
The beautiful and touching project “Send Me Love” by Shirley Wu creates interactive trees based on conversations within the app. This is an incredible and mesmerizing project, one of my favorite data art piece.
The author here does not control the final result, as it depends on external data that changes. So, at some point, the project was “released” to live its own life. The author only sets the initial algorithm and provides symbols used by the project. There is a certain element of generation and freedom for the machine in this project.
In the same Quarantine Portrait project, it’s also released to float freely. The author also doesn’t know precisely what the project generated for each survey participant. However, the rules are very strict, so only different combinations of options set by the author can be obtained.
But there are projects that receive not only a manually made database and generation rules but also the final project is later “finished by hand,” for example, elements are assembled into a certain poster, like the Literary Constellations project by Nicholas Rougeux. Graceful and refined work!
Some generative art projects can also be further refined by the author manually.
Data legibility
Is it possible to return from the project to the data using the legend? In other words, is there a way to decipher the project back to the data? Or is it a one-way cipher, where data, for example, is translated into a numerical sequence, which is visualized, and from which the original dataset cannot be reconstructed? There may be a possibility of partially restoring the original data.
An example could be the data art project by Kimley Scott — MomzWhoViz — with badges of community members composed of their data. This project is done in the Tableau BI package, so it’s semi-automatic. But I really like it, as well as other projects of this kind. I want to replicate something similar for our local community.
But that’s not the point; it’s about the legend. With a clear legend, it is easy to identify which data is represented in each object. In data art, legends are often encountered, while in generative art, they are rare.
Of course, this is more of a characteristic of data art; not only do they contain data, but many data arts can be deciphered according to a strict legend, although not all. In generative art, however, the ability to decipher the project is rare.
The ability to discern meaningful patterns in the data within the final project, the possibility of analytics
The data in the project is not so distorted, and the form in which they are collected preserves the ability to read certain patterns, correlations, from which something can be understood. These conclusions may be modest, but they exist.
An example could be another incredible project by Nicholas Rougeux - “Off the Staff”, where despite the absence of a clear legend, one can discern features of some works — whether many or few instruments are involved, whether long or short notes are prominent.
Not all projects allow for understanding anything about the original data.
Presence of errors
(at which stages of development errors were made and how significant they are). This applies to all types of projects, even analog ones. Errors can arise from rounding numbers, mistakes in data collection, accuracy in rendering values, and so forth. The involvement of humans in the project usually increases the number of errors, which may exceed the errors of any instrument and rounding result.
I cannot recall specific examples here, but even in analog data arts, there is a chance of error. Automation of the project reduces it. The main sources may remain rounding and instrument measurement errors.
But in generative art, there is the possibility of “amplifying” the error, meaning that during generation, it worsens at each stage of the project and with each iteration. Here, I am speaking in general terms and cannot provide specific examples at the moment. I would appreciate your assistance in formulating this thought more precisely.
Degree of data distortion
(“third derivative of the original data”), when the meaning of the data is distorted and lost during their processing. Both data art and generative art are guilty of this. This isn’t inherently bad, but it greatly hinders the meaningfulness of the data, their decipherability, and especially the ability to see patterns and some analytics.
Data can be distorted through averaging, multiplication, truncation, and transformation. The more and the more complex the manipulations, the harder it will be to return to the data from the project’s result, and the less sense there will be in that result.
An interesting example can be seen in the elegant project made by mentioned earlier Nicholas Rougeux — Sonnet Signatures. In this project, a literary work — Shakespeare’s sonnet — receives a unique figure resembling a signature. However, it’s impossible to decipher the original work from the signature or understand anything from it. Despite the meaningful text in the original data, the method of information processing transforms it into unique and beatiful but meaningless symbols.
Well, variables such as the number of letters in a string and the sum of letters in a string divided by their count are used. Letter numbers, let alone the average of their sum, don’t have much meaning. At most, we can judge the lengths of strings and the average letters used, which is quite indistinct and rather meaningless. However, it results in this interesting art project based on a specific algorithm using specific data.
Here are some thoughts that came to mind after our broadcast! I invite everyone interested in such projects or who has their own thoughts on the topic to write in the comments under the article or on social media! I would be glad to discuss this with someone.
This is not yet a finished classification, but rather a collection of reflections on the topic, so please forgive me for the lack of clear definitions and free verbal constructions!
Thanks to all the creators of wonderful data art projects that are so delightful to study and discuss!
Thank you for reading!
Original rus version of the article: here
My Website with data art and comics projects: https://eolay.tilda.ws/en
My LinkedIn: https://www.linkedin.com/in/eolay/
