Ur-Modernism

Introduction
I studied Philosophy in college and Art Practice in graduate school, and the genres of Modernism and Post-Modernism were the core of my education. Through both degrees I studied Machine Learning and Robotics. We learned categories for everything, and then we learned how to test the boundaries of those categories. A decade later there is are two terms I’d like to add to this sequence — Un-Modernism and Ur-Modernism. 

I’ll start by defining my terms, then talk about why Machine Learning in particular can benefit from this framework. My terminology represents the best of my understanding and memory from school, but these definitions have taken on more personal meaning over time; I am open to feedback. I’ve chosen three of my favorite practitioners of each genre as an illustration, not as a definitive ranking.

 

Modernism
Naming objects and phenomena is what Science is all about. The brain likes pattern recognition — it’s how we form an internally consistent and externally coherent understanding of the world. Modernism as a genre in Philosophy, Art, and Machine Learning is a strategy of categorizing and indexing information. As an artist I like to think of myself as part of the Hudson River School, with their Modernist practice of meticulously studying pigments and materials.

Immanuel Kant - Built a system of epistemological categorization

Anne Truitt - Gave individual pigments and gestures physical form

W.E.B. DuBois - Visualized otherwise hidden sociological data

 

Post-Modernism
Any categorical system will have its flaws. Because the brain likes patterns it’s easy to lean too hard on them, and important to challenge those patterns frequently. There is some overlap in the motivations and creations of the two genres, and most practitioners oscillate between them. Intention is what I consider the defining factor — Modernists are contributing to a shared system of knowledge, and Post-Modernists are challenging it.

Ernst Bloch - Explored Utopia as a construct, both failed and possible

Coral Woodbury - Added women to a male-dominant canon

Kehinde Wiley - Added black people to a white-dominant canon

 

Un-Modernism
I believe there is a third strategy to incorporate here, and that is to cluster rather than categorize — this is what I am labeling Un-Modernism. I’m not sure exactly what this looks like in Philosophy and Art, but I know what it looks like in Machine Learning.

 

Un-Modernism in Machine Learning
The way machines learn is by training on datasets. The programmer tells the machine, “intake these data, find patterns, and report back.” The programmer evaluates that report, and fine-tunes the machine’s data analyzing engine, getting more and more meaningful reports in return. 

Problems of bias arise quickly in all Modernist systems — categorizations encode the biases of their creators. Racism and sexism are the most insidious and common examples, and there is a constant stream of articles published to confirm:

Machine Learning has a strategy to mitigate this — clustering. It’s a way of finding commonality among objects in a complex dataset, without sorting everything into strict categories. There are two algorithms in particular, K-Means Clustering and Principal Component Analysis, that I like. I’ll discuss those in another post.

To detach the conversation from emotion as much as possible, consider baking as an example. Where is the line between bread and cake, or muffin and scone? 

Let’s say we want to build a baking robot. We give it a pantry full of items, and a set of recipes:
Ingredients = [ water cups, milk cups, flour cups, salt tsps, sugar cups, yeast, baking powder tbsps, vanilla tsps, eggs, butter tbsps]
Bread = [ 1, 3, 1, 0, 3, 0, 0]
Cookies = [1, 1, 1, 8, 1, 1, 1]
etc.

We’d then define some functions for combining and cooking those ingredients. This system works well enough for the basics, but what happens when things get more complex? How do you account for a biscuit, which is so regionally specific? Clustering allows you to drop the strict categorizations and understand recipes globally, detached from linguistic hangups. It defines things relationally, like, “this recipe for naan is 80% similar to this recipe for fry bread.”

 

Ur-Modernism
Ur-Modernism is the cycle of categorizing, deconstruction, and clustering. In Philosophy it’s about reading old books until you have enough background to read contemporary scholarship, then writing scholarship of your own. In Art it’s about understanding the raw materials available and the sensory experiences their combinations provide. In Machine Learning it’s about generating comprehensive data, writing elegant algorithms, and making graceful recommendations.

 

Conclusion
Language is descriptive not prescriptive, and that is important for all genres of Modernism to remember. When we teach machines about ourselves linguistically, we impart all of our bad linguistic habits. Machines are best at analyzing numbers, animals are best at analyzing sensations, humans are best at analyzing language. All the genres of Modernism try to balance these aspects, and it’s a good opportunity to leverage the concept of the Harawayan Cyborg.

2021American CyborgbluebirdComment