Tides of Information Flow: Visualizing Our Digitally-Mediated Reality

I consume a lot of news, especially recently.

But often I feel like I have a hard time grasping the shape of it. What is different this time? What am I missing? What have I learned? Where did that other thing go?

I’m also interested in maps.

So, I built this:

Or if you freeze frame and zoom in:

Or if you prefer, darker:

Kaleidoscope is an experiment in realtime, abstract cartographic representation of current events. Each element is a news item or trending hash tag, with spatial nearness, visual design, and overall layout determined by semantic similarity dimensions of the content.

The question I want to explore is:

What does the sum total of information I consume look like?

More precisely: How do I develop my ability to comprehend the virtual spaces we increasingly come to inhabit? What do my patterns of media content and other virtual resource consumption look like? How does this affect me? How is it different today than yesterday? And beyond that, if I zoom out what does the territory look like? What is its macrostructure, the shape of the information artifact we are building and inhabiting?

“New research suggests use of digital platforms such as tablets and laptops for reading may make you more inclined to focus on concrete details rather than interpreting information more abstractly.”— This means that when you read things on a tablet or laptop you really only pay attention to the objective points of the material instead of making more complex connections.

The conclusions drawn from this study are probably overreaching, but if true—even to a small degree—indicates to me that we need a better understanding of abstract thinking in digital spaces.

Cartography of Digital Spaces

The visual languages of architecture, cartography and geography (a la Jacques Bertin’s Semiology of Graphics) give us a broad set of tools to map physical space.

Can we take these tried and true cartographic principles and apply them to more abstract virtual spaces? Spaces made entirely out of information, e.g. the evolving landscape of online media, flamewars on reddit, or my internet browsing habits?

This is a bit tricky. For example, the principle abstraction of equiareal representation of the sphere is no longer valid as the information on the internet does not have a useful physical dimension — or rather, a useful heuristic for physical dimension must first be discovered.

In order to synthesize raw media data flows into a form of visual abstraction we need to develop new cartographic primitives that can work automatically without the need for constant human input. This suggests a purely mechanical approach to digital cartography, mediated by machines, that is nevertheless capable of addressing the fundamental issues of cartographic precision:

• How well can the map represent the territory?
• What ground must be given?
• How can we accommodate or embrace biases in the human visual system to better represent the “truth” ?

Introducing Kaleidoscope

Kaleidoscope is an automated system for mapping and exploring information spaces. It consists of three components:

1. A flow of raw information with some underlying driving principle (e.g. the news cycle, topicality, or the interests and desires of a particular person as perceived through the lens of their browsing habits);
2. Sufficient natural language processing and other ML primitives to find patterns in this data, and
3. Somewhat new ways of thinking about cartography and visual representation to render those patterns understandable to humans.

In its current incarnation, kaleidoscope.news uses a neural-network based content embedding to map the most salient textual and structural data features into a shared low-dimensional vector space.

The spatial nearness structure evolves over time as new content is added and older content drops off, with a meandering evolutionary path between birth and death. Thus, new topics bubble up over time, become dominant and die out, and the whole cycle can be made visible over longer time-horizons.

Tools, Inspirations, and Debts of Gratitude

I could fill several more blog posts on how this was built, but briefly I’m incredibly indebted to the developers of: tensorfow, fasttext, mapnik, mapbox-gl-js, tilelive, tippecanoe, postgis, and Google cloud platform. If anyone on any of these teams is reading this, hmu I will gladly spot you a beer.

Design-wise Kaleidoscope draws most heavily on Jacques Bertin’s Semiology of Graphics, the awesome cartographers at Stamen design, and the Ethiopian-born artist Julie Mehretu, whose “story maps of no location” depict “an imagined, rather than actual reality.”

Equally compelling and inspirational are the purely generative artwork of inconvergent.net, and the deep exploration of visual information design of the Dear Data project.

Then and now

As international trade developed in the ancient world there arose a need to pool together and abstract cartographic knowledge.

In a similar vein, as we become increasingly reliant on the intricate and interconnected world of realtime digital media, we now more than ever need to bring more cognitive power to bear on the problem of understanding the world, and better maps of the space we’re moving into.

I can’t claim to solve this problem — or even push the state of the art forward much — but kaleidoscope does give a lens into a particular visually intuitive view of the mass media landscape.

If we view ourselves from a great height, it is frightening to realize how little we know about our species, our purpose and our end.” — W.G. Sebald, The Rings of Saturn