Double Hamster on The Rocks

I’m surprised by the number of shear lines— regions in the air where curtains of blowing snow lie directly on top of each other and pull in opposite directions. They always occur during transition points. The snow, looping about in the middle of the street, shifts direction and blows back on itself before smoothing out and standardizing: straight down in the middle of the street, rising and rolling near the surface of the houses and frozen, floating at points in the middle. I perceive a neutral pattern that gets pushed out of equilibrium only to return but I know that this isn’t really the case. The snow blows about in a cyclic fashion and it’s the cycle of back and forth that is maintained in equilibrium.

I'm reminded that the human brain is remarkable for its pattern recognition capacity. If we see the first five steps in a process, we will predict the sixth.

Claude Shannon, an engineer at AT&T, had a great insight into pattern recognition while trying to understand the static in wire-based telephone signals. It was important to predict this static in order to smooth out the data that was sent across each wireline. He surmised that it was better to think about the static patterns in terms of probability. His conjecture? Improbable events provide more information to the viewer while probable events provide less. Pattern recognition reflects an attempt to make our own environment more probable.

I’m bad at probability but I like the concept of probability and statistics. It provides a way of seeing complex phenomenon by stripping out data. This clashes with our own habit of forcing all available data into manageable compartments.

Take the snow outside. It is still blowing around. Last night an experimental composer put a fifty foot sensor screen at the far end of the street. Each time a fraction of snow hit a sensor, the screen registered a sound based on flake size, speed and direction of approach. She came by this morning with a laptop that held all of the captured snowblowin’ data. When she played it back to me, it just sounded like static. Using a filter, she stripped out the smaller flakes and the flakes of snow that hit at oblique angles. Now I heard intermittent static. Using a clustering algorithm, she put the snow data into buckets and threw any data out that did not fit. Now the snow sensor yielded a rolling, pattern of loud and soft noise. A miniature end-of-the-street STOMP production.

For my part, I’d spent the morning with my video camera, tracking the snow as blew around in the space just over my porch. I assigned different colors to different flakes based on size and speed. The swarm of sow yielded a rolling, coiling surface of mixed red and green. It was as if someone had dumped a container of cream in a clear glass of coffee. My brain automatically picked up a first-order pattern. I could recognize curl boundaries and even predict motion a few seconds into the future.

The composer smiled when I showed her my attempt to render snow visually. She admitted it was pretty but it offered little predictive power. Her device was not as pretty but she could predict the sound of the snow minutes into the future.

That, of course is the tradeoff. We are good at visualizing data but bad at using more than a few variables to make predictions. Statistical methods can be used to derive patterns that (possibly) have better predictive value but these methods require that we throw out data. In a sense, data causes paralysis. The more we see, the less we are able to predict. There is, I think, an ethical component to this observation but I'm too dense to draw it out.

At any rate, it is still snowing. And if you ask me what I think about when I see the snow, I’ll tell you that I think about sensor screens and drum productions. I’m lying though. Mostly, I think “What if those flakes were actually number 2 pencils… I’ll bet the ScanTron people would really love that.”