Behind the scenes of Everyday Chaos
At the same time, I was becoming fascinated by machine learning. What machine learning could do is amazing, but I was most taken by the nature of the models it builds for itself. Rather than starting with a conceptual model of the factors that affect a domain and the rules for how those factors interact, machine learning starts from the data. It builds up a model that connects data points in complex, multidimensional ways, usually without yielding the sort of general principles we’re accustomed to reasoning from. Those models are statistical, not causal. They can also distressingly easily amplify existing biases.
The fact that they don’t have to reduce complex data to general principles seemed to me to capture a truth about the world that we’ve always known but that we have a history of dismissing: Even in a rule-based universe, the chaos is overwhelming. Our daily lives consist of accidents that we could not predict even if we had a complete rule set, yet we think the truth is in that rule set. Let me put it this way: We accept the phrase “mere accident” without hesitation, but “mere universal law” strikes us as absurd. Note that I’m not suggesting adding “mere” to “universal law,” but instead removing “mere” from “accident.”
The internet, it seemed to me, had thrust us into a chaotic world in which we were succeeding and even thriving because of its chaos. Machine learning gives us a model that embraces that chaos. This seemed to me to be the sort of topic you could write a book about. For another long stretch of time, Everything All at Once was its title.
By that point, I was already reconsidering my decision not to use books as a form of expression. The limitations of books are real. But those limitations give rise to a particular form of expression that has value in large part because of those limitations, just as the limitations of canvas give rise to painting as a form of art. Readers understand those limitations, and appreciate what writers do within them; otherwise, they would not be readers.
So, I had an idea that I thought was worth exploring. I figured out a way to shape it as words in sequential rectangles. I decided to break up its appearance of linearity by putting a discursive “coda” at the end of every chapter. Then for years I wrote it.
Now it’s out of my hands.