A clockwork butterfly
Presenting to the American Association for the Advancement of Science in December 1972, MIT meteorology professor Edward Lorenz asked, “Does the Flap of a Butterfly’s Wings in Brazil Set Off a Tornado in Texas?”
Over a decade earlier Lorenz had observed how small and seemingly insignificant changes in the starting conditions of his weather modelling caused outsized and unpredictable outcomes. This fascinated him and formed the basis for a branch of mathematics known as chaos theory. That question he asked overturned our understanding of how the world works that had stood for three hundred years.
Seventeenth and eighteenth century scientists, people like Sir Isaac Newton and Pierre-Simon Laplace, believed in a clockwork universe. They argued that unpredictability has no place in the universe, and that if we knew all the physical laws of nature, then “nothing would be uncertain and the future, as the past, would be present to our eyes.”
These different ways of understanding how one thing leads to another present a bit of a conundrum.
If I drop a plate on the floor, it breaks. That’s predictable. Cause and effect. But the way the plate shatters, the size and shape of the broken parts, is unpredictable and unrepeatable. No two plates ever break in exactly the same way.
If you tap the ‘a’ key on your keyboard, the ‘a’ character appears on your screen. That’s predictable. But who reads what you wrote, what effect it has on them, what it causes them to think, to feel, to do, that is unpredictable. You can never know the effect you have on people. Lots of responsibility, huh?
If a deep learning computer model is trained on 84,743 photos of retinas, it can identify a person’s sex with a high degree of accuracy. That wasn’t predicted. Scientists had no idea that it might be possible to identify gender from the retina. But the data set only contained photos from a generally healthy Caucasian population, which is not exactly a diverse data set. How might such a limited data sets affect how algorithmically identifying sex happens in the future. Who has responsibility then?
Our understanding of technology is increasingly moving from Newton’s predictable world to Lorenz’s unpredictable world. Unpredictable outcomes aren’t an inherently bad thing, they are part of how the world works. The problem is when we assume predictability for unpredictable things. Then we get surprised by the outcomes. Rather than accepting we don’t know what effect new and emerging technology might have, we assume we can predict it. Not only is our prediction wrong, but so is our thinking about predictability.
Does the flap of a clockwork butterfly’s mechanical wings in Brazil set off an unpredictable algorithm in Texas?