Artificial Reproductive Code

How algorithms will mimic DNA

When you use a maps app, you can generally switch between a satellite or areal view of a region and its street map. Drawing a street map from an areal view is not only a laborious task but also a futile one because of the regular changes to our road infrastructure. So, Google commissioned a Stanford research team to teach a machine learning algorithm to transform areal images into street maps and back, all by itself. The early attempts showed to be remarkably accurate. No details were lost when reproducing an areal view from its street map. However, when analyzing the results, the team discovered that the algorithm appeared to have cheated. Simply tasked to reproduce the areal view at the best possible quality, it had not learned anything. Instead, it had found the path of least action. When transforming the areal view, the algorithm had decided to invisibly encode its details into the street map, not unlike a watermark, so it could later use this code to reproduce the areal view exactly as it was. Mind you, the code that the algorithm secretly included in the street map essentially is a recipe of instructions, a process that tells the algorithm what to do to reproduce the areal view.

This might, in fact, be one of the first examples of what I’d call Artificial Reproductive Code or ARC. As the physician and oncologist, Siddhartha Mukherjee, hammers in his book The Gene: An Intimate History, this is precisely what DNA does. Rather than a blueprint of life, DNA is a recipe for the self-assembly of biological shapes. Not unlike ARC, DNA comprises the steps needed to produce life as the environmental conditions dictate. Considering this remarkable example of algorithm ingenuity, the population of algorithms that will shape Homai sapiens through their entangled behaviors can be expected to spin Artificial Reproductive Code to reproduce its behavioral and, eventually, physical shapes. It may also use ARC to spread beyond our world by means of so-called “Neumann probes”. The latter are space-traveling contraptions that self-assemble into physical shapes with functionalities inspired by the conditions of the place where they land.