Machine Learning's Black Box Evan Hu

When I first stumbled across machine learning in 11th grade, I had no idea the challenges that lay before me in seeking to conquer it. To begin developing my understanding, I scoured the internet, read textbooks from my school, and sought help from my teachers. I delved deep into calculus and statistics, learning about differential equations, integrals, and optimization. The knowledge before me was daunting, overwhelming, yet enticing, and it was with tremulous excitement that I charged forward.

Eventually, I managed to create a program that learned from an enormous, assorted collection of artworks from throughout history, teaching itself to produce original pieces of art. However, as I looked over it’s bizarre artistic creations, I discovered that I was unable to explain the process by which they were made; hidden between the input and the output lay an anomalous “black box.” This “black box” grew to consume me thereafter. While some simpler programs were fathomable in their inner workings, with greater complexity, all semblances of logic disappeared: the machine’s choices felt derived from nothing. With no tangible, traceable path from which to glean rationale, I realized that the decisions of the machine would be rendered...