Artificial Intelligence is Like a Perpetual Stew

Abstract

Artificial intelligence is inescapable. It is in our phones, fridges, and most of the businesses we engage with use it to “improve” their services. From deciding on what YouTube video to watch next to driving vehicles or firing weapons, artificial intelligence is a lynchpin in our society. But what is artificial intelligence? And, more importantly, why does that matter?

It matters because we are currently unprepared to deal with the paradigm-shifting legal issues brought about by artificial intelligence. And without this understanding, we are nearly certainly going make mistakes; mistakes like failing to fully appreciate what copyright infringement means when artificial intelligence expresses the power of creativity or failing to understand how moving vehicle law, in and of itself, is ill-suited to encourage the safety of autonomous vehicles. The bright side, the former question, is that artificial intelligence is not complicated. Artificial intelligence—more accurately machine learning—is built on simple and intuitive concepts revolving around: (1) a particular type of machine known as a neural network; and (2) getting that machine to learn through a process of, or similar to, stochastic gradient descent. What is more, these concepts may be illuminated with a simple analogy: artificial intelligence is like a perpetual stew. All models (i.e., stews) are created by following a recipe (building the machine). The models are then set to live (i.e., a stew may simmer forever, if kept above a certain temperature). And as long as these models are maintained (i.e., re-trained on new data) they will continue pumping out accurate—tasty—results. The aim of this Essay is to provide a low-level, accurate, and easy-to-digest explanation of what artificial intelligence is. In turn, the Essay provides the legal community with an accurate vantage point from which to analyze the many AI-based issues that will not be on the horizon for much longer.

Keywords: Artificial Intelligence, Machine Learning, Neural Networks, Deep Neural Networks