The idea of a universal approximator has been around since the early 1900s. The mathematical proofs required to prove that you can transform any function into a set of inner and outer functions (approximation) was shown by Andrey Kolmogorov and Vladimir Arnold (his student) in the early 60s. It was called the Kolmogorov-Arnold Representation Theory and was later used to develop what we now know as “Deep Learning”.
In Academia Letters, I published a letter that draws a a comparison between the development pathway of deep learning and the development pathway of agent-based modeling titled “Are Agent-based Models Universal Approximators?”. In the paper, I discuss the historical development of both disciplines and point to a seldom considered paradox in the way we view agent-based model outputs.