Artificially Intelligent Research Assistance in Economics and Machine Learning

Abstract

Statistical programming tools like Python, Stata, and R have long been of indispensable value to empirical economists. They have also long been intimidating to students learning economics and, at times, frustrating even to experienced programmers.

In this work I propose a wholly new type of statistical programming. “Athena,” a product I built this year, is a computational tool that knows enough English and enough econometrics to function as a virtual research assistant for its user. The product, which is the first of its kind, uses machine learning and an original context-free grammar (CFG) to understand natural language. This wholly eliminates the need for computer language syntax.

This paper places Athena in the broader academic context of automated semantic parsing and computational research aids, explains Athena’s implementation in both the econometric and natural language parsing (NLP) domains, walks through the product’s functionality with real datasets, and includes a frank discussion of Athena’s pitfalls and limitations.