Exploring Machine Learning’s Contributions to Economic Productivity and Innovation

Abstract

What role does computational power play in economic productivity and innovation? How will machine learning and AI change this? Building off previous work quantifying historical computational power levels, we explore the relationship of metrics for computing power with US GDP and US internet sector GDP from 1960 to today. We develop forecast scenarios incorporating machine learning development using internet data production volumes and forecasted growth of computational power. The goal of the research is not to build a full model of productivity that incorporates computational power, but to begin to get a sense of potential role and impacts of artificial intelligence on the economy. The research has three main findings. First, we find a modest, but statistically significant relationship between computational power and economic productivity, linked to approximately 0.3-0.7% of GDP per capita and to approximately 2-3% of internet sector GDP per capita. Second, and as expected, the relationship is stronger for internet sector GDP per capita, which is linked more closely to AI. Third, and as expected, when we narrow our window of analysis to more recent windows of analysis in our regressions and in robustness tests, we see a strengthening of the relationship between computational power and GDP per capita.