Data Augmentation with BERT

Many of the most significant breakthroughs of AI and ML in the 2010s were theorized and described many decades ago. A few of the greatest ingredients to the AI gold-rush of the past decade are the perceptron (1958), backpropagation (1975), and (for NLP) recurrent neural networks (1986) [1, 2, 3].

How do seemingly obscure innovations from 1958, 1975, and 1986 become critical to a science and tech revolution in the 21st century?

This is a companion discussion topic for the original entry at