Incrementalism and Discontinuities in Scientific Progress

When playing around with OpenAIs GPT-3, I was struck by how "guild-like fields," that accumulate improvements very gradually (say, semiconductor manufacturing or ML for image classification) seem incremental, until they don't. That is, the field is overlooked until it reaches a singular critical threshold, at which point activity explodes. Take image classification using neural networks; NNs were outperformed by many methods like SVMs until the mid 2010s (à la AlexNet). Shortly after the deep learning revolution began, performance approached that of humans, and the field exploded.

Some, like Peter Norvig, believe science advances incrementally, by the gradual accumulation of facts and positing of theories to accomodate those facts that bring the broader picture of truth more sharply into focus. Others, like Thomas Kuhn, believe that the history of science is lined with a few, seminal discontinuities where individual minds advance the state of the art substantially in a matter of weeks by bucking against consensus thinking. In this essay, we explore the possibility of a third modality, "incrementalism as discontinuity," where there is clearly incremental progress in a field, but when the incremental progress passes a certain critical threshold, the entire field is upended and re-evaluated.

other examples?