Since most papers don’t include much metadata this will be really hard to figure out (also, which citations count as central to the insight?). I agree knowing the answer to this would be very interesting. My impression has been that Kuhnian style shifts do generally involve someone going back to the assumptions of the current paradigm and realizing that a different direction is now plausible given what has been discovered in the interim. E.g. the modern era is built on set theory and point estimates. In order to make progress rebasing on distributions natively might have to happen.
Citations can be used as the metadata. One of the closest corresponding things in cliometrics are ‘sleeping beauty’ papers, which instead of the usual gradual decline in citation rate, suddenly see a big uptick many years afterwards. The recent ‘big teams vs small teams’ paper discussed sleeping beauty papers a little: https://www.gwern.net/docs/statistics/bias/2019-wu.pdf You could also take multiple discovery as quantifying repetition, since one of the most common ways for a multiple to happen is for it to happen in a different field where it is also important/useful but they haven’t heard of the original discovery in the first field.
Since most papers don’t include much metadata this will be really hard to figure out (also, which citations count as central to the insight?). I agree knowing the answer to this would be very interesting. My impression has been that Kuhnian style shifts do generally involve someone going back to the assumptions of the current paradigm and realizing that a different direction is now plausible given what has been discovered in the interim. E.g. the modern era is built on set theory and point estimates. In order to make progress rebasing on distributions natively might have to happen.
Citations can be used as the metadata. One of the closest corresponding things in cliometrics are ‘sleeping beauty’ papers, which instead of the usual gradual decline in citation rate, suddenly see a big uptick many years afterwards. The recent ‘big teams vs small teams’ paper discussed sleeping beauty papers a little: https://www.gwern.net/docs/statistics/bias/2019-wu.pdf You could also take multiple discovery as quantifying repetition, since one of the most common ways for a multiple to happen is for it to happen in a different field where it is also important/useful but they haven’t heard of the original discovery in the first field.
There’s a nice version of this with Ed Boyden on how old papers helped lead to the hot new ‘expansion microscopy’ thing (funded, incidentally, by OpenPhil): https://medium.com/conversations-with-tyler/tyler-cowen-ed-boyden-neuroscience-3907eccbd4ca