Paper short abstract:

One of the ways computers entered scientific cultures is as ideal number-crunching machines which, given the same input, always deliver the same output. While rarely realized in practice, I will argue that this ideal may have contributed to the emergence of the "reproducibility crisis".

Paper long abstract:

Since the turn of the millennium there has been increasing alarm regarding a "reproducibility crisis" in science, leading to calls for overarching requirements to be imposed on scientific practice, one of the most prominent being that of "computational reproducibility," characterized as "obtaining consistent results using the same input data, computational steps, methods, code, and conditions of analysis" (2019 Report on Reproducibility and Replicability in Science of the U.S. American National Academies of Science, Engineering and Medicine). Yet the variety of "data," "code," or "conditions of analysis" in the practices of different cultures of research lets the operationalisation of such normative statements appear a hopeless challenge, shedding doubts that such reproducibility may have been achieved in the past. How did computational reproducibility then attain and maintain its prominent position? I will argue that its rise went hand in hand with the emergence of the reproducibility crisis, and that both were linked to the expectations of reproducibility raised by the introduction of computer-assisted methods in a growing number of research cultures around the end of the 20th century. Starting point for my presentation will be a paper published in1992 by geophysicists John Claerbout and Martin Karrenbach, which is today often quoted as marking the beginning of the "reproducibility crisis," but where in fact the authors welcomed "word processing and software command scripts" as providing the chance of improving reproducibility in computationally-aided science.