Science appears to evolve over time, but the sense in which science “progresses” is still debated by historians, sociologists, and philosophers of science. Despite the plurality of views concerning scientific progress, there is one point of consensus: any complete understanding of scientific progress must account for the ways scientific concepts develop over time. To appreciate the primacy of scientific concepts, consider how scientific explanations and theories-- two categories of research output commonly supposed to encapsulate scientific knowledge-- are constituted and preconditioned by scientific concepts. In this project, we define a protocol for combining traditional methods from the history, sociology, and philosophy of science with recent methods from machine learning to model the evolution of scientific concepts.
For this meeting, the team will focus on the “microbial biofilm” concept. To study the evolution of this concept, we will analyze a corpus of scientific publications related to microbial biofilms from 1929–1974. First, we will scrape text and images from the corpus publications. Next, we will investigate ways to computationally represent these heterogeneous concept components (which we call descriptions) and their social relationships (which we call relations). Finally, we will analyze our model by comparing the results to traditional historical narrative, performing further historical investigation on instances in the model that “look like” conceptual evolution, investigating the identity of our generated categories with an expert systems approach, and examining how our interdisciplinary group reacts to the novel representation with an extended computational case method.