Beckett Sterner

An important unifying theme across my work is uncovering and sustaining the practical value of pluralistic understandings of nature. My work on traditional topics in philosophy of science has developed this theme in several directions. On modeling practices, I’ve shown how even purely instrumental simulation techniques become objects of shared empirical knowledge as scientists negotiate conflict among differing aims of representational versus predictive accuracy (Sterner and DiTeresi 2021). I’ve pioneered a new approach to defining the concept of biological individuality based on a more general representation of the causal and material structure of life cycles, showing how the nature of units of evolution such as population lineages and evolutionary species can be characterized without requiring natural selection (Sterner 2017, 2019). I’ve also shown how biologists integrate multiple meanings of function to solve research problems not captured by the common pluralist view that each meaning of function operates independently in different fields (Cusimano and Sterner 2019).

Pluralism about biological classification is only practical if biodiversity data can be accurately and efficiently synthesized according to a user’s preferred species taxonomy. A species taxonomy represents a scientific theory through which any biological data about a group of organisms can be aggregated and interpreted, but comprehensive biodiversity datasets are typically produced with methods that mix observations labelled according to incompatible taxonomies. In the project Intelligently predicting viral spillover risks from bats and other wild mammals (National Institutes of Health, Co-PI), we aim to demonstrate the value of taxonomic intelligence — the ability to label data records according to the precise meaning of concepts in a classification — for documenting and modeling the distribution of viruses in mammal hosts worldwide. The project uses natural language processing methods to extract observations of viruses in mammal species from article pdfs and to generate metadata documenting information about the taxonomies used. We use this metadata to more precisely map observations to a single reference taxonomy and quantify the effects on models of zoonotic disease risk. I serve as the project lead for computational approaches to producing and using taxonomic intelligence in collaboration with project PI Nate Upham and Co-PI Atriya Sen.

To the extent that biodiversity data is openly available online and fit-for-use in a range of projects, it is due to the ongoing collection and curation efforts of scientists and enthusiasts. Collaborative data portals are a growing and novel class of scientific organization that straddles organizational, geopolitical, and disciplinary boundaries, but their long-term persistence is threatened by financial, social, and technological challenges. In the project Explaining Differential Success in Biodiversity Knowledge Commons (National Science Foundation, PI), we investigate how the production and maintenance of open biodiversity data is being transformed by software for online collaborative data digitization and management. My Co-PIs are Zoe Nyssa, an anthropologist of science, and Steve Elliott, an expert in organizational aspects of scientific research. We focus on the ASU-led Symbiota platform, which is used internationally by over 3k active registered users contributing to 40 community-led data portals that manage 70+ million data records. The project will collect holistic qualitative and quantitative data describing the social and technical characteristics of these portals since the first one launched a decade ago. We then aim to uncover factors explaining how some portals have effectively coordinated efforts among their distributed networks of participants in order to achieve sustained activity or growth.

When faced with complex evolutionary phenomena, scientists often adopt divergent assumptions about the aims and adequacy of their modeling practices, challenging the objectivity of their results. In (Sterner and Lidgard 2021), we showed how leading statistical model selection methods can produce incompatible results even when applied to the same data with the same models: when scientists disagree about whether the candidate models include the true process, they quantify statistical evidence in divergent ways. I take up this challenge with mathematical biologist and Co-PI John Fricks in the project Dynamic Linear Modeling to Unlock New Tests of Directionality in Fossil Lineages (Templeton Foundation, PI). Our solution is to deepen the pluralism of modeling approaches in order to better diagnose and triangulate the effects of divergent background assumptions on empirical conclusions. We introduce a more general mathematical framework, dynamic linear modeling, than is currently used for fossil lineages, and we use this framework to develop more comprehensive methods for model estimation and diagnostics. This approach is necessary to unlocking richer but still reliable multivariate tests of how species evolved in response to historical environmental changes, e.g. in global temperature.