Use Case
PolyGraphs data animation from Northeastern's Center for Design; PolyGraphs, Northeastern University, CC-BY-NC-ND 4.0
Polygraphs’ methods and approaches have been expanded by the addition of custom data ‘processors’, code modules which essentially extend the range of analyses that can be performed with simulation data to broaden research on epistemic attitudes across domains.
Project Description
PolyGraphs is an ongoing computational humanities project in social epistemology. It uses computational methods to simulate the effects of mis- and disinformation on communities of rational agents, exploring the relative roles of social network structures, informational environments, and information processing strategies in influencing epistemic attitudes.
With financial support from the Royal Society and others under an APEX Award from 2021-2023, a scalable framework for philosophical simulations was developed in Python, and is available on GitHub (https://github.com/alexandroskoliousis/polygraphs). This enables researchers to perform experiments – effectively, batches of simulations of the (practical and theoretical) behaviour of these communities under various configurations, i.e. sets of values for the independent variables. Running these simulations generates synthetic data, which need to be analyzed and interpreted.
As a DISKAH Fellow and through engagement with DRI, Prof Ball’s research on the PolyGraphs simulation framework focuses on the potential to be further generalized – notably to new models (of rational agents, of the communities they belong to, and of the informational environments in which they operate), and to new empirical/real-world and not merely artificially generated data sets, e.g. to model climate mis- and disinformation, or decision-making in business contexts.
Skills and technical support requirements
Skills enhancement through training, allocation of study and experimentation time, as well as technical support through DISKAH has enabled Prof Ball to enhance digital skills capacity to effectively work with Polygraphs. This includes more efficient coding in Python and software refinement, improved data management practices (e.g. through Globus), better understanding of data analysis and visualisation libraries in Python and improved familiarity with computational environments, including the ability to overcome challenges, such as library dependency issues. Moreover, access to dedicated RSE support has been crucial to optimise Polygraphs’ containerisation across systems with different architecture/s, benchmarking CPU and GPU usage to design efficient HPC batch simulation running and development of suitable documentation ensuring workflow reusability.
User roles and DRI requirements
Given the nature of Polygraphs, which runs large scale simulations generating synthetic data for analysis and interpretation, the project was already computationally intensive and HPC dependent.
As such, the user-roles within this project and during the DISKAH fellowship require the researcher to work first as an AI/compute developer to support Polygraphs’ transferability to HPC systems with different configurations and improve software reusability through optimisation and documentation. Consequently, by adopting a computational critical interpreter role, the fellow runs batches of simulations in HPC under various configurations to generate synthetic data. More precisely, running simulation sub-batches (25 simulations instead of 100) in parallel allows Prof Ball to perform further experiments with variables, finding new correlations influencing epistemic attitudes.
Having completed this phase, the Fellow works as a data-shaper to improve data management practices and ensure large-scale datasets’ transferability and sharing across clusters and personal devices (using Globus). Then, the produced simulation data are further processed using the analysis module from PolyGraphs, enabling the researcher to act as a computational critical interpreter and/or AI/compute developer (when working on custom data ‘processors’) to analyse the data.
Such analysis strengthens the research potential to model mis- and disinformation, as well as decision-making in different contexts, allowing the Fellow to expand his research on epistemic attitudes across various domains and disseminate findings as a scholarly communicator.
Highlights
Polygraphs’ methods and approaches have been expanded by the addition of custom data ‘processors’, code modules which essentially extend the range of analyses that can be performed with simulation data to broaden research on epistemic attitudes across domains. Building on these enhancements, the research has the potential to equip researchers with the necessary means to create their own processors to address specific research questions.
Future work
Future work for the project includes extending Polygraphs to new models and empirical or real-world datasets to model mis- and disinformation or decision-making in other contexts.
Project Outputs