Bayesian Argumentation via Delphi
BARD is a multi-year project funded by IARPA and forms part of the larger Crowdsourcing Evidence, Argumentation, Thinking and Evaluation – “CREATE” – program. Monash University aims to build a tool that uses causal Bayesian networks as underlying structured representations for argument analysis. The tool will have automated Delphi methods to help groups of analysts develop, improve and present their analyses. This involves researching and designing new means of interacting with Bayesian networks, including new means of assessing their potential in causal explanations.
At Monash, we’ve been working on Bayesian networks in Artificial Intelligence since the early 1990s, producing a wide range of models for applied sciences (e.g., environmental modelling, biosecurity, epidemiology) and tools for the automated learning of Bayesian networks from data.
In BARD, we are designing and producing Graphical User Interfaces (GUIs) for using causal Bayesian networks as the underlying engines for arguments, allowing analysts to build and test competing or complementary arguments and to examine the impact of different pieces of evidence in an intuitive environment based on the principles of Delphi.
Delphi methods have been used for fifty years to help bring experts to improved opinions in domains of uncertainty, minimizing group think and other biases using anonymising moderation.
We’ve developed automated support for the Delphi construction of Bayesian networks, which we will further enhance in BARD. BARD’s principal investigators include experts in Delphi from the University of Strathclyde and experts in the psychology of causal reasoning from Birkbeck College London and University College London.
Crowdsourcing Evidence, Argumentation, Thinking and Evaluation
The CREATE program by IARPA seeks to develop, and experimentally test, systems that use crowdsourcing and structured analytic techniques (STs) to improve analytic reasoning. These systems will help people better understand the evidence and assumptions that support or conflict with conclusions. Secondarily, they will also help users better communicate their reasoning and conclusions.
The CREATE program draws upon the strengths of academia and industry through collaborative teaming. The teams are multidisciplinary and include social and behavioural scientists, experts in informal logic and computer scientists from universities around the world.
Projects that received funding include: SWARM: Smartly-assembled, Wiki-style Argument Marshalling by University of Melbourne, TRACE: Trackable Reasoning and Analysis for Collaboration and Evaluation by Syracuse University & Co-Arg: Cogent Argumentation System with Crowd Elicitation by George Mason University