BARD: A Structured Technique for Group Elicitation of Bayesian Networks to Support Analytic Reasoning

Area of Science:

• Artificial Intelligence
• Decision Science
• Cognitive Science

Background:

• Human reasoning about causation and uncertainty is complex and error-prone.
• Bayesian networks (BNs) offer a powerful AI tool for probabilistic and causal reasoning.
• Existing BN methodologies lack comprehensive training, guidance, and collaborative features.

Purpose of the Study:

• To introduce Bayesian Argumentation via Delphi (BARD), a novel methodology and application for Bayesian network (BN) development.
• To address limitations of current BN tools by integrating training, guided construction, automated reporting, and collaborative features.
• To enable groups without prior BN expertise to build, validate, and reason with causal models.

Main Methods:

• Developed BARD, an end-to-end online platform combining e-courses, stepwise BN construction, automated reporting tools, and a multiuser web interface.
• Integrated Delphi-style social processes for collaborative model building.
• Conducted initial experiments to evaluate BARD's effectiveness in reasoning and reporting.

Main Results:

• BARD provides integrated training, guided model building, automated explanations, and collaborative features for Bayesian network development.
• Initial experiments indicate BARD effectively aids in causal reasoning and analytic reporting for suitable problems.
• The combination of BN-building and collaboration within BARD shows beneficial and cumulative effects.

Conclusions:

• BARD offers a comprehensive solution for accessible and collaborative Bayesian network modeling.
• The platform empowers users with limited expertise to leverage causal reasoning for complex problem-solving.
• BARD demonstrates potential for improving decision-making and analytical reporting through enhanced AI methodologies.