Towards a compositional semantics for quantitative confidence assessment in assurance arguments

2026-05-23 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Expert, extended

Summary

Assurance arguments provide a clear and structured way to explain why stakeholders should trust that a system satisfies certain properties. This article proposes a compositional semantics for quantitative confidence assessment in such arguments, addressing the lack of operational semantics in notations like Goal Structuring Notation (GSN) for deriving "assurance confidence". The approach represents argument elements (goals, evidence, assumptions) as Subjective Logic (SL) opinions (belief, disbelief, uncertainty) and maps relations (support, context) to SL operators (conditional deduction, conjunction, disjunction, fusion). This effectively transforms an argument into an analyzable confidence network, enabling systematic confidence propagation, explicit warrants, and principled handling of context. The paper provides practical guidance, illustrating the method with an exemplary GSN argument and demonstrating how confidence in top-level claims is computed from underlying evidence under various scenarios, including full uncertainty, full confidence, and partial confidence, while also showing how assumptions influence the overall assessment.

Key takeaway

For Assurance Engineers developing complex system arguments, this compositional semantics offers a robust method to quantify confidence. You can systematically trace how belief, disbelief, and uncertainty propagate from evidence to top-level claims. This enables identifying weak links, assessing the impact of new evidence, and making explicit the influence of assumptions, enhancing auditability and supporting continuous re-assessment in iterative AI system development.

Key insights

A compositional semantics using Subjective Logic quantifies assurance argument confidence by mapping elements to opinions and relations to operators.

Principles

• Assurance arguments can be modeled as confidence networks.
• Inferential steps require explicit warrants via conditional opinions.
• Contextual information conditions argument element validity.

Method

The proposed method involves identifying reasoning strategies, eliciting SL opinions for elements and conditional opinions for relations, propagating confidence using SL operators, and iteratively reviewing results, including assumption impacts.

In practice

• Represent GSN elements as Subjective Logic opinions.
• Map GSN relations to SL operators for confidence propagation.

Topics

• Assurance Arguments
• Subjective Logic
• Goal Structuring Notation
• Confidence Propagation
• Quantitative Semantics
• System Safety

Best for: AI Scientist, Research Scientist

Related on AIssential

• Towards a compositional semantics for quantitative confidence assessment in assurance arguments — A compositional semantics uses Subjective Logic to model assurance argument elements and relations as a confidence network.
• Quantifying Faithful Confidence Expression in Large Reasoning Models — Faithful confidence expression is a distinct, significant challenge for Large Reasoning Models, requiring new quantification methods.
• On the Formal Limits of Alignment Verification — AI alignment verification faces a trilemma: soundness, generality, and tractability cannot be simultaneously achieved.
• SoK: AI-Augmented Binary Reversing — AI-augmented binary reversing complements conventional methods, leveraging artifacts as an interface for learning-based semantic inference.
• Concerning Uncertainty -- A Systematic Survey of Uncertainty-Aware XAI — UAXAI integrates uncertainty into AI explanations to improve transparency, reliability, and human trust in high-stakes decision-making.
• Trust-Aware Multi-Agent Traceability: Confidence-Calibrated Knowledge Graphs for Consistent Software Artifact Management — Calibrated confidence scores enable trust-aware coordination and consistency detection in multi-agent software engineering pipelines.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.