ExplainabilityTrustworthinessTransparencycs.AI
ARIA: A Causal-Aware Framework for Rescuing LLM Reasoning in Trustworthy Materials Discovery
Yi Cao, Liaoyaqi Wang, Jieneng Chen, Benjamin Van Durme, Alan Yuille, Paulette Clancy
June 21, 2026
Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2
Key Takeaways
-
LLMs can suffer from "contextual tunneling": The paper identifies a specific failure mode where language models become over-reliant on narrow pieces of retrieved information while ignoring broader physical principles, potentially leading to scientifically invalid recommendations in materials discovery.
-
Causal reasoning improves trustworthiness: ARIA's three-tier approach—prioritizing complete causal chains, then analogical reasoning, then explicit fallbacks—demonstrates that structuring AI systems around mechanistic completeness rather than pattern matching produces more reliable and physically grounded outputs.
-
Auditability is essential for high-stakes applications: The framework's ability to produce "auditable causal traces" means scientists can see why the AI made a recommendation, not just what it recommends. This transparency is crucial for responsible deployment in domains like materials discovery where decisions have real-world consequences.
-
Knowledge graphs alone aren't sufficient safeguards: Simply augmenting LLMs with structured knowledge (2,839 Process-Structure-Property relations in this case) doesn't automatically prevent reasoning failures—the paper shows that how models use that knowledge matters as much as having it available.
-
Hybrid human-AI workflows require explicit fallback mechanisms: ARIA's design includes explicit acknowledgment of uncertainty through parametric fallbacks when evidence is incomplete, rather than forcing the model to generate answers anyway—a practical principle for responsible AI in specialized domains.
Abstract
Generative models have revolutionized the process of materials discovery, yet they often fail to satisfy underlying physical causality. Through an analysis of Large Language Models (LLMs) augmented with knowledge graphs derived from current literature, we uncover a phenomenon termed contextual tunneling, where models "over-anchor" on narrow, retrieved evidence while suppressing global physical reasoning. To address this problem, we introduce ARIA, a causal-aware framework that conditions knowledge use on mechanistic completeness. ARIA routes each query through a three-tier cascade: (i) direct causal reasoning when complete evidence chains of Process-Structure-Property (PSP) are available, (ii) physics-informed analogical transfer for sparse or novel material systems, and (iii) explicit parametric fallback when external evidence is incomplete. As a proof of concept, we construct a Knowledge Graph (KG) containing 2,839 extracted PSP relations from peer-reviewed articles in the materials literature and evaluate ARIA on forward prediction and inverse design tasks for two-dimensional (2D) materials. ARIA mitigates contextual tunneling, improves over unaugmented and naive KG-augmented baselines, and provides further gains when an online literature search is used for evidence enrichment. Crucially, ARIA produces auditable causal traces, enabling physically grounded and trustworthy AI-assisted materials discovery.
Accepted to the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD 2026)
Paper Info
arXiv ID2606.22375
PublishedJune 21, 2026
Citations
0
Relevance78%
VenueProceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2
TypeJournalArticle, Book, Conference, Review
Categorycs.AI
Responsible AI Topics
ExplainabilityTrustworthinessTransparency
Links
Citation data from Semantic Scholar