Deterministic Event-Graph Substrates as World Models for Counterfactual Reasoning
The paper discusses deterministic event-graph substrates as models for counterfactual reasoning. These models represent agent states as logs of RDF triples and can answer counterfactual queries through structured interventions. The study demonstrates the effectiveness of these substrates in various benchmarks, outperforming existing models in several categories.
- ▪Event-graph substrates represent agent states as append-only logs of typed RDF triples.
- ▪The study formalizes the class of substrates and proves a duality between explanatory and counterfactual queries.
- ▪The substrate exceeds the NS-DR symbolic oracle on all four per-question categories and introduces a benchmark that outperforms Llama-3.1-8B.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.15967 (cs) [Submitted on 15 May 2026] Title:Deterministic Event-Graph Substrates as World Models for Counterfactual Reasoning Authors:Fabio Rovai View a PDF of the paper titled Deterministic Event-Graph Substrates as World Models for Counterfactual Reasoning, by Fabio Rovai View PDF HTML (experimental) Abstract:We study event-graph substrates: a class of world models that represent agent state as an append-only log of typed RDF triples and answer counterfactual queries by forking the log under a structured intervention vocabulary. Substrates are inspectable at the triple level, support exact counterfactuals, and transfer across domains without learned components.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
