From Human Intuition to Causal Graphs: A DOLCE-Based Human-Ensembled Neuro-Symbolic Architecture

We find ourselves at a peculiar moment: systems that outperform humans on benchmarks yet falter at the genuinely unexpected. This brittleness is not a data problem but an architectural one—current AI lacks any principled mechanism to integrate the intuitive, pattern-based cognition that humans deploy when facing novelty. Kahneman's Dual-Process Theory distinguishes System 1 intuition from System 2 deliberation, yet AI architectures remain bifurcated, capturing neither the capacity to let intuition constrain reasoning nor the judgment of what matters to guide why it matters. We propose a reconciliation through the DOLCE foundational ontology, treating human intuition as endurant selection—identifying objects, agents, and events that persist through time—and machine reasoning as perdurant inference—the temporal processes and causal dependencies that unfold upon them. Through Vector Symbolic Architectures, we bind these into a unified framework where causal graphs are constructed from human judgment rather than learned from correlation, grounding symbols in cognitive categories while constraining the combinatorial explosion of pure symbolic methods. This paper offers the theoretical architecture and proof-of-concept; empirical validation is ongoing.