The Core Vision
A central challenge in neuro-symbolic AI is the selection and justification of symbolic inductive bias. Currently, the choice of a symbolic regime—whether logic, equations, or procedural code—is often treated as an external design decision rather than one derived from the structure of the task itself.
Our proposal is that Ruliology can provide a foundational framework at this level. By studying classes of abstract rules and the macro-behaviors they generate, we can identify which symbolic formalisms are most appropriate for capturing the regularities of a given world [WOLFRAM-2026].
Reifying Inductive Bias
Symbolic AI already possesses several robust local reifications of inductive bias. In **Inductive Logic Programming (ILP)**, bias is embodied in background predicates and hypothesis language [ILP-REVIEW-2021]. In **Symbolic Regression**, it is found in operator sets and structural priors [SR-REVIEW-2025]. In **Program Synthesis**, it is grammars and search strategies [PS-REVIEW-2025].
These methods are powerful, but they usually assume the surrounding formal regime in advance. If a task is cast in ILP, the logical regime has already been selected. Ruliology addresses the prior problem: how do we select among these candidate formalisms before committing to any one of them?
Ruliology as a Comparative Science
Wolfram characterizes Ruliology as the study of what abstract rules do. This shift relocates the discussion from formalism-internal bias to a broader comparative study of computable structure. We ask which classes of rules support locality, compositionality, or useful coarse-graining, and which ones are best captured by compact equations or discrete relations.
This comparative use of Ruliology helps categorize rule spaces. A rule space that reliably supports compact quantitative regularities makes symbolic regression a plausible downstream regime. A space supporting discrete compositional relations favors logic-based induction.
Discovery through Observation
To test this mechanism, we've explored "weak-observer" frameworks. In these experiments, an observer with no prior knowledge of symbol semantics induces theory profiles from simple execution behavior.
In a Boolean world, the induced theory is dominated by concepts like involution and absorption, suggesting a logic-oriented regime. In a transformational world, the profile is dominated by composition and identity, suggesting a synthesis-oriented regime. This shows that structural signals relevant to regime selection can arise from the discovered theory itself rather than from an external human choice.
References
- [WOLFRAM-2026] Stephen Wolfram. What Is Ruliology? 2026.
- [ILP-REVIEW-2021] Zheng Zhang et al. Inductive Logic Programming Techniques for Explainable AI. 2021.
- [SR-REVIEW-2025] Junlan Dong, Jinghui Zhong. Recent Advances in Symbolic Regression. ACM. 2025.
- [PS-REVIEW-2025] Zurabi Kobaladze et al. A Comparative Review of Program Synthesis Paradigms. 2025.