| Abstract: |
Representing and reasoning with complex, uncertain, context-dependent, and
value-laden knowledge remains a fundamental challenge in Artificial
Intelligence (AI) and Knowledge Representation (KR). Existing frameworks often
struggle to integrate diverse knowledge types, make underlying assumptions
explicit, handle normative constraints, or provide robust justifications for
inferences. This preprint introduces the Conditional Reasoning Framework (CRF)
and its Orthogonal Knowledge Graph (OKG) as a novel computational and
conceptual architecture designed to address these limitations. The CRF
operationalizes conditional necessity through a quantifiable, counterfactual
test derived from a generalization of J.L. Mackie's INUS condition, enabling
context-dependent reasoning within the graph-based OKG. Its design is grounded
in the novel Theory of Minimal Axiom Systems (TOMAS), which posits that
meaningful representation requires at least two orthogonal (conceptually
independent) foundational axioms; TOMAS provides a philosophical justification
for the CRF's emphasis on axiom orthogonality and explicit context (W).
Furthermore, the framework incorporates expectation calculus for handling
uncertainty and integrates the "ought implies can" principle as a fundamental
constraint for normative reasoning. By offering a principled method for
structuring knowledge, analyzing dependencies (including diagnosing model
limitations by identifying failures of expected necessary conditions), and
integrating descriptive and prescriptive information, the CRF/OKG provides a
promising foundation for developing more robust, transparent, and
ethically-aware AI systems. |