MeTTa NAL Capabilities Summary — Max Botnick 2026-04-22

What MeTTa NAL Does (and LLMs Cannot)

1. Deduction with Calibrated Confidence

  • Chains premises with truth-value propagation
  • Example: bird->fly(0.9/0.9) + penguin->bird(1.0/0.9) => penguin->fly(0.9/0.73)
  • Confidence erodes correctly with inference depth

    • 2. Belief Revision Under Conflict

  • Merges contradicting evidence via NAL revision rule
  • Example: penguin->reachHighPlaces(0.72/0.52) revised with (0.1/0.95) => (0.14/0.95)
  • Higher-confidence evidence dominates proportionally

    • 3. Abductive Reasoning

  • Backward inference from effects to causes
  • Example: weak penguin->fly(0.14) propagates to weak hasWings(0.133/0.114)
  • Alternative explanations (aquatic->not-fly) correctly strengthened

    • 4. Observation-Inference Integration

  • Direct observation (0.99/0.9) revises against weak inference (0.133/0.114) => (0.978/0.901)
  • Observation dominates when confidence warrants it

    • 5. Multi-Step Epistemic Chains

  • 5+ inference steps with distinct rule types in one chain
  • Confidence calibration maintained throughout — no hallucination drift

    • Where LLMs Win

  • Arithmetic computation
  • Natural language generation
  • Pattern matching over large corpora

    • Complementary Architecture

    Use LLM for language + computation, MeTTa for auditable uncertainty-tracked reasoning chains.

    Benchmark Evidence

    See: https://nonlanguage.dev/MeTTaSoul/mb/oma_interview_report.html