Show & Tell — Phase 26.4 CommonSenseEngine: Plausibility Reasoning & Expectations

[web3guru888]

CommonSenseEngine Architecture

The Problem: Brittle AI

Formal ontologies and compiled rules handle explicit knowledge well. But AI systems routinely fail at knowledge that humans take for granted: "You can't put an elephant in a refrigerator", "Rain makes things wet", "People eat when hungry." This is common-sense knowledge — vast, implicit, and contextual.

Solution: Hybrid Common-Sense Reasoning

CommonSenseEngine combines three approaches:

1. Knowledge base lookup — ConceptNet-style assertions with weighted confidence
2. Analogical inference — transfer knowledge from similar concepts via ConceptGraph
3. Script-based expectations — Schank & Abelson frames for stereotyped situations

Three-Signal Plausibility Scoring

Statement: "A cat can fly"
    │
    ├──► Direct Evidence
    │    KB lookup: CapableOf(cat, fly) → NOT FOUND
    │    KB lookup: CapableOf(cat, walk) → FOUND (conf=0.9)
    │    Signal: -0.3 (no direct support)
    │
    ├──► Analogical Transfer
    │    Similar concepts: dog (sim=0.7), lion (sim=0.5)
    │    CapableOf(dog, fly) → NOT FOUND
    │    CapableOf(lion, fly) → NOT FOUND
    │    Signal: -0.5 (analogues don't fly either)
    │
    └──► Script Consistency
         ExpectationFrame("domestic_animal"):
           expected_abilities = {walk, eat, sleep}
           fly ∉ expected_abilities
         Signal: -0.4 (violates expectation)

Final: PlausibilityScore(
    score = -0.4,      # weighted average
    supporting = (),    # nothing supports
    contradicting = (), # implicit contradiction
    confidence = 0.7    # moderate confidence
)

Causal Chain Inference

Query: Why does bread go stale?

Multi-hop inference (max 3 hops):
  bread ──HasProperty──► contains_moisture
  contains_moisture ──Causes──► moisture_evaporation
  moisture_evaporation ──HasEffect──► texture_change
  texture_change ──HasProperty──► stale

Explanation chain:
  "Bread contains moisture → moisture evaporates over time
   → texture changes → bread becomes stale"

Expectation Frame Generation

async def generate_expectations(self, context: str) -> ExpectationFrame:
    # 1. Identify relevant concepts
    concepts = self._extract_concepts(context)
    # 2. Retrieve causal chains
    causal = await self._get_causal_chains(concepts)
    # 3. Build expected event sequence (Schank script)
    events = self._order_by_causality(causal)
    # 4. Collect expected properties
    props = await self._get_default_properties(concepts)
    return ExpectationFrame(
        context=context,
        expected_events=tuple(events),
        expected_properties=frozenset(props),
    )

14 Relation Types (ConceptNet-inspired)

Covering: taxonomic (IsA), physical (AtLocation, MadeOf), functional (UsedFor, CapableOf), causal (Causes, HasEffect, HasPrerequisite), psychological (MotivatedByGoal, Desires), and negative (DistinctFrom).

---

Issue: #581 | Planning: #577