Analogical Logic: A Formal System for Analogical Reasoning

Imagine you're explaining something new to a friend. You might say "the atom is like a tiny solar system" or "the brain works like a computer." We use these comparisons—analogies—constantly to understand unfamiliar things through familiar ones. They're how Darwin explained evolution (like selective breeding), how Rutherford explained atomic structure (like planetary orbits), and how we navigate everyday life.

But here's the puzzle: while we have rigorous mathematical systems for logical deduction (if A then B), probability (how likely is X?), and other forms of reasoning, we've never had a formal system for analogy. When is an analogy actually valid? How much confidence should it give us? Can we combine multiple analogies? These questions have lived in philosophical limbo for over a century.

What This Paper Does

This paper creates the first complete logical system for analogical reasoning—essentially, the "mathematics of analogy." Just as probability theory gives us precise rules for reasoning under uncertainty, Analogical Logic (AL) gives us precise rules for reasoning by similarity.

The Core Insight

The key idea is that analogies aren't about surface similarities—they're about structural correspondences. A whale looks like a fish (similar shape, fins, lives in water), but that's a weak analogy because their deeper structures differ fundamentally (mammals vs. fish, lungs vs. gills, warm vs. cold-blooded). Meanwhile, the atom and solar system look nothing alike at the surface level, but make a powerful analogy because their relational structures match: a central massive body attracts smaller bodies that orbit it.

The system captures this by separating:

Relational structure: How things relate to each other (orbits, attracts, causes)

Surface properties: What things are like individually (hot, charged, massive)

How It Works

The paper builds a complete formal system with five components:

A language for precisely describing domains (like the solar system or atom) and mappings between them

Five axioms that characterize how analogies behave:

Every domain is perfectly analogous to itself

If A is analogous to B, then B is analogous to A

Analogies can be chained, but get weaker with each link

Valid analogies must preserve relational structure

Surface properties affect analogy strength but not validity

Five inference rules for deriving new knowledge:

Transfer relations from source to target

Transfer properties (with reduced confidence)

Recognize when differences weaken analogies

Generate hypotheses by transferring explanations

Strengthen conclusions when multiple analogies converge

A strength metric (Σ) ranging from 0 to 1 that quantifies how good an analogy is, combining structural alignment with property similarity

Soundness proofs showing that valid analogical arguments produce reliable conclusions with calculable confidence levels

What Makes It Non-Obvious

Some surprising results emerge:

Non-monotonicity: Unlike deductive logic, adding true information can invalidate previous analogical conclusions. The whale/fish analogy weakens dramatically when you learn whales are mammals—new knowledge can break old analogies.

Weak transitivity: If A is analogous to B and B is analogous to C, then A is analogous to C, but more weakly. Information degrades through analogical chains.

Structure trumps properties: A perfect structural match with zero property overlap (Σ = 0.70) creates a stronger analogy than perfect property match with weak structure (Σ < 0.50).

Seeing It In Action

The paper works through historical scientific analogies in detail:

Rutherford's atom (like a solar system): Calculates Σ = 0.80 (strong analogy), shows which inferences were valid (inverse-square force law) and which failed (continuous electron trajectories—quantum mechanics revealed this disanalogy)

Darwin's natural selection (like artificial breeding): Calculates Σ = 0.88 (very strong), shows how the analogy generated the theory of evolution despite the key disanalogy (no intentional "breeder" in nature)

Electricity (like water flow): Shows a moderate analogy (Σ ≈ 0.70) that's useful for engineering despite microscopic differences

Why It Matters

This isn't just theoretical housekeeping. The system:

For AI: Provides foundations for machines to reason by analogy rigorously, with confidence estimates

For science: Formalizes how analogies drive discovery and when to trust them

For philosophy: Resolves century-old debates about the nature of similarity and analogical inference

For education: Helps evaluate teaching analogies (which ones support learning vs. create misconceptions?)

For everyone: Makes explicit the implicit reasoning we use constantly