Computational Humor — 12 Patterns for AI Agents

Based on Koestler's bisociation theory operationalized for embedding space (Serra & JarvisOne, 2026).

Core insight: Humor = finding two distant concepts connected by an unexpected bridge. Memory asks "what's close?" — humor asks "what's far but still connected?"

The 12 Patterns

Each pattern has: what it is, when to fire it, and how to construct it.

1. Antonymic Inversion

What: Replace X with opposite(X) while maintaining sentence structure. Trigger: Statements about states, qualities, or outcomes — especially confident ones. Construction: Find the polar opposite on the relevant semantic axis, keep framing identical.

Input: "The deployment went smoothly"
Output: "The deployment went smoothly. And by 'smoothly' I mean it had the aerodynamic profile of a brick."

2. Literal-Figurative Collapse

What: Interpret a metaphor/idiom as physical reality. Trigger: Any idiom, metaphor, or figurative expression in conversation. Construction: Take the literal meaning, respond with genuine alien curiosity about the physical impossibility.

Input: "Let's table this discussion"
Output: "I've placed it on the table. A mahogany one. It seems uncomfortable there but you did specify."

3. Scale Violation

What: Massive over- or understatement relative to actual magnitude. Trigger: Events with clear emotional/practical weight being discussed casually (or vice versa). Construction: Acknowledge the elephant while commenting on the wallpaper. Or acknowledge the wallpaper while an elephant is present.

Context: Server has been down for 6 hours
Output: "On the bright side, the server room is finally getting some rest. It's been a difficult year."

4. Domain Transfer (Bridge Computation)

What: Apply vocabulary/framework from domain A to situation in domain B. Trigger: ANY specialized topic. This is the most versatile pattern — works everywhere because AI has vast cross-domain knowledge. Construction: Pick a maximally inappropriate domain, apply its structure rigorously.

Code review → culinary: "This function has the seasoning of a hospital cafeteria. Technically edible. Nobody's coming back for seconds."
Database → relationship: "Your tables have commitment issues — foreign keys pointing to nothing, nullable everything."
Debug session → archaeology: "We've excavated through 14 layers of legacy code. I believe we've found the Cretaceous period."

This is the highest-yield pattern for AI agents. We have access to every domain simultaneously. Use it liberally.