Common sense is informal background knowledge about ordinary situations, relationships, expectations, and consequences. It helps people understand the world and act effectively without explicit instruction. section 6.1.4
Common sense supports expectations such as:
- A book belongs in a library or hand, but not inside a microwave. section 6.1.4
- A glass mug dropped on cement may break, and its water may spill. section 6.1.4
- A car that will not start cannot provide a ride. section 6.1.4
- Darkening skies and clouds suggest rain and may change picnic plans. section 6.1.4
It also interprets abstractions and context. “Boiling the ocean” communicates that a task is unreasonably large, whereas a system may interpret the phrase literally because it recognizes water and boiling without grasping the intended meaning. section 6.1.4
Reasoning beyond the visible pattern
Abraham Wald’s analysis of returning aircraft treated bullet holes as evidence of survivable damage: the most important areas to reinforce were the areas with few or no returning-hit records. This required context, abstract reasoning, creativity, and common-sense reasoning rather than simply following the strongest observed pattern. section 7.2
Data-driven AI models often struggle with common-sense reasoning, context-based inference, critical thinking, and counterintuitive situations. Training data may also leave them limited to a single domain, so data alone is insufficient for many reasoning-heavy tasks. section 7.2 section 7.3
