The hardest knowledge to give a machine is the knowledge no one bothers to say. That water is wet, that objects fall, that people have intentions, forms a vast unspoken background. Its absence is why capable systems can be strangely foolish.
The principle
Common sense is the immense body of ordinary knowledge that humans share and rarely articulate. It underlies understanding of language, action, and cause. Because it is assumed, it is almost never written down.
The mechanism
The difficulty is that common sense is tacit and enormous. It cannot be captured by a finite list of facts, and it resists explicit encoding. What everyone knows is precisely what no one records.
An unexpected turn
The turn is that fluency can mask its absence. A system may produce plausible language while lacking the grounding that would prevent basic errors. Eloquence is not comprehension.
The hidden cost
Learning common sense from data is partial. Statistical exposure to text conveys much but leaves gaps that surface as bizarre mistakes. The background remains uneven and brittle.
The limit
The implication is that intelligence is more than skill at tasks. It includes a robust model of how the everyday world works, without which competence is shallow. The unstated is foundational.
The larger point
Common sense is the vast, tacit knowledge humans share and never state, underlying real understanding. Its absence makes fluent systems foolish in ordinary ways. Robust intelligence rests on the background no one bothers to record. What makes the idea durable is not that it settles a question but that it reframes many. It teaches where to look and what to discount, which is often more valuable than any particular answer it yields. Understood in this spirit, it becomes a habit of attention rather than a doctrine, and habits of attention are what distinguish deep comprehension from mere knowledge.
