Why AI Cannot Possess “Common Sense” — Common Sense as a Collection of Discontinuities, and Common Sense as Ontology —

AI knows an enormous amount.

Wider than an encyclopedia.
Faster than a specialist.

And yet, at a certain moment, we feel:

“This AI has no common sense.”

This is not a lack of capability.
It is a structural issue.

What Happens When We Feel “It Has No Common Sense”?

When AI output feels “strange,”
it is rarely because:

the grammar is broken
the facts are incorrect

More often, it is because:

it says something that should not be said
it ignores a sequence that should not be violated
it steps over an implicit assumption

In other words,
a boundary has been crossed.

Common Sense Is Not Knowledge

The word “common sense” is vague.
But we must separate it from knowledge.

Common sense = knowing many things
Common sense = knowing facts

This is a misunderstanding.

Common sense is:

Knowing what not to do.

Common Sense Is Not Continuous — It Is Made of Discontinuities

AI excels in continuity.

Similarity.
Proximity.
Smooth transitions.

But common sense draws cut lines within continuity.

Do not cross this point.
Do not explain that.
Not that topic, not now.

These sudden drops — these sharp edges —
are the essence of common sense.

Common Sense = A Collection of Discontinuities

Structurally speaking:

Common sense is
the collection of countless discontinuities
drawn across the world.

They are:

not logically deduced
not fully written down
dependent on place and context

Yet when broken,
they produce a powerful sense of discomfort.

Why AI Cannot Learn Discontinuities

AI learns patterns:

frequency
similarity
co-occurrence

But discontinuities are composed of:

what does not happen
what is not said
what is deliberately avoided

In other words:

They rarely appear as data.

AI can learn the density of the world.
It cannot learn the cuts in the world.

Common Sense Is Ontology

Let us shift perspective.

If we treat common sense as “rules” or “manners,”
we will inevitably fail.

Common sense is:

An understanding of how the world is segmented.

That is ontology.

What counts as inside.
What counts as outside.
What is related.
What is irrelevant.

This segmentation itself is common sense.

Ontology Is a Premise of Judgment — Not Judgment Itself

This is crucial.

Evaluation functions, probabilities, optimization —
all of them operate only on top of an ontology of common sense.

If the ontology shifts,

a “correct” judgment
can appear wildly inappropriate.

When AI says something strange,
often it is not the decision that is wrong —
it is the prior segmentation of the world.

Common Sense Can Be Explained After the Fact, But Not Fully Written in Advance

This is the difficulty.

When broken, it can be explained.

But it cannot be exhaustively specified beforehand.

That is why people say:

“That’s just common sense.”

This is not laziness.
It is structurally inevitable.

What Happens If We Try to Encode Common Sense Into AI?

If we attempt to formalize common sense as rules:

exceptions explode
context-dependence becomes unsolvable
judgment becomes rigid

We end up with:

“Common sense rules” that have lost common sense.

Common Sense Is Not Something to Embed in AI

This is the conclusion.

Common sense is:

not something to give to AI
not something to train into AI

It is a foundational structure
that humans must continue to bear.

What AI can do is:

indicate where common sense may be violated
warn when it is approaching a boundary

That is all.

Summary

Common sense is not knowledge.
Common sense is a collection of discontinuities.
Discontinuities rarely appear as data.
Common sense is ontology itself.
AI does not possess a prior world-segmentation.

AI can become intelligent.

But the act of deciding
how the world is to be divided —
the assumption structure itself —
can only be borne by humans.

For technical approaches to ontology and knowledge,
please refer to discussions under
“Ontology Engineering” and “Knowledge Information Processing Technologies.”