Cognitive Science

Why Can't Semantics Be Computed?

Sean3 min read

Why Can't Semantics Be Computed?

Core Argument

This essay argues that semantic understanding cannot be reduced to computation due to its experiential nature. Semantics comprises five dimensions: relationality, structurality, dynamicity, contextuality, and experientiality. While the first four are partially computable through knowledge graphs, ontologies, and neural networks, the fifth—experientiality—remains fundamentally non-computable.

The Experientiality Problem

Experientiality emerges from the intersection of non-computable dynamic and contextual elements, creating subjective meaning. It manifests through:

Individuality

Each person's understanding is uniquely personal, referencing Levinas's concept that "the Other is fundamentally unknowable." When you read the word "home," the images, emotions, and memories that arise are entirely different from what another person experiences.

Privacy

Following Nagel's bat thought experiment, conscious experience cannot be fully observed or communicated. Even if we could map every neuron in your brain, we still couldn't know what it feels like to be you.

The Temporal Uniqueness Principle

The author proves that uniqueness—which characterizes experientiality—must be temporal rather than spatial. This distinction is crucial:

  • Static structures are replicable: Shannon's information theory shows that any information pattern can be duplicated
  • Temporal processes are irreversible: Prigogine's dissipative structures demonstrate that time-bound processes cannot be reversed

This creates the equation:

Uniqueness = Temporality = Non-replicability = Experientiality

Each moment of understanding is a unique event in time, never to be exactly repeated—even by the same person reading the same text twice.

Computational Incompatibility

Turing computation requires three fundamental properties:

  1. Determinism: Same input always produces same output
  2. Verifiability: Results can be checked and validated
  3. Finite states: System can be described with finite parameters

These properties are fundamentally incompatible with unique, irreproducible temporal experience:

Computation Experientiality
Deterministic Non-deterministic
Repeatable Unique
Spatial information Temporal experience
Objective Subjective

Computation handles spatial information; semantics depends on temporal experience.

Practical Implications

This analysis explains why RAG and similar systems fail at true semantic understanding:

  • They treat semantics as computable information
  • They ignore the experiential core of meaning
  • They cannot replicate the temporal uniqueness of understanding

The solution is not better algorithms, but recognizing that AI systems should complement rather than replicate human understanding. We need:

  • Systems that augment human cognitive capabilities
  • Tools that preserve and transmit semantic richness
  • Architectures that respect the non-computable nature of meaning

About the Author

Deepractice - Making AI at Your Fingertips

This is the second in the Monogent theory series. Monogent is dedicated to building true AI individual cognitive systems, enabling each AI to have its unique cognitive world.