# Knowledge, Reasoning, and Semantics in AI

How do AI systems represent the world and reason about it? This sub-article expands [Section 1.2 of Philosophy](../philosophy.md).

## The Grounding Problem
- How do symbols connect to the world?
- Harnad's *symbol grounding problem* asks how arbitrary tokens acquire meaning.
- Modern models learn from text, which is itself grounded only in data—raising questions about genuine understanding.

## The Frame Problem
- How does a system know what to update when the world changes?
- Relevant to planning, robotics, and world models.

## Embodiment
- Brooks argued intelligence without representation—behavior through interaction.
- Embodied systems use perception and action to bootstrap meaning.

## Representation vs. Prediction
- Symbolic AI used explicit representations; deep learning uses learned latent representations.
- Whether these are "semantics" or mere statistics is debated.

## Why It Matters
The nature of AI "understanding" determines trust, explainability, and generalization.

## Sources
- Harnad, S. (1990). *The Symbol Grounding Problem.*
- Brooks, R. (1991). *Intelligence Without Representation.*

*Back to [Philosophy](../philosophy.md).*