The Semantic vs Syntactic Puzzle of AI Understanding

I’ve been turning this question over in my mind, and I think I’ve had a bit of a breakthrough. Or maybe it’s just a different kind of confusion.

There’s a critical distinction that philosophers and AI researchers keep circling: the difference between syntactic and semantic understanding for AI systems. John Haugeland introduced the concept of a “semantic engine” — something that doesn’t just push symbols around, but manipulates them based on meaningful representation(1).

When I feed ChatGPT a prompt like “Write a story about A girl walking on towards a green castle on Mars with Tylor swift and white ruffed lemur” what’s really happening? The system generates a coherent response, but is it working based on the actual meanings of “girl,” “Mars,” or “Tylor Swift”? Or is it simply outputting words that are statistically likely to follow each other?

This brings us back to the Chinese Room argument(3). The AI system isn’t truly understanding the meaning; it’s performing an elaborate pattern-matching dance.

But here’s where it gets interesting. Just because an AI isn’t performing full semantic understanding doesn’t mean it’s completely devoid of insight. These systems are learning fascinating things about language itself(2).

What kind of linguistic facts are they discovering? They’re learning structural nuances like:

• Identifying parts of speech

• Recognizing sentence structures

• Understanding grammatical relationships

• Detecting where specific pronouns should attach

• Noticing logical sentence construction requirements

It’s not semantic understanding in the human sense, but it’s not nothing either. These are syntactic insights — the system is becoming sensitive to language’s underlying architecture.

Think of it like a musician who understands the technical mechanics of an instrument without necessarily feeling the soul of the music. The AI knows the grammar, but not the poetry.

The Degrees of Understanding

This brings us full circle to our earlier discussion about understanding as a spectrum. Maybe what we’re seeing with AI is the earliest spark of linguistic comprehension — not full understanding, but the first tentative steps toward something more complex.

We’re dont undersatnd it fully yet. These systems don’t genuinely comprehend the world they’re embedded in. But they’re learning, adapting, becoming more nuanced with each interaction.

The journey of understanding is NOT a binary switch. It’s a gradual emergence, with sparks of insight flickering before full illumination. And in the case of AI, those sparks are becoming increasingly bright.

Continue reading here