Signal and sign.
Why machines can't be smart.
AI is fluent. It sounds like it understands. Fluency and understanding are not the same thing.
Welcome
AI is fluent. It sounds like it understands. But fluency and understanding are not the same thing.
Tonight we sit in that gap — between signal and sign, between math and meaning.
Where we left off.
Last salon: AI is exceptional at predicting objective questions, and useless at disambiguating subjective ones.
Tonight we go a layer deeper — not what AI does, but how we communicate with it, what it is, and what it isn't.
A quick recap,
then the deeper question.
Prediction: pattern matching from past data.
Ambiguity reduction: framing, interpretation, judgment.
Machines do the first. Humans do the second.
What we didn't ask last time — when a model produces a sentence that sounds meaningful, where is the meaning located?
Tonight's answer: nowhere.
The Jennifer
Aniston neuron.
In 2005, researchers at UCLA studying epilepsy patients found something strange. They placed electrodes deep in the medial temporal lobe of a patient and one single neuron fired for Jennifer Aniston.
- Her photograph
- Her name in text
- A line drawing
- A still from Friends
One cell for a symbolic concept.
Recognition
without reduction.
Human cognition does something specific: it takes a flood of sensory signals and binds them to a concept — a parcel of meaning.
The neuron isn't reacting to pixels. It's reacting to Jennifer Aniston as an idea you carry around with you.
We can think of it as a topology of meaning, with peaks and valleys across people — individual signs in the territory of significance.
What a language model
does instead.
A language model has no Jennifer Aniston neuron. It has no neurons at all — just weights.
When you type Jennifer Aniston, it doesn't retrieve a concept. It computes a probability distribution over what tokens tend to follow those tokens, given everything it has read.
The output can sound like understanding. But without symbolic topology, there can't be comprehension.
Signal vs. sign.
| Signal | Sign | |
|---|---|---|
| What it is | Data — a measurable input | A meaning — a thing that stands for something |
| Where it lives | In the medium | In a mind |
| What it requires | A sensor | An interpreter |
| Example | Smoke | Something is burning |
| Who handles it well | Machines | Humans |