Neural Activity and Mental Representation

8 minute read

Published: August 21, 2026

A neuron fires.

What does that firing mean?

Does it represent:

a line,
a face,
a place,
a decision?

Neuroscience often speaks of neural representation.

But representation is not simply the same as correlation.

Neural Coding

A neural code is a mapping between:

patterns of neural activity

and:

variables relevant to the organism.

Examples include:

stimulus orientation,
spatial location,
movement direction.

The code may be distributed.

Receptive Fields

A sensory neuron’s receptive field describes which inputs influence its activity.

In vision, a neuron may respond strongly to an edge with a particular:

orientation,
location.

This supports a representational interpretation.

Feature Detectors

Early descriptions called some neurons:

feature detectors.

The idea was that a neuron responds to a specific feature.

This is useful.

But higher cognition rarely maps one concept to one cell.

The “Grandmother Cell”

A famous caricature imagines one neuron representing:

your grandmother.

If that neuron dies, the concept disappears.

Brains probably do not work this simply.

Representation is usually distributed and redundant.

Sparse Coding

Some neurons can respond selectively to highly specific stimuli.

Sparse coding uses relatively few active units for a representation.

This is not identical to one-neuron-one-concept.

A sparse population can still be distributed.

Distributed Coding

In distributed coding, information is represented by patterns across many neurons.

Each neuron may participate in many representations.

This provides:

capacity,
robustness,
generalization.

Population Vector

Motor systems can represent movement direction through activity across populations.

No single neuron determines the movement.

The weighted pattern carries the information.

Rate Coding

Information can be represented by firing rate.

More spikes over a time window may correspond to a stronger variable value.

This is one possible neural code.

Temporal Coding

Information can also depend on precise spike timing.

Two patterns with similar average rate may carry different information if timing differs.

Neural coding is not purely frequency-based.

Synchrony

Relative timing among neurons may matter.

Synchronized activity can indicate coordinated processing.

But synchrony is not a universal code.

Place Cells

Hippocampal place cells fire in relation to locations in an environment.

They provide a compelling example of neural activity representing spatial structure.

Yet place-cell activity also depends on context and task.

Grid Cells

Grid cells show periodic firing patterns across space.

Their collective activity supports spatial representation.

The code is geometric.

Head-Direction Cells

Some neurons fire according to the animal’s facing direction.

Together with place and grid cells, they form a rich navigational system.

Representation is distributed across cell types.

Sensory Coding

Neural systems encode:

color,
motion,
sound frequency,
touch.

But coding is often nonlinear and context-sensitive.

There is no universal one-variable-one-neuron scheme.

Decoding

Researchers train statistical models to predict:

stimulus or behavior

from:

neural activity.

If decoding succeeds, neural signals contain usable information.

But decodability alone does not prove the brain itself uses exactly that code.

Encoding Models

An encoding model predicts neural activity from known stimuli.

A decoding model predicts stimulus or behavior from neural activity.

These are complementary directions.

Correlation Is Not Representation

Suppose a neuron fires whenever the animal is hungry.

Does it represent hunger?

Maybe.

But the activity could be downstream of another process.

Representation requires more than correlation.

Causal Role

A stronger case arises if manipulating the activity changes:

perception,
decision,
action

in ways consistent with the represented content.

Causal involvement supports representation.

Consumer-Based Accounts

One philosophical approach says a signal represents something partly because downstream systems use it as carrying that information.

Representation depends on a consumer mechanism.

Meaning is functional.

Teleosemantics

Another approach links representation to biological function shaped by evolution.

A signal represents what it was selected to indicate.

This connects semantics to evolutionary history.

Information-Theoretic Accounts

A neural state can carry Shannon information about a stimulus if they are statistically dependent.

But statistical information is broad.

A shadow carries information about an object’s shape.

That does not mean the shadow mentally represents the object.

Representation Needs Normativity

Mental representation seems capable of being wrong.

A belief can misrepresent reality.

This suggests representation involves conditions of correctness.

Pure correlation does not capture error.

Misrepresentation

If an animal mistakes a harmless object for a predator, its neural state may represent:

predator

even though none is present.

Representation must allow mismatch between:

content

and:

world.

Mental Content

A mental representation has content.

For example:

“There is food over there.”

How neural activity acquires such content remains a major problem.

Localist Representation

Some models use explicit units.

One unit may correspond to:

dog.

This is easy to interpret.

But it can be fragile and biologically unrealistic in strong form.

Distributed Representation

A concept may instead correspond to a vector-like pattern across many neurons.

The same units contribute to many concepts.

This resembles distributed representations in neural networks.

Superposition

In artificial neural networks, features can sometimes share dimensions.

One unit can participate in multiple encoded features.

This makes representation efficient but less interpretable.

The brain may use analogous mixed selectivity.

Mixed Selectivity

Neurons in some brain regions respond to combinations of variables.

A neuron may encode:

stimulus,
task rule,
context

jointly.

This creates high-dimensional representational spaces.

High-Dimensional Coding

Mixed selectivity can make different task states linearly separable.

A population can represent many combinations even if individual neurons are difficult to interpret.

The population is the meaningful unit.

Representational Geometry

Researchers analyze distances and directions among neural population states.

Similar stimuli may form clusters.

Abstract relationships can appear as geometric structure.

Neural Manifolds

Large neural populations often occupy lower-dimensional structured regions of their enormous possible state space.

These are sometimes called neural manifolds.

The concept helps describe coordinated activity.

Mental Representation Is Not a Picture

A representation need not resemble what it represents.

The word:

cat

does not look like a cat.

Likewise, a neural code need not resemble its object.

Representation is relational.

Maps in the Brain

Some neural systems do preserve spatial relationships.

Retinotopic maps preserve neighborhood relations from the visual field.

Somatotopic maps preserve aspects of body layout.

These are structure-preserving representations.

Topographic Maps

A topographic map can be useful because nearby represented locations correspond to nearby neural locations.

But not all concepts use spatial maps.

Higher cognition often uses more abstract distributed codes.

Concept Cells

Some medial temporal lobe neurons have shown highly selective responses to particular people or concepts across different images or representations.

These are sometimes called “concept cells.”

They are not simple proof of grandmother-cell coding.

They operate within distributed systems.

Binding

If color, shape, and location are encoded separately, how does the brain represent:

this red ball here?

The binding problem asks how features become integrated.

Representation requires coordination.

Attention and Binding

Attention may help bind features associated with one object.

But binding likely involves multiple mechanisms.

No single universal solution is established.

Working Memory Representation

When a stimulus disappears, neural activity can preserve information about it.

This supports working memory.

Sometimes information is maintained through persistent firing.

Other times it may be stored in hidden synaptic or dynamic states.

Silent Memory

Recent theories distinguish active firing from latent states that can later reactivate.

Mental representation need not always appear as sustained obvious activity.

Predictive Representation

The brain may represent not only current input but expected future states.

Prediction can shape perception.

Representations are temporally extended.

Action-Oriented Representation

Some theories argue representations are organized around possible actions.

An object is represented partly in terms of:

what can be done with it.

This connects cognition to embodiment.

Representation and Consciousness

A neural representation can exist without being consciously experienced.

Visual processing can occur without awareness.

So representation is not identical to consciousness.

Representation and Thought

Thought likely depends on flexible manipulation of representations.

But the neural code may change across:

tasks,
contexts,
individuals.

Mental identity need not correspond to one fixed physical pattern.

Multiple Realizability Again

If different neural patterns can implement the same cognitive state, then representation is partly abstract.

This resembles software implemented by different hardware configurations.

But biological details still matter.

Neural Decoding and Privacy

As decoding improves, ethical questions arise about:

consent,
inference,
privacy.

Brain data is not a transparent window into exact thoughts.

But increasingly powerful inference raises real concerns.

The Philosophical Lesson

Neural representation is not one simple code.

Brains use:

rates,
timing,
population patterns,
topographic structure,
high-dimensional geometry.