Higher-Order Theories: Does Awareness Require Representation?

A visual system can represent a red apple without you necessarily seeing it consciously. Higher-Order Theories ask whether awareness appears only when the mind also represents itself as being in that first-order state.

Seeing and knowing that you see may not be the same event

A visual signal reaches the brain.

Edges are detected.

Color is encoded.

Shape is categorized.

The object influences behavior.

Yet suppose there is no conscious experience of it.

What is missing?

A first-order sensory representation of the object already exists.

Higher-Order Theories propose that consciousness requires something more.

The system must, in some sense, represent itself as being in that state.

This does not necessarily mean verbal self-reflection.

You do not need to think:

“I, Kyle, am currently experiencing red.”

The higher-order representation can be nonverbal, fast and implicit.

The central idea is simpler:

a mental state becomes conscious when it is appropriately represented by another mental state.

First-order representation is about the world

Your visual system represents:

a face;

a line;

a color;

movement.

Your auditory system represents:

pitch;

voice;

location;

rhythm.

These representations can affect behavior without entering awareness.

Blindsight is a famous example.

Some people with damage to primary visual cortex report no visual experience in part of the visual field yet can perform above chance on certain visual tasks there.

Something in the system contains information.

But the person says they do not see.

Higher-order theorists use cases like this to motivate a distinction between:

representing X

and

being consciously aware of representing X.

The higher-order state is not necessarily another little observer

A common criticism imagines an infinite regress.

If first-order state A becomes conscious because higher-order state B represents it, does B need state C to make B conscious?

Then D?

Then E?

Higher-order theorists reject that requirement.

The higher-order state does not itself need to be conscious in order to make the first-order state conscious.

It performs a representational function.

There is no homunculus sitting inside the brain looking at another mental screen.

Still, the intuition can be difficult.

Why should an unconscious representation of another state make that state feel like anything?

That question remains one of HOT’s deepest philosophical pressure points.

There is not one Higher-Order Theory

The family contains different versions.

Higher-Order Thought theories

A conscious state is accompanied by a suitable thought or conceptual representation that one is in that state.

Higher-Order Perception approaches

The higher-order state may resemble an inner perception or monitoring process rather than a thought.

Higher-Order State-Space and related models

Contemporary neuroscience versions often frame the higher-order representation in terms of learned state spaces, probabilistic inference or metarepresentational coding.

These approaches agree on a broad architecture.

They disagree on what the higher-order representation is and how it is physically implemented.

So saying “HOT predicts X” always needs caution.

Which HOT?

Higher-order does not mean “high intelligence”

This is one of the biggest misunderstandings.

A system does not need advanced language, philosophy or explicit self-reflection.

Hakwan Lau and David Rosenthal have emphasized that higher-order processing need not be cognitively elaborate.

The relevant representation can be relatively minimal.

This matters for animals, infants and clinical cases.

If HOT required the explicit concept:

“I am currently in mental state M,”

it would imply implausibly narrow consciousness.

Most proponents avoid that.

It is also not simply confidence

Researchers often ask participants:

How confident are you that you saw the stimulus?

Confidence is related to metacognition.

Higher confidence can correlate with awareness.

But confidence is not identical to consciousness.

A person can be consciously uncertain.

A person can have misplaced confidence.

Higher-order theorists generally do not define consciousness as confidence rating.

Instead, confidence tasks provide one experimental window into how the brain represents its own perceptual states.

The window can be informative without being the thing itself.

Prefrontal cortex became the obvious neural candidate

If the brain must represent its own sensory state, where might that happen?

Prefrontal cortex is heavily involved in:

• monitoring;
• decision-making;
• metacognition;
• working memory;
• flexible behavior.

This made frontal systems natural candidates for higher-order representations.

Some experiments have linked prefrontal activity or damage with changes in subjective report or metacognitive accuracy.

But the inference is contested.

Frontal cortex is also involved in preparing reports.

Again, the report confound appears.

A region can be crucial for telling us about experience without being crucial for experience itself.

Lesions create both evidence and ambiguity

If damage to a brain area destroys conscious experience while leaving first-order processing intact, that would strongly support a higher-order role.

But human lesions are messy.

Brain damage affects networks, not isolated theory boxes.

Compensation occurs.

Tasks differ.

A patient may lose metacognitive accuracy without losing basic conscious experience.

Or lose the capacity to report reliably.

Therefore lesion evidence rarely maps cleanly onto a single philosophical theory.

It can constrain models.

It rarely settles them.

No-report experiments challenge strong frontal accounts

Some no-report studies find conscious-perception-related activity in posterior sensory regions without the frontal signatures seen when participants explicitly report what they saw.

This appears more comfortable for Recurrent Processing or posterior-hot-zone accounts.

Higher-order theorists respond in several ways.

Internal higher-order states may still occur without overt report.

Frontal signals may be more subtle than standard measures detect.

Some higher-order processing may be distributed rather than confined to classic prefrontal areas.

The debate shows why theories must distinguish their functional claim from one convenient neural localization.

The 2025 adversarial study also matters here

The large Cogitate experiment formally tested GNWT and IIT, not HOT as the primary adversaries.

But the paper explicitly notes overlap.

Some HOT implementations share predictions involving prefrontal cortex.

The study’s limited representation of certain conscious dimensions in PFC therefore creates indirect pressure on strong frontally localized higher-order accounts.

Indirect is the key word.

The experiment was not a decisive test of the HOT family.

It tells theorists what future versions have to explain.

What would higher-order representation actually add?

Suppose a first-order visual state encodes a faint red circle.

A higher-order representation might encode something like:

the system is currently in a visual state with these properties.

Why is that useful?

Because the brain needs to distinguish:

signal from noise;

perception from imagination;

reliable from unreliable evidence;

self-generated from external information;

strong from weak representation.

A higher-order layer can evaluate the state of the first-order system.

This is functionally similar to monitoring.

The controversial step is identifying that monitoring architecture with consciousness.

The theory has an attractive explanation for hallucination

Imagine the higher-order system represents the brain as seeing something even when the first-order sensory state is weak, noisy or absent.

What would experience be like?

Some HOT models allow the higher-order representation to influence conscious appearance.

This provides a framework for:

• hallucination;
• perceptual confidence;
• mismatch between sensory evidence and experience.

But explanatory flexibility is not evidence by itself.

A model can explain many phenomena after the fact.

The strongest tests require predictions that distinguish it from competing models.

Emotions create an even harder test

You feel fear.

Is the feeling produced directly by subcortical threat circuits?

Or do those circuits provide nonconscious inputs that are assembled into a higher-order conscious state?

Joseph LeDoux and Richard Brown proposed a higher-order account of emotional consciousness in which subcortical survival circuits are not themselves the conscious feeling.

Instead, conscious emotion emerges through broader cortical representation.

This approach sharply separates:

defensive behavior

from

the conscious experience of fear.

That distinction has influenced debates in affective neuroscience.

It remains contested.

HOT attacks the “magic transition” problem differently from GNWT

GNWT says information becomes conscious when globally broadcast.

RPT says recurrent sensory processing may be enough.

IIT says intrinsic causal structure is consciousness.

HOT says a first-order state becomes conscious when the system represents itself as being in that state.

Every theory therefore places the transition somewhere different.

This is useful.

It converts the vague question:

“Where does consciousness happen?”

into several competing mechanisms.

But does the higher-order state explain feeling, or only awareness of feeling?

This is the philosophical objection in its sharpest form.

Suppose the brain has:

first-order red representation;

higher-order representation that it is in a red state.

Why should either representation produce the phenomenal redness?

Perhaps HOT explains awareness of a mental state but not the existence of experience itself.

Higher-order theorists reject that split.

They argue that being in the right higher-order relation is what makes the first-order state conscious.

The dispute mirrors the access-versus-phenomenality problem from Global Workspace Theory.

Animal consciousness becomes a demanding case

Many animals display:

perception;

learning;

pain behavior;

decision-making.

Do they possess the higher-order architecture HOT requires?

If the required representation is minimal, perhaps many do.

If it requires sophisticated metarepresentation, perhaps fewer.

The theory therefore needs biological specificity before it can carry ethical conclusions.

You cannot infer:

“this species lacks consciousness”

merely because one human-style metacognitive task fails.

Failure of task performance may reflect motivation, motor limits or task design.

AI exposes the same boundary

A future AI could represent both:

the external world;

its own internal states;

its uncertainty;

its confidence;

its processing history.

Would that satisfy HOT?

Depends on the theory.

And even if the functional architecture were present, philosophers would still disagree about whether implementing the representation is sufficient for genuine experience.

This is why consciousness theories should not be converted directly into AI-consciousness verdicts.

They are frameworks under test.

The DarkBrain model: first-order, higher-order, access, evidence

First-order

What does the system represent about the world or body?

Higher-order

Does it also represent its own state?

Access

What new behavior or report becomes possible?

Evidence

What observation would show that the higher-order representation is necessary for consciousness rather than only for report or metacognition?

That final question is the empirical heart of HOT.

What the evidence does not support

The evidence does not support these universal claims:

• prefrontal cortex is proven to generate consciousness;
• confidence equals consciousness;
• metacognition equals consciousness;
• animals without human-like self-reflection are unconscious;
• every higher-order representation is conscious;
• every conscious state requires explicit thought;
• no-report studies have disproved HOT;
• the 2025 adversarial study directly tested and refuted all HOTs.

The theory family remains active because the distinction between processing a state and being aware of that state is genuinely difficult to explain away.

Why this matters

Higher-Order Theories make a strange proposal feel almost obvious.

The brain does not merely model the world.

It models its own relation to the world.

Somewhere in that second layer may lie the difference between information passing through the system and information appearing to the system.

Whether that extra representation is sufficient for consciousness remains unsettled.

But the question changes how we think about awareness.

Perhaps consciousness is not only seeing.

Perhaps it is the brain, in some minimal sense, knowing that seeing is happening.

Continue exploring

Next: Recurrent Processing Theory: Can Local Feedback Be Enough for Awareness?

The final theory pushes back against both global broadcast and higher-order representation. It asks whether the critical transition may happen much earlier, inside sensory cortex itself, when feedforward activity begins to loop back through the system.