Fairy Note
Predictive Worlds: Karl Friston, Worldhood, AI, Dementia and the Cognitive Image
Part I: Karl Friston, Predictive Processing, and Why Reality Is Not a Hallucination
Opening Position: A Conversation with William
This note begins with a conversation rather than a text.
While discussing predictive processing with my father, the cognitive scientist and psycholinguist Professor William Marslen-Wilson, I found myself struggling to understand the relationship between contemporary neuroscience, phenomenology, dementia, and artificial intelligence. During the conversation he repeatedly referred to Karl Friston’s work on predictive processing and explained that perception operates through interactions between higher and lower levels within a hierarchical system. Unfortunately I failed to record the most detailed part of the explanation, but enough remained to identify the conceptual structure he was describing.
The conversation became particularly interesting when the phrase “reality is a hallucination” emerged. This phrase, often associated with Anil Seth’s popular presentations of predictive processing, immediately provoked a strong reaction. My father dismissed it as misleading. His objection was not to predictive processing itself but to the philosophical implications people often derive from it. Reality, he insisted, is not a hallucination. Hallucinations are something quite different. Yet at the same time he fully accepted that perception is a constructive process rather than a passive recording of the world.
This apparent contradiction reveals something important.
To understand why, we need to begin with Karl Friston.
Karl Friston and the Predictive Brain
Karl Friston is arguably the most influential neuroscientist working today. His Free Energy Principle and associated theories of predictive processing have become increasingly central across neuroscience, cognitive science, psychiatry, artificial intelligence, and philosophy of mind.
The traditional picture of perception imagined the brain as something like a camera.
Reality exists.
Light enters the eye.
Information travels inward.
The brain reconstructs what is already there.
Predictive processing reverses this sequence.
The brain does not begin with sensation.
The brain begins with prediction.
According to Friston, perception emerges through the continual interaction between expectations and incoming sensory information. The brain is constantly attempting to predict what the world will be like before sensory information fully arrives.
This is why predictive processing is often described as a model-based theory of perception.
The brain is continuously constructing models of reality.
Sensory information does not create these models from scratch.
Instead, sensory information updates them.
Perception therefore becomes a process of continual negotiation between prediction and correction.
The Hierarchical Structure
The part of the conversation I found most difficult involved what William described as higher and lower levels within the system. Although I failed to record the full explanation, this corresponds closely to the hierarchical architecture proposed within predictive processing.
Very roughly speaking, the hierarchy works as follows.
Lower levels process relatively immediate sensory information.
These might involve:
- edges
- colour contrasts
- movement
- sounds
- bodily sensations
- tactile information
Higher levels process increasingly abstract structures:
- objects
- faces
- rooms
- people
- situations
- memories
- narratives
- expectations about the world
The important point is that information moves in both directions.
Predictions flow downward.
Prediction errors flow upward.
The higher levels generate expectations about what should be present.
The lower levels compare incoming sensory information against those expectations.
When prediction and sensation align, the model is reinforced.
When they fail to align, prediction errors move upward through the hierarchy and the model must be adjusted.
Perception therefore becomes neither purely bottom-up nor purely top-down.
It emerges through their continual interaction.
The world is not simply imposed upon the brain.
Nor is the world simply invented by the brain.
Reality emerges through negotiation.
Why the Brain Must Predict
One reason predictive processing has become so influential is that it solves a fundamental computational problem.
The sensory world contains vastly more information than any nervous system could process directly.
Every second we are confronted by:
- visual information
- auditory information
- bodily information
- proprioceptive information
- environmental information
- social information
If the brain attempted to process all incoming information equally, cognition would become overwhelmed.
Prediction provides a solution.
Instead of analysing everything, the brain predicts most things and focuses attention on unexpected deviations.
The world therefore becomes manageable.
The system conserves energy.
Uncertainty is reduced.
Meaning emerges.
In Friston’s language, the system attempts to minimise prediction error and reduce uncertainty about the world.
The Controlled Hallucination
This brings us to Anil Seth.
Seth’s famous phrase that perception is a “controlled hallucination” has become enormously influential. It appears repeatedly in popular science writing, AI discourse, neuroscience podcasts, and discussions of consciousness.
At first glance it seems to follow naturally from predictive processing.
If perception emerges through internally generated models, then perhaps what we experience is a kind of hallucination constrained by sensory input.
Yet this is precisely the point at which William objected.
And I increasingly think his objection is important.
The phrase is memorable.
The phrase is provocative.
The phrase is probably philosophically misleading.
The problem lies in the word hallucination.
Hallucinations ordinarily refer to perceptual experiences that are not constrained by the world.
Predictive processing does not claim this.
Predictive processing claims that perception is actively constructed, but that construction remains constrained by reality.
Reality pushes back.
Reality generates prediction errors.
Reality forces models to adapt.
Without this continual correction the system would indeed drift into hallucination.
The very reason predictive processing works is because the world continually resists prediction.
The phrase “controlled hallucination” therefore risks overstating the internal side of the equation while underplaying the role of the world itself.
This is precisely where phenomenology begins to re-enter the discussion.
Why Phenomenologists Are Uncomfortable
At first glance predictive processing and phenomenology seem remarkably similar.
Both reject passive perception.
Both reject simple stimulus-response models.
Both recognise that perception is structured and active.
Yet there is an important difference.
Predictive processing often begins with the brain.
Phenomenology begins with lived experience.
Merleau-Ponty famously argued that the body is our general medium for having a world.
This statement changes everything.
Perception is not something a brain does inside a skull.
Perception is something an embodied being does within a world.
The body is already immersed in meaning.
Already oriented.
Already situated.
Already engaged.
Thomas Fuchs extends this position into contemporary cognitive science.
Fuchs repeatedly criticises accounts of cognition that reduce human experience to internal computation.
For him, meaning emerges through embodied engagement with a world rather than through detached information processing alone.
This is where predictive processing and phenomenology begin to diverge.
Friston asks:
How does the brain model reality?
Merleau-Ponty asks:
How does a living body inhabit reality?
Fuchs asks:
How does meaning emerge through embodied worldhood?
These are not contradictory questions.
But neither are they identical.
Why This Matters for Dementia
The significance of this distinction becomes visible through dementia.
Dementia reveals something extraordinary.
It demonstrates that perception is not merely sensory.
It is world-constituting.
When Lolly mistakes William for Fake William, the problem is not that she cannot see him.
She sees him perfectly well.
The problem lies elsewhere.
The predictive structures through which he becomes William have destabilised.
Recognition fails.
Meaning fails.
Worldhood fails.
Similarly, when she asks repeatedly to go home, the problem is not geographical.
The house remains physically present.
The difficulty is phenomenological.
Home no longer functions as an orienting structure within her world.
The predictive model and the embodied world have drifted apart.
This is why dementia increasingly appears less like memory loss and more like disruption of world-maintenance.
The world remains present.
Its coherence begins to fracture.
In this sense dementia becomes an extraordinary test case for predictive processing.
It reveals how much labour is normally required to maintain a stable reality.
Preliminary Conclusion
Karl Friston’s predictive processing offers one of the most powerful contemporary accounts of perception available.
It explains:
- why perception is active
- why prediction matters
- why the brain constructs models
- why uncertainty is central to cognition
Yet predictive processing alone remains incomplete.
It explains how perception is organised.
It does not fully explain why certain things matter.
It struggles with grief.
It struggles with attachment.
It struggles with mourning.
It struggles with the collapse of home.
It struggles with Fake William.
This is why phenomenology and psychoanalysis remain necessary companions.
Predictive processing explains the architecture.
Phenomenology explains the world.
Psychoanalysis explains the emotional investment that makes that world matter.
For Phantom Mirror this distinction becomes crucial.
The project does not simply investigate perception.
It investigates what happens when perception, recognition, attachment, memory, and worldhood begin to unravel together.
Predictive processing provides one part of that story.
The next task is to understand how worldhood itself emerges, and why phenomenologists such as Merleau-Ponty and Thomas Fuchs remain indispensable for explaining what Friston alone cannot.