How the Brain Constructs Reality: Carl Friston on Predictive Processing

---

---

---

---

---

Executive Summary

The Setup

This is an interview with Carl Friston, a prominent neuroscientist who developed the "free energy principle"—a mathematical framework for understanding how brains (and all self-organizing systems) work. The conversation covers how our brains construct our experience of reality through prediction rather than passive observation.

The Core Idea: Prediction-First Perception

Your brain is constantly guessing:

• The content explains that your brain doesn't wait for sensory input to build a picture of reality
• Instead, it continuously generates predictions about what it expects to sense
• When sensory data arrives, your brain subtracts the prediction from the actual input
• The difference (called "prediction error") is the only newsworthy information
• Your brain then updates its internal model to reduce future prediction errors

Why this matters:

• Your brain is "skull-bound"—trapped inside your head with no direct access to the outside world
• You only get sparse sensory information through limited channels (eyes, ears, skin, etc.)
• The brain solves this problem by building an internal model and constantly testing it against reality

The Vision Example: Palpating the World

How we actually see:

• The content notes that our eyes "palpate" the world every 250 milliseconds (four times per second)
• We only see a tiny portion of our visual field clearly at any moment
• The brain stitches together a coherent scene by predicting what it would see if it looked at different spots
• If predictions match reality, the brain accumulates evidence that its internal model is correct

The Free Energy Principle

The mathematical framework:

• Friston explains this as a physics principle that applies to all self-organizing systems
• "Free energy" in this context means prediction error or surprise
• Any system that maintains its existence must minimize free energy (keep prediction errors low)
• This applies to everything from single cells to humans to potentially economies

Two ways to minimize prediction error:

1. Perception: Change your internal model to match what you're sensing

2. Action: Change what you're sensing by moving/acting to match your predictions

The scale question:

• The content claims this principle applies to anything you can distinguish from its environment
• At very small scales (quantum) or very large scales (planetary motion), different rules apply
• At the "intermediate scale" (roughly human-sized), something interesting happens: action becomes separated enough from internal processing that creatures can plan—they infer their own future actions

Consciousness and Agency

What makes something an agent:

• Friston distinguishes between simple systems (like thermostats) and true agents
• Agents have internal models of their own future and can select between different possible actions (policies)
• The content suggests that consciousness requires agency—you need this kind of deeply structured internal model
• Not all agents are conscious, but all conscious beings are agents

Levels of awareness:

• The content references philosopher Thomas Metzinger's distinction between "transparent" and "opaque" perception
• Transparent: You just see something
• Opaque: You know that you're seeing something—awareness of your own perception
• This requires hierarchical brain structure where higher levels can observe lower levels

Attention as Internal Action

What attention actually is:

• The content explains attention as the brain choosing what prediction errors to ignore
• Higher brain levels contextualize lower levels by selecting newsworthy information
• This is "action on the inside"—you're doing something to your own mental processes
• This internal action is key to moving from mere sensation to conscious awareness

Applications: Sleep, Dreams, and Creativity

Why we dream:

• Friston offers a speculative explanation based on the free energy principle
• Free energy can be broken down into accuracy (matching data) and complexity (simplicity of model)
• When you sleep (no sensory data coming in), accuracy is off the table
• The brain can still minimize free energy by minimizing complexity
• Dreams might be the brain testing whether it can explain imagined scenarios more simply
• This allows "synaptic pruning"—removing unnecessary neural connections

The daily cycle:

• Eyes open: accumulating associations and explanations
• Eyes closed: housecleaning, removing redundant connections
• This prepares a simpler, more generalizable model for tomorrow

Mental Illness as False Inference

Psychiatric symptoms as prediction errors:

• The content frames psychopathology as "false inference"—bad belief updating
• Type 1 errors (inferring something is there when it isn't): hallucinations, delusions
• Type 2 errors (inferring something isn't there when it is): neglect, dissociative symptoms

The mechanism:

• Mental illness might stem from failing to ignore irrelevant information
• If you can't filter out unimportant prediction errors, you'll generate false explanations
• Example given: Feeling unexplained arousal while watching TV news, then concluding the newscaster is talking to you personally
• This is mathematically optimal given the unattenuated sensory signal—it's the best explanation available if you can't ignore the data

Illusions and Prior Beliefs

Why illusions persist:

• The content explains that illusions reveal the brain's prior commitments—built-in assumptions from evolution and learning
• Even when told something is an illusion, you can't override the automatic inference
• There's a divide between transparent qualitative experience and declarative knowledge
• You can't "talk yourself out" of an illusion because the mechanisms are distributed throughout the brain

The Metaphysics: Are We Trapped?

The reality question:

• Friston acknowledges we are "trapped in our heads" in some sense
• There's a world outside, but we never directly access it
• Instead, there's "mutual entrainment" between brain states and world states
• Uses the analogy of Huygens's clocks—two pendulum clocks on the same wall eventually synchronize
• This is a "balanced dance," not one-way control from world to brain

Active perception:

• The content emphasizes that perception is "quintessentially active"
• We choose what to sample from the environment
• We cause our own sensations by moving our eyes, turning our heads, etc.
• Friston pushes back against "controlled hallucination" terminology—prefers "entrainment"

---

Key Takeaways

1. Your brain predicts first, senses second — Perception works "inside-out" by generating predictions and comparing them to incoming data, not by passively absorbing information.

2. Prediction errors drive learning — The difference between what you expected and what you sensed is the only information your brain uses to update its model of reality.

3. Action and perception serve the same goal — Both minimize prediction error: perception changes your model, action changes what you sense.

4. Planning requires self-modeling — True agency emerges when systems are large enough that they can infer their own future actions as part of their world model.

5. Mental illness may be attention dysfunction — Psychiatric symptoms might result from failing to ignore irrelevant prediction errors, leading to false but mathematically optimal explanations.

6. Dreams might simplify your brain — Sleep may allow the brain to minimize model complexity by pruning unnecessary neural connections, preparing a cleaner model for the next day.

7. We're synchronized with reality, not controlled by it — The relationship between brain and world is mutual entrainment, like pendulum clocks that gradually swing together.

---

Should You Watch?

Yes, if:

• You're curious about cutting-edge theories of consciousness and perception
• You want to understand how neuroscience connects to physics and philosophy
• You're interested in computational psychiatry or AI/machine learning parallels to brain function
• You can tolerate some jargon for the sake of encountering genuinely novel ideas

No, if:

• You want concrete, evidence-based neuroscience without speculative frameworks
• You prefer practical applications over theoretical models
• You find heavy technical terminology frustrating (terms like "Markov blanket" and "generative model" appear frequently)
• You're looking for settled science rather than contested theories

---

---