Intro to Predictive Processing

Slide 1 of 17

Opening Framework

Predictive Processing

Active Inference

Embodied Cognition

Generative Mental Model

Predictive Coding

lots of fancy names, but it comes down to

Your brain doesn't wait for the world to come to it,
it goes out to find it

the system works like this:

  1. Build internal models of reality
  2. Generate predictions from those models
  3. Compare predictions to sensory input
  4. Prediction errors = the difference
  5. Precision weighting = how much you trust the error
  6. Update model OR ignore the error
  7. Action follows the model

not passive reception
active construction

your perceptions ARE your predictions
until proven wrong

The Architecture

hierarchical structure
bidirectional processing

forward connections carry ONE thing

  • prediction errors
  • that's it
  • nothing else goes up

backward connections carry PREDICTIONS

  • from higher levels
  • to lower levels
  • constantly

this is where it gets interesting

error units - superficial pyramidal cells

encode precision-weighted prediction errors


representation units - deep pyramidal cells

encode predictions about causes


functionally distinct populations
same level, different jobs

the prediction travels DOWN
the error travels UP

when prediction matches input?
error units go quiet
explaining away

when prediction misses?
error signal propagates
demands attention
forces update

The Architecture

BUT

neural connections BACK 4 to 1
four times more backward than forward

WHY

neurons are EXPENSIVE

2% body weight
20% of your energy budget

why waste that on redundant bottom-up processing

evolutionary answer:

don't send what can be predicted


only send SURPRISE
only send ERROR
only send what matters

the brain implements efficient coding
bandwidth is precious
prediction errors are compressed information

this architecture answers:

  • why so many backward connections
  • why context modulates everything
  • why attention changes what you see
  • why expectations shape perception

predictive brain = metabolically efficient brain

The Mechanism

processing cycle runs continuously

  1. higher level generates prediction of lower level activity
  2. prediction transmitted via recurrent connections
  3. lower level computes mismatch
  4. if error is precise enough - propagates forward
  5. higher level adjusts to minimize error

all levels simultaneously
all the time
no central coordinator

precision weighting is the key

not all errors matter equally
brain estimates reliability of prediction errors

  • inverse of variance
  • confidence in the signal

high precision = high gain on error units

  • signal amplified
  • forces model update
  • this IS attention

low precision = low gain on error units

  • signal dampened
  • prediction dominates
  • error ignored

neuromodulators encode precision

  • dopamine
  • acetylcholine
  • serotonin
  • maybe neural synchronization

attention = precision optimization

you're not amplifying signal
you're changing confidence in prediction error

Perception as Inference

visual input is 2D
perception is 3D

that extra dimension?
generated by your model

photons hit retina

  • ambiguous
  • inverse optics problem
  • infinite possible causes

your percept = maximum a posteriori estimate

  • given sensory evidence
  • given prior expectations
  • given current context
you never see the world directly
you see your brain's best guess
constrained by sensory input

Think of visual illusions.
They aren't failures of the system.
They are optimised solutions based on
environmental stimuli and past experience.

Action and Motor Control

ideomotor theory on steroids

throw something
catch something

you don't calculate trajectories
you don't compute joint angles
you don't plan muscle activations

you PREDICT proprioceptive consequences

  • what successful catching feels like
  • what the arm position should be
  • what joint angles would occur

then your body minimizes prediction error in the motor domain

prediction: arm here, ball arriving there
sensory feedback: arm NOT here yet
error signal triggers movement
movement continues until error resolves

you downplay sensory evidence that you aren't moving
then generate movement that fulfills the prediction

perception: adjust predictions to match world
action: adjust world to match predictions


same computational principle
opposite direction of fit

like we are the puppet and the puppeteer
both at once

when you miss?
massive prediction error
attention increases
precision weights adjust
repeated action

learning = updating model based on precision-weighted error

The Survival Logic

This makes even more sense when you think about the system priority - Survival

imagine processing every photon
every sound wave
every tactile input
from scratch
every millisecond


metabolic bankruptcy

evolutionary solution:

build models
predict input
process only deviations

  • saves time - instant responses
  • saves energy - minimal computation
  • builds skill - models improve with experience
identifies potential threat

BANG!!!!

your system predicts: TIGER
massive error signal
attention spike
action cascade


turns out - Dave dropped something

system updates:

  • false alarm
  • reduce threat weighting
  • adjust precision

if you think you don't do this - bad news
you're descended from ancestors who predicted tigers in bushes and ran,
not from ancestors who carefully evaluated each stimulus and got eaten

false positives? survived
false negatives? eliminated


your threat detection is biased
by design
not by malfunction

Nociception as Prediction Error

Imagine you are walking along about to stand on a nail…

current model

  • ground safe
  • skin intact
  • tissue healthy
  • walking possible

sensory input arrives

  • sharp object
  • tissue deformation
  • nociceptor activation

prediction error generated

  • ground NOT safe
  • skin potentially damaged
  • danger present

this is nociception
information about actual or potential tissue damage
NOT pain yet

the error signal is precise

  • clear source
  • unambiguous meaning
  • high precision weighting
attention

increases gain on this error
changes model
compels action

subconscious level:

  • inflammatory response
  • protective mechanisms
  • tissue repair initiation

conscious level:

  • pain perception
  • protective behavior
  • movement alteration

automatic - not a choice

new model assembled:

  • bad things present
  • protection required
  • altered action necessary

action follows:

  • limping
  • avoidance
  • reduced loading

Acute Injury Loops

GOOD - SAFE - HEALTHY
EVIL - DANGEROUS - PROBABLY DYING
GOOD - SAFE - HEALTHY

perception/action loops
updated by prediction errors

middle state:

  • systems alarmed
  • infection risk
  • danger signals
  • imminent threat model

but also

new sensory information arrives:

  • tissue healing
  • inflammation resolving
  • weight bearing tolerated
  • movement improving

new prediction errors generated:

  • healing occurring
  • threat decreasing
  • function returning

precision weighting shifts:

  • danger signals downweighted
  • recovery signals upweighted

model updates:

  • reverts toward baseline
  • safety predictions increase
  • threat predictions decrease

off we go

this is normal acute pain
error-driven updating
model plasticity intact
system functioning

Failed Updating - Uncertainty

GOOD - SAFE - HEALTHY
EVIL - DANGEROUS - PROBABLY DYING
UNSURE - UNSURE - UNSURE

what happens here

same injury
same healing process
same objective improvement

BUT

less certain information
OR
new information, less weighting

due to:

  • previous negative experience
  • conflicting sensory signals
  • competing information sources
  • contextual factors
  • prior beliefs about healing
  • other stressors present
no model updating occurs

system maintains danger weighting
downgrades safety information
uncertainty persists

precision estimation fails:

  • can't determine reliability
  • can't weight signals appropriately
  • can't resolve ambiguity

prediction errors exist
but don't drive change

model stuck
danger predictions maintained
protective actions continue

Failed Updating - Escalation

GOOD - SAFE - HEALTHY
EVIL - DANGEROUS - PROBABLY DYING
THE DEVIL - MORTAL PERIL - IMMINENT DEATH

perception/action INCREASES pain prediction
actions fit this catastrophic model

mechanisms:

  • avoidance behavior
  • disuse
  • fear of movement
  • hypervigilance
  • protective guarding

anything positive gets downweighted:

  • improvement signals ignored
  • functional gains dismissed
  • recovery evidence minimized
in the presence of strong priors
evidence gets filtered

self-fulfilling prophecy:

  • prediction = pain likely
  • attention increases on confirming information
  • precision increases on threat signals
  • action predicts more pain
  • pain occurs
  • model confirmed
again - automatic

not conscious psychological choice
not willful catastrophizing

biological variation in system function:

  • precision weighting abnormal
  • prior beliefs too strong
  • model updating impaired
  • learning rate altered

influenced by everything everywhere all at once:

  • stress
  • sleep
  • inflammation
  • previous trauma
  • social context
  • beliefs about pain
  • healthcare messages
  • life circumstances

Chronic Pain as Prediction

pain shouldn't persist when tissue heals
but it does

predictive processing explains this:

hierarchical priors become entrenched

high-level predictions:

  • "my back is damaged"
  • "movement causes harm"
  • "pain will persist"
  • "recovery impossible"

these generate lower-level predictions:

  • proprioceptive threat
  • anticipated nociception
  • protective motor patterns

which generate prediction errors:

  • not enough danger signal arriving
  • system INCREASES precision on threat-consistent input
  • system DECREASES precision on safety-consistent input

model becomes self-maintaining

interoceptive predictions also critical:

  • predictions about internal body state
  • inflammatory status
  • tissue condition
  • autonomic activation

emotional feelings = interoceptive inference
anxiety, fear, distress = prediction errors in interoceptive domain

anterior insular cortex:

  • interoceptive prediction errors
  • affective content integrated with predictions
  • emotional experience constructed
whole system predicting threat
at multiple hierarchical levels
sensory, motor, interoceptive, affective

all reinforcing
all self-confirming

Precision in Chronic Pain

attention alters everything

chronic pain shows precision dysfunction:

too much precision on:

  • threat-related signals
  • pain-consistent information
  • danger cues
  • movement-related sensations

too little precision on:

  • safety signals
  • improvement indicators
  • functional capacity
  • positive sensory input

this isn't cognitive
this is built in system function

gain settings on error units:

  • threat errors amplified
  • safety errors dampened

neuromodulation potentially abnormal:

  • dopaminergic signaling altered
  • noradrenergic tone changed
  • serotonergic function modified

neural synchronization patterns:

  • gamma band activity increased in pain networks
  • might encode precision
  • might maintain threat predictions
the system CAN'T update properly
not WON'T
CAN'T

precision weighting prevents learning
even when evidence available
signals don't have appropriate impact

Therapeutic Implications

if pain = maladaptive prediction
treatment = update the model

provide prediction errors that can drive change

exposure-based approaches:

  • generate safety prediction errors
  • movement doesn't cause harm
  • function possible without catastrophe
  • tissue can load without damage

BUT - precision matters


if precision on threat too high
safety errors get ignored
exposure fails

need to:

  • reduce precision on threat predictions
  • increase precision on safety predictions
  • create conditions for model updating

how:

context manipulation:

  • safe environment
  • trusted therapist
  • clear explanation
  • graded progression

attention training:

  • redirect to safety-consistent information
  • reduce hypervigilance
  • modify precision weighting

pain education:

  • update high-level priors
  • "pain doesn't equal damage"
  • "the system can learn"
  • changes conceptual landscape

multisensory integration:

  • visual feedback
  • mirror therapy
  • virtual reality
  • creates rich prediction errors across modalities
the goal:
shift precision landscape
enable model updating
restore adaptive prediction

not "dealing with pain"
not "managing symptoms"
updating the generative model

Action as Inference

motor control = active inference

you don't execute motor commands
you fulfill proprioceptive predictions

prediction: hand at cup
current state: hand not at cup
minimize error:

  • option 1: change prediction (perception)
  • option 2: change state (action)

in motor domain - option 2

classical motor control:

  • plan trajectory
  • compute commands
  • execute movement
  • compare feedback
  • adjust

active inference:

  • predict sensory consequences
  • suppress prediction errors through movement
  • movement continues until prediction fulfilled

mathematically equivalent to optimal control
but mechanistically inverted

reflex arcs explained:

  • spinal prediction errors
  • local minimization
  • no cortical involvement needed

cerebellar function:

  • forward models
  • predicting sensory consequences of movement
  • providing predictions to minimize error

basal ganglia:

  • action selection
  • which predictions to pursue
  • precision weighting on motor predictions
whole motor system:
hierarchical predictive control
all minimizing prediction error
just in different domains

Implications Summary

perception = inference about causes
action = inference through consequences

attention = precision optimization
learning = error-driven model updating

the brain is not passively representing
it's actively constructing

based on:

  • evolutionary priors
  • lifetime learning
  • current context
  • precision-weighted evidence

your experience of reality:

not what's out there
what your model predicts
constrained by sensory input

Clinical Reality

think of some patients

unhealthy, overweight, unemployed, smoking and drinking for Australia
chronic pain persisting
multiple failed treatments
every intervention somehow makes it worse

predictive processing lens:

their model predicts threat

  • hierarchical priors entrenched
  • precision massively weighted toward danger
  • safety information filtered out
  • every experience confirms threat

their actions fit that model

  • avoidance maintains beliefs
  • disuse prevents disconfirmation
  • protective behavior predicts harm

context reinforces everything

  • life stress
  • poor sleep
  • inflammation
  • social isolation
  • economic pressure
  • previous trauma
  • healthcare catastrophizing

influenced by everything everywhere all at once

not psychological weakness
not poor coping
biological variation in system function

precision weighting abnormal
model updating impaired
learning mechanisms disrupted


automatic
not conscious choice

the system doing what systems do:
minimize prediction error
given the priors it has
given the precision landscape
given the available evidence

bad things happen
models persist
suffering continues

this is where treatment meets theory
this is why understanding mechanism matters
this is predictive processing in the clinic
↓ Scroll to read more