W6 - Predictive Processing Flashcards
(103 cards)
1
Q
What is the premise of the Theory of Predicative Processing (Fristen)?
A
- the brain only has indirect access to information from its environment
- We do not actively perceive reality as it is, we perceive the inputs from our senses, eg., Plato’s cave
- The brain must gather information from a noisy, complex and uncertain environment
- Process information and act based on information for continued survival
- To achieve this, the predictive processing framework suggest the brain functions as a hierarchical bayesian inference device
2
Q
Why is predictive processing hierarchical?
A
Each part of the brain has a model about the inputs it expects to receive, organised in cortical hierarchy
brain regions processing sensory information mainly process low level isolated percepts of sensory inputs
while more frontal regions process complex syntheses of information from many sensory inputs, combined with memories, predictions and goals
3
Q
Why is predictive processing baysian?
A
- beliefs are updated based on the the strength of the prior belief and the strength of new evidence
- e.g., you confidence in the original belief + new evidence + strength/precision of evidence = determines if you update your hypothesis
Example = Seeing a cat clearly with new evidence = cat
Seeing a cat in corner of eye + have a dog = might be a dog
4
Q
Why does predictive processing have Inference device?
A
our brains can’t directly access the environment, so everything is done based on inferences based on our sensory inputs
Predictive processing must be done efficient, without spending all body’s finite energy, and can’t update continuously
5
Q
What is the Metaphor: Zoom calls?
A
- zoom/call does NOT send through every pixel through the screen, and only transmits the pixels that changes, = maximises efficiency, saves energy
- If you stay very still, all the pixels will keep displaying as they were
It would take a lot of bandwidth energy to continuously update every pixel in a zoom call
6
Q
When does PP model update its beliefs?
A
it ONLY UPDATES PREDICTIONS BASED ON ERRORS FROM SENSORY INFORMATION WHERE THE SENSORY INFORMATION WASN’T PREDICTED
If the model already knew the sensory input, it doesn’t update at all, maximises efficiency
7
Q
What is a ‘surprise’?
A
the long term average of these prediction errors - brain tries to minimise
reading: non preferred outcomes
8
Q
What is ‘free energy’?
A
its energy it uses associated with dealing with prediction errors
9
Q
How long do prediction errors take?
A
about 50ms
10
Q
What happens when predictions are incorrect and how is it evaluated?
A
- prediction errors are passed UP the cortical hierarchy via synaptic depolarisation, updating the predictive model
- In a hierarchical bayesian manner, predictions are updated depending on the top-down predicted PRECISION of the EVIDENCE for the error relative to the STRENGTH of the PRE-EXISTING belief/prediction
- This bayesian inference allows the higher region to decide whether to update the predictive model or not
11
Q
What do Top down predictions do?
A
Top down predictions modulate strength of the prediction error AND direct precision of processing based on internally held goals,
12
Q
What kinds of predictions are in the frontal lobe?
A
- More abstract models of a person’s interactions, abstract thought and planning are made by frontal regions
- Abstract predictions are transmitted to the posterior regions, to modulate the precision of prediction errors/synaptic gain for specific sensory information, like an attention function
13
Q
What do posterior regions do?
A
Posterior regions process sensory information - when the model is incorrect, prediction errors from sensory regions pass up to the frontal regions
14
Q
Does the brain does passively receive information? -
A
- NO -it receives information AND change the environment so the environment will produce new evidence/info that might be more aligned with our predictions and future state it wants to find itself in
15
Q
What are the two types of inference the brain makes to align with its internal goals?
A
- Perceptual inference: Updates the predictive model based on new prediction errors
- Active inference = Use motor control functions to act, to attempt to allow future sensory inputs to be more aligned with predictions
*Designates actions to best lead to a desired future state
* minimise average prediction error
16
Q
Why might active inference produce temporary in prediction error to minimise overall prediction error in the long term?
A
sometimes active inference might involve producing temporary increases in prediction error by seeking information in uncertain environments,
allowing updated trajectories towards desired states, rather than just missing short term prediction error
17
Q
A reflex response from jumping back from touching a stove is an example of a…
A
Simple example of active inference, to match prediction that we are free from burning pain
18
Q
Wanting to get a good grade, implementing sequences of action trajectories to update predictive models with info to pass your exams is a…
A
Complex example of active inference
19
Q
What is the IMPLICATIONS OF PREDICTIVE PROCESSING regarding entropy?
A
By adhering to this driving force of trying to minimise long-term average prediction error by performing active inferences - the brain can continue to exist despite the tendency of the universe to learn towards entropy
Counter to law of entropy, we manage to continuously self-organise via performing predictive processing, causing active inference using energy to maintain order predicted in one’s own body
20
Q
What happens if we don’t follow the prompts of the prediction errors coming in and implement new active inference policies to get into a better state of prediction?
A
you may find yourself in a state of complete environmental reality that doesn’t match our reality
EG., why trauma is so destructive
21
Q
What does dynamical systems theory claim?
A
- That complex systems like humans pursuing limited sets of preferred / predicted states, are called attractors, “attracted to predictive states to allow us to continuously perceive evidence aligning with our current reality”
Inference is close to a theory of everything, evolution, consciousness and life = Foundation of what it means to be alive
22
Q
What are preferred predicted states called?
A
attractors
23
Q
HOW DID INFERENCE HAPPEN BEFORE INFERRERS EXISTED, IN AN ENTROPIC UNIVERSE?
Think - evolution
A
- Our brain system has evolved to continues update its predictive model from the environment to make active inferences and verify its predictions
- Brains that were not effective at doing predictive processing DID NOT SURVIVE and pass on code without predictive processing
24
Q
What is an organism’s survive based on?
A
- Choice organism that are more likely to occupy their ecological niche
adaptive fitness is thus the likelihood of finding a phenotype in its environment - An organism’s survival is about the likelihood of being a good model for its niche, e.g., different birds on Galapagos Islands
25
What implications does PP theory have on the past/present/future?
it needs all conception of past, present and future
1. Brain cannot infer the consequences of its action unless it has a model of its FUTURE
2. To make better predictions of the future, we must also form a model of the PAST, for context and processing sensory input - need priori previous experience and genetic history
3. To enable the brain to take action to ensure its prediction for continued existence, it must also be aware of the PRESENT, its current internal state and bodily state, blood levels
26
Why do Predictive process components look a lot like the state of consciousness?
1. Conscious beings can comprehend the future, remember the past and introspect the current state of being
2. Conscious beings can also act to change their circumstances to make their internal preferences / predictions
27
What type of inference is consciusness supposed to be?
a process of active inference - connects past to present to future to better match our internal predictions for own continued existence
28
What is the most deeply held prediction most resistant to change?
the prediction of “I exist” - influences all other predictions
29
What 3 examples of altered consciousness change predictive processing?
1. Anesthesia
2. Psychedelic drugs
3. Trauma
30
What less damaging altered state of consciousness is said to impact predictive processing?
1. Long term meditation = can reduce a sense of self, from a conscious choice not to act in response to escalating prediction errors
thus updating the predictive model of the process leading to consciousness is voluntarily dissolving
31
Why does anesthesia disrupt ability to do predictive processing?
Anesthesia seems to be disrupting the brain’s ability to generate its predictive model, thus process of consciousness ceases
32
Why does psychedelic drugs disrupt ability to do predictive processing?
Psychedelic drugs drastically increase the synaptic gain such that
1. prediction errors SPEED UP the cortical hierarchy
2. top down generation is REDUCED
3. leading to increased sensory noise 4. A lack of prediction error suppression, leading to a distorted sensory experience
33
Why does TRAUMA disrupt ability to do predictive processing?
1. Trauma disrupts our predictions that we will be safe and free of pain and experience dignity and respect
2. Trauma reduces our ability to make active inferences to match environment to predictions and develop a perceived lack of agency or control over their own circumstances and bodies
34
PP model and attention: How do Top-down predictions change depending on our goals?
Top-down predictions can INCREASE the synaptic gain for goal-relevant predictions
And DECREASE synaptic gain for non-relevant predictions, increasing neural signal of relevant information based on internal goals, more evidence to make better predictions
35
PP model and attention: How do salient prediction errors influence our attention?
1. Highly salient prediction errors automatically engage processing prediction errors, going up cortical hierarchies while also SUPPRESSING other prediction errors
EXAMPLE - a loud noise distracts you from a person, as the prediction errors from the unexpected sound suppress other focuses of PP mechanisms
36
What are the 2 main concepts that predictive processing uses to help help explain consciousness, attention and behaviour?
1. Thermodynamics physiological truths
2. bayesian statistics
37
How are alpha waves involved in predictive processing?
Cortical travelling alpha waves from EEG progress from posterior to frontal regions seem to index prediction errors progressing up the cortical hierarchy
38
How to determine whether alpha travels from back to front of head?
We can measure whether there is INCREASING PHASE DELAY across the electrodes from the front to the back of the brain
39
How can prediction errors be measured in EEG?
Thus predictive errors (from sensory processing to updating prediction models in frontal regions) can be measured / indexed through alpha cortical travelling waves with increasing phase delay from electrodes from front to back of head
40
alpha CTW waves from posterior to frontal might be involved in...
the spreading inhibition of non-relevant neural activity to protect processing of prediction errors,
by increasing the signal to noise ratio for the purpose of processing prediction errors
41
Conversely, alpha CTW waves from frontal to posterior may reflect ...
top-down inhibitory activity to suppress predicted inputs, e.g., like the Zoom model receiving signals that frontal regions have already predicted, we don’t need to process them and they are suppressed
42
What 2 ways does taking DMT change the strength of different cortical travelling waves?
1. DMT/ psychedelic INCREASES the strength of forward CTW
* eliciting visual processing with eyes closed “seeing with the eyes closed” gives illusion of visual hallucinations and feelings of not existing
2. DMT DECREASES the strength of BACKWARDS CTW
43
Closing eyes: how does CTW differ when on DMT or not?
1. No drugs with eyes closed leads to backwards waves increase, and forwards waves decrease (sensory processing decreases) -
2. DMT with eyes closed leads to forward waves increase in sensory processing, eliciting visual processing with eyes closed, visual hallucinations
44
If prominent alpha waves that occur during eyes closed, what happens during DMT
frontal to posterior alpha waves are almost eliminated from the signal under DMT eyes closed
45
How might long term meditation influence predictive errors?
Long term experience with meditation involves continual attention to one’s sensations, and thus likely trains their interoceptive awareness / predictive errors of sensory information
46
Considering that meditation encourages non judgemental acceptance, reduced stress and rumination, and improved attention
what were the 2 hypotheses of meditator's cortical travelling waves?
1. Hypothesis 1 = meditators will show increased bottom up processing (forwards CTW), more processing sensory information and more attention to and synaptic gain for processing prediction errors
2. Hypothesis 2 = weaker top-down predictions in meditators, due to their trained awareness that all sensations/prediction errors are temporary, and might not need to take active inferences to change them, thus weaker backward travelling waves
47
What was the mediators/EEG task and how does it relate to cortical travelling waves?
1. Focus on centre cross, press button to a happy face, not press button to a sad face, hard to inhibit due to habit of pressing
2. You are making cognitive control mechanisms about the instructions, upcoming stimuli and predictions from top down CTW
48
What were the 2 main findings on mediators forward and backward cortical travelling waves?
1. Meditators stronger forwards travelling alpha waves (suggesting increased sensory processing)
RESTING
2. Meditators show weaker backwards alpha waves (suggesting reductions in abstract predictions in memories, future predictions) - when those top-down references aren’t useful in resting / eyes closed
49
What is the major difference in CTW between mediators and non-mediators?
In go/no-go task
1. Mediators had stronger forwards CTW (increased attention/precision of prediction eros and sensory processing)
2. the same strength of backwards waves as the non mediators when they were resting
3. non meditators had the SAME strength of backwards waves/top down predictions while resting AND doing the task! - suggests we use additional predictive processing energy when we DO NOT NEED TO
50
When do mediators engage in top down processing?
Thus mediators engage in top-down processing when they are doing a task, but not when they are just resting -
51
What does the lower backwards CTW / top down processing suggest in mediators?
Overall suggests that reductions in backwards CTW with predictive processes could be a marker of reducing ruminations and worries
52
Limitations
1. CTW effects did not have the strongest effects detected
2. PROs = Used time series feature extract approach, analysing EEG features comprehensively in 7000 different ways
3. PROs = Had many multiple comparison controls, to reduce risk of high false positive rate
53
What cognition differentiated the mediators from non mediators the best?
The brain topography associated with somatosensory attention differentiated the 2 groups the best
Found a single component that could differentiate the 2 groups meditators vs. non-meditators with 67% accuracy (above chance level)
54
What does the stationarity of dynamic properties (suggesting neural stability) ask?
How stationary is the predictability of the time series?”
55
What did stationarity of dynamic properties find about mediators?
Distribution of data was more consistent and stable across time in meditators vs. non mediators in the region that processes SOMATOSENSORY activity
Increased stability in meditators brain activity correlates with high minutes per week of meditating
Practising 35-45 minutes per day seems to be threshold, aligns with recommendations on how much meditation to do
56
How long to mediator to see neural effects in practice?
Main literature says you must do 100-150 hours before you see any health benefits, more stable after 4 years of practice
57
What types of neural oscillations did the mediators have more of?
1. theta = cognitive control,
2. alpha = inhibitory signal
3. gamma = intensive cognitive/perceptual processing
58
How could we use the EEG/capturing PP to analyse neural differences in people with schizophrenia?
1. Engages auditory processes, compare brain activity in predictive processing mechanisms in people with and without schizophrenia
In schizophrenia patients
2. Find that the mismatch negativity between the tones is reduced in amplitude in schizophrenics
3. They do not develop the same strength of prediction or update their prediction based on new sensory information as strongly
4. weaker PP mechanisms might explain behaviours of schizophrenia
58
How do you engage predictive processing mechanisms in their brain in an EEG study?
Method
1. Repeat a simple boring tone, while EEG records, then play a deviant tone which is higher/lower
2. By using simple predictable sounds and then an unexpected sound in the sequence, the EEG can record mismatch negatively
3. Have a stronger reaction to the deviant tone, which reflects the predictive model updating
59
What is the mismatch negativity spike involved in?
1. produced by the auditory cortex about 200ms after the mismatch tone
2. sensitive to very early processes in brain function, extracts relevant info from an irrelevant background
3. Indexes glutamatergic neurotransmitter activity - excitatory processes, known to be affected in schizophrenia
60
What clinical symptoms is the mismatch negativity amplitude involved in?
(Remember, the MMN amplitude correlates with symptom severity and psychosocial outcomes)
1. MMN at TEMPORAL areas correlate with verbal memory symptoms and POSITIVE symptoms, eg., delusions and hallucinations
2. MMN at FRONTAL regions correlates with NEGATIVE symptoms, e.g., cognitive impairment in schizophrenia
61
What are the main 2 limitations of making assumptions about the MMN amplitude and schizophrenia?
1. MMN is not specific to schizophrenia / also found in bipolar disorder
2. MMN is also altered by alcohol and cannabis use
62
What is the issue with the common lack of specificity of EEG markers?
1. MMN might just be a mechanism involved in impaired function
2. EEG markers vary between patients and are often not specific to one mental illness
3. Poses difficulties in drawing mechanistic inferences about neural activity and treatment
63
What is Error Related Negativity?
When people make an error, people become aware and develop a neural response called error related negativity
ERN seems to be an emotional neural activity response of a conflict between internal goals and outward behaviour
64
What brain regions are implicated in ERN?
ERN is a negative voltage deflection in frontal areas, linked with theta brain waves and made by dorsal ACC
65
What clinical disorders have heightened/altered Error Related Negativity?
1. There is an increased ERN amplitude in OCD, not always found
2. also sometimes found in non-affected siblings of individuals with OCD
2. It is also altered in depression, anxiety, trait worry and ADHD
66
What is the relationship between theta and beta waves and ADHD?
1. Individuals with ADHD show increased theta waves, mean of 132% higher than control
2. But LOWER beta power, and thus an increased theta : beta ratio
3. SOLUTION = increased ratio can be normalised with Ritalin treatment, but some individuals have increased beta rather than decreased
67
How would we test more complex lower- or higher order predictive expectations in EEG studies?
1. Repeating stimulus paradigms MMN can be adjusted:
1. Lower order expectations can be broken with SIMPLE CHANGES in a beep pattern
2. Higher order expectations can be broken by BREAKING SEQUENCES OF PATTERNS instead of a singular stimulus
Higher order, eg., beep-beep-bop, beep-beep-bop, beep-beep-bom
68
How do we test whether the low and high order expectations are involved in a cortical hierarchy?
1. Comparing lower with higher order expectations supports to notion of the cortex is organised in a temporal hierarchy
* whereby frontal regions process longer predictions
* while posterior regions process immediate simpler stimulus predictions
69
How can the McGurk effect be used to test predictive processing?
1. Since the certainty of visual information from their month predicting auditory info can vary
2. More certain visual information will generate stronger predictions for the incoming sound
1. By using differences in prediction strength to test neural responses, generates stronger or weaker prediction errors, when predictions don't match the later sound more or less
2. Also can compare how response to stronger/weaker prediction errors differ in clinical and healthy populations
70
What McGurk effect method did Method from Romero et al., 2016 use to study predictive processing?
1. Make a face expression like they were going to produce sound
2. Then played them the sound whereby the sound either matched or did not match the face
3. Make predictions that are stronger/weaker depending on whether it was ‘p’ or ‘g’ sound
71
What is different in the theta waves of schizophrenia patients?
1. Theta oscillations in frontal regions are reduced, which is involved in conflict resolution between predictions of different sounds
2. This suggests they CANNOT UPDATE HIGHER ORDER PREDICTIVE MODELS AFTER AN AMBIGUOUS MULTISENSORY PREDICTION ERROR LIKE THE MCGURK EFFECT
72
Do schizophrenia patients have intact or impaired prediction error processing in auditory region?
Schizophrenia patients have intact prediction error processing in auditory regions and CAN make predictions based on visual information
73
What are the 2 main implications for lower theta waves in schizophrenia?
1. reduced efficacy at integrating multisensory information, making cognitive functioning harder
2. may be involved in delusion, hallucination symptoms, with a less effective predictive model might increase vulnerability to having experiences that are mismatched to actual sensory inputs
74
What are the 2 ways that depression seems to have impaired network connectivity?
1. Default mode network (self, rumination, worry) seems to be overconnected
2. Cognitive control regions seem to be under connected
75
What regions is TMS applied to in depression?
(remember: TMS applies a strong magnetic field to the scalp with no resistance and generates neural activity, causes DEPOLARISATION)
1. TMS is applied to
* dorsolateral PFC, involved in attention and cortical control can regulate connectivity between DLPFC and other brain regions
76
How quickly can rTMS can alter cognition and connectivity?
1. rTMS can alter cognition and connectivity for 45 minutes after a single session, over weeks can produce lasting neuroplastic change, increases connectivity
77
How can Transcranial alternating current stimulation tACS give personalised treatment?
1. tACS applies weak electrical field to brain, strong enough to affect membrane potentials, applied at a selected frequency, to enhance specific neural oscillations
2. If you find specific under/overreactive oscillations in a specific condition and they are casual mechanisms, you can increase or reduce oscillations and give personalised treatment
78
How can we predict responses to TMS treatment for depression?
1. Applying a time series comparison approach to pre-treatment EEG with machine learning algorithm produced 68% accuracy at predicting response to TMS treatment
79
What were the 3 main findings in mediator's CTW?
1. Meditators showed increased forward wave strength during tasks, supporting enhanced sensory processing.
2. Had lower backward wave strength at rest, possibly due to meditation’s focus on present-moment awareness reducing internal predictions or rumination.
3. However, during tasks, meditators could still generate strong backward waves when needed, suggesting flexibility in predictive processing.
80
What were the 3 main findings regarding working memory test and CTW?
1. Forward waves increased during stimulus encoding and predicted better performance.
2. Backward waves rose during the recall phase, possibly helping to reconstruct memory representations.
3. Lower forward CTW during recall predicted better performance, perhaps because strong forward waves during recall could reflect being distracted by new sensory input
81
What is the pattern of backward / forward waves in schizophrenia and PTSD?
1. Schizophrenia (linked to excessive backward waves)
2. PTSD (linked to reduced forward waves)
82
What were the 2 main methodological constraints in the EEG mediator study?
1. Only midline travelling waves were studied.
2. Possible confound: participants may have meditated during the “rest” condition.
* EEG might not reflect actual cortical travelling waves; more direct measures are needed.
83
What did researchers explain why there was no correlation between Go/No-go performance and wave strength?
1. possible ceiling effects
84
What specifically did Roa Romero test?
1. high density EEG analysis to investigate crossmodal PE processing generated by audiovisual speech (McGurk effect) in schizophrenia and healthy patients
85
What were the 2 main findings of the Roa Romero study?
1. Higher ERPS found in HIGH over LOW predictive symbols in both groups, suggesting intact audiovisual incongruence detection in auditory cortex in both healthy and SZP
2. However, healthy controls (but not SZP) had enhanced theta oscillations (4–7 Hz) in HIGH compared with low-predictive syllables in the frontal cortex
abnormal frontal theta band oscillations reflect cross modal PE processing deficits in SZ (top-down multisensory processing deficit)
86
What is believed to be the cause of weird theta in SZP + implications?
dysfunctional neural oscillations (unsynchronised oscillatory activity) might underlie altered crossmodal PE processing in SZ.,
leading to altered processing and SZ symptomatology
87
How did Roa Romero use the McGurk effect to test congruent/incongruent trials and low/high predictions?
1. By altering temporal delay of visual and auditory information, the brain forms phonological predictions based on visual info before the sound:
1. incongruent (Ba / Ta) and congruent trials (Ba / Ba)
2. words that elicit high (Ba) and low predictions (Ga) of the incoming word
88
How did HC outperform SZP in reaction times on congruent/incongruent trials of sounds?
1. Reaction times to RTs to audiovisual target syllables were slower in SZP than in HC
2. Only HC showed quicker RTs to congruent trials, while Incongruent and congruent RTs did not differ in SZP
3. Indicating that incongruent trials had a stronger impact on performance in HC than in SZP
89
How does lower theta impact SZP performance and ability to engage in PP errors?
1. Theta oscillations are involved cognitive control induced by conflicting and erroneous information from the environment
2. Deficits in theta point to insufficient updating of complex multisensory predictions and altered top-down processing in SZ
2. Thus the lack of frontal theta-band power enhancement in the present study might reflect, at least in part, temporal processing deficits at later processing stages in SZP
90
What is “expected free energy” in active inference, and why is it important?
A: Expected free energy is a measure that combines expected prediction error, epistemic value (information gain), and preferences. Minimizing it guides decision-making under uncertainty by favoring actions that are both informative and rewarding
91
Q: Why do Smith et al. emphasize the mesoscale level of brain organization?
A: Mesoscale circuits (neural populations with structured connectivity) bridge micro-level neural dynamics and macro-level cognitive function, making them crucial for understanding how disorders emerge from neural computations
92
Q: How does predictive coding explain core symptoms of depression?
A: Depression may involve maladaptive precision estimates, such as overly confident beliefs about negative outcomes, leading to biased inference and persistent negative affect.
93
What role does interoception play in depression within the predictive coding framework?
A: Faulty prediction of internal bodily states (interoceptive inference) can lead to disrupted mood regulation and affective experience, contributing to symptoms like anhedonia and fatigue.
94
What refers to the disorder when high precision is attributed to negative interoceptive predictions, leading to chronic expectation of harm and heightened autonomic responses, even in safe environments.
anxiety conceptualized in active inference models
95
How do predictive models account for somatic misinterpretations (e.g., panic attacks)?
A: Misattribution of bodily signals—like interpreting benign arousal as a medical emergency—occurs when the brain has maladaptive priors and overweighted prediction errors about bodily threat.
96
What are the hypothesized computational mechanisms underlying autism?
A: Autism may involve over-precise sensory prediction errors and under-weighting of abstract priors, leading to difficulties in generalizing, forming social inferences, and tolerating unpredictability.
repetitive behaviors in autism may serve as self-generated, predictable stimuli that reduce uncertainty and minimize overwhelming prediction errors in a sensory world perceived as overly volatile.
97
How do active inference models explain hallucinations in schizophrenia?
A: Hallucinations may stem from over-precise priors (e.g., expecting speech) combined with underweighted sensory evidence, leading to false percepts being treated as real.
98
What mechanism is proposed for delusion formation in schizophrenia?
A: Delusions may arise from maladaptive updating due to noisy or aberrant precision signals, resulting in the persistence of implausible beliefs despite contradictory evidence.
99
How does policy precision relate to negative symptoms like avolition?
A: Reduced policy precision (low confidence in action selection) can result in indecision, inaction, or motivational deficits often observed in depression and schizophrenia.
100
What is the role of interoceptive simulations in modeling fatigue and depression?
A: They show how allostatic failure (inability to meet internal physiological needs) can lead to persistent fatigue by generating chronically unresolved prediction errors.
101
How is “risk” defined in active inference models?
A: Risk refers to the anticipated cost or negative outcomes associated with certain policies or actions. It is one component of expected free energy and reflects how much predicted future states deviate from preferred outcomes.
102
What is the role of epistemic value (uncertainty reduction) in balancing risk and surprise in policy selection?
A: Epistemic value promotes exploration by favoring actions that reduce uncertainty. In active inference, the brain balances epistemic gain (information-seeking) against pragmatic risk (goal achievement), enabling adaptive behavior even in uncertain environments.
