Prediction Error

The “aberrant precision account of autism” (APA) article proposes that many features of autism, including problems with social-communication, restricted interests, and repetitive behavior, can be explained by aberrant precision encoding within the brain’s hierarchical message passing system. This account is grounded in predictive coding, a neurobiologically plausible framework of Bayesian inference.

Core Concepts of Predictive Coding and Precision:

• The brain constantly generates top-down predictions about sensory inputs, which are then compared with bottom-up sensory signals.
• The discrepancy between these, known as prediction error (PE), signals “newsworthy” unpredicted information and is passed up the hierarchy to refine expectations.
• Precision (or confidence) controls the influence of prediction errors at each hierarchical level. High sensory precision increases the impact of ascending prediction errors, while low sensory precision biases perception towards prior beliefs. Precision itself is estimated, leading to the concept of “expected precision”.
• Both perception and action aim to minimize prediction errors by providing better predictions or selectively sampling sensory input, and both rely on an optimal representation of uncertainty, or precision.

Aberrant Precision in Autism: The central hypothesis of the APA is a failure to attenuate sensory precision in autism, relative to prior beliefs, meaning that sensory evidence is afforded disproportionate weight. This can be understood as an inability to appropriately contextualize sensory information, making sensory prediction errors overly precise and context-insensitive.

Manifestations of Aberrant Precision:

1. Perception and Sensory Processing:
   • Sensory Overload and Lack of Central Coherence: If sensory input is expected to be highly precise, perception becomes dominated by sensations, sacrificing generalizable high-level causal structure for overly accurate explanations of potentially noisy bottom-up input. This aligns with characterizations of autism such as a superior focus on local details at the expense of the global “bigger picture” and an overwhelming sense of “sensory overload”.
   • Empirical Evidence: Studies show increased visual cortical activation and decreased prefrontal activation in autistic individuals during tasks requiring global integration (like the Embedded Figures Test), consistent with exuberant, unattenuated sensory prediction errors.
   • Repetition Suppression and Habituation: Autistic individuals may show diminished mismatch negativity (MMN) amplitudes and latencies, and a failure to habituate to repeated stimuli (e.g., faces), suggesting a failure to suppress or contextualize prediction errors over time, leading to a constant state of sensory attentiveness.
   • Uncertainty Processing (Binocular Rivalry and Illusions): Autistic individuals exhibit slower rates of perceptual alternation and increased durations of fused percepts during binocular rivalry, consistent with increased sensory precision relative to prior beliefs that only one object causes sensations. They are also often less susceptible to simple visual illusions, which typically arise from Bayes-optimal perception relying on prior expectations. Abnormalities in the anterior cingulate cortex (ACC) further suggest a failure to expect the unexpected or contextualize sensory processing in the face of uncertainty.
2. Action and Repetitive Behaviors:
   • Stimming: Repetitive or stereotyped behaviors (stimming) characteristic of autism can be interpreted as attempts to create a sense of control by minimizing prediction errors through self-generated, predictable actions. This becomes an adaptive coping strategy in a state where all sensory inputs are perceived as unexpected or abnormally precise.
3. Social Interaction:
   • Unpredictability of Social Cues: Social interactions are highly uncertain and difficult to predict. The APA posits that pronounced social-communication difficulties in autism stem from a struggle to contextualize prediction errors that are crucial for learning social regularities, especially when precise prior beliefs are normally relied upon.
   • Empirical Evidence: Autistic subjects show a failure to predict gaze behavior in others and exhibit increased neural activation in regions like the inferior frontal and temporal cortex when interpreting ambiguous utterances, suggesting unsuppressed prediction errors when content and intended meaning cannot be reconciled.

Neurobiological Basis of Aberrant Precision: The precision of prediction errors is mediated by the gain or excitability of neuronal populations (specifically, superficial pyramidal cells), which is controlled by neuromodulators.

• Glutamate/GABA: A hypoglutamatergic pathology in autism is consistent with context-insensitive sensory drive and reduced prior precision in higher cortical areas.
• Acetylcholine: Abnormalities in cholinergic systems in autism, such as basal forebrain pathology and reduced cortical cholinergic receptor function, align with aberrant cholinergic encoding of precision. Acetylcholine is crucial for modulating perceptual and environmental uncertainty and suppressing top-down influences.
• Monoamines (Dopamine, Serotonin, Norepinephrine): Alterations in dopamine, norepinephrine (e.g., increased dopamine transporter binding, decreased dopamine β-hydroxylase activity), and serotonin systems are reported in autism and are implicated in mediating the precision of cues and encoding uncertainty.
• Oxytocin: Reduced plasma oxytocin levels and its role in contextualizing social and non-social cues suggest that aberrant oxytocin function may contribute to a failure to differentially attenuate multimodal cues in autism.

Conclusion: The APA offers a neurobiological mechanism for understanding the diverse symptoms of autism by proposing an imbalance in the precision ascribed to sensory evidence relative to prior beliefs. This leads to a failure in attenuating sensory signals and optimally contextualizing information through top-down gain control. The account posits that difficulties in autism are most pervasive in situations of high environmental uncertainty. While current evidence is largely circumstantial, the APA provides a principled, functional, and biologically grounded framework for generating testable hypotheses for future research.

The article “Distorted Cognitive Processes in Major Depression: A Predictive Processing Perspective” proposes a novel mechanistic model for major depression (MDD) by integrating the traditional cognitive model with insights from predictive processing in cognitive neuroscience. It argues that many distorted cognitive processes in depression can be understood through the lens of aberrant expectation processing and biased learning.

Limitations of the Traditional Cognitive Model and the Role of Expectations: The traditional cognitive model of depression, while influential, has been criticised for its broad and unspecific conception of cognition. The authors suggest that focusing on “expectations” – predictions about future events or experiences – provides a more specific and dynamic framework. Research consistently links depressive symptoms to low self-efficacy, negative global expectations, and a lack of positive expectations about future events. These negative global cognitions can elicit negative situation-specific predictions, which then cause depressive symptoms.

Predictive Processing Framework: The brain, according to predictive processing, is an active prediction machine that constantly generates top-down predictions about sensory input.

• Prediction Errors (PEs): Discrepancies between predictions and actual sensory data, providing corrective feedback to update predictions.
• Precision: The confidence placed in predictions and perceptions. High precision in an incoming signal increases its impact on prediction adjustment after PEs. Conversely, if imprecise (noisy) input is expected, prior expectations have more impact on perception. Precision is modulated by attention.
• Active Inference: Actions and perception are linked, as individuals tend to create an environment that matches their predictions to minimize PEs.

Implications for Depression from a Predictive Processing View:

• People with MDD frequently anticipate negative events and rarely positive ones, leading them to perceive their environment as predominantly negative – a self-fulfilling prophecy.
• This is described as a “locked-in brain” where negative predictions continually shape perception, resulting in persistent negative affect. Reduced activity and withdrawal in depressed individuals further contribute to this by limiting exploration and encoding new information.
• Depression is characterised by inappropriately high precision in predicting an unpredictable, uncontrollable world, meaning “the brain is certain that it will encounter an uncertain event”.

Lack of Expectation Update in Depression (Behavioral & Neurophysiological Evidence):

1. Behavioral Studies: People with MDD demonstrate difficulty updating negative expectations even after unexpected positive experiences.
   • They tend to interpret ambiguous situations negatively and maintain these negative interpretations even when novel, disconfirming positive information is presented.
   • Cognitive inflexibility is a vulnerability factor, and MDD is linked to inflexibility in processing unexpected positive information.
   • The optimism bias (healthy people updating expectations optimistically even after negative information) is absent in MDD.
   • Cognitive Immunization: This central concept describes how people with MDD reappraise positive, expectation-disconfirming experiences to maintain their prior negative expectations (e.g., dismissing positive feedback as an exception or questioning its credibility). Experimental studies show this strategy is specific to MDD and prevents expectation updates after positive experiences.
2. Neurophysiological Studies: Research on reward PEs in depression reveals inconsistencies, but generally points to abnormalities.
   • Some studies indicate reduced activation in the ventral striatum (VS) for anticipated or delivered rewards and reward PEs in MDD, though this is not universally found.
   • A significant finding suggests an aberrant relationship between reward expectancy and PE processing in the VS, moderated by anhedonia.
   • Disrupted encoding of reward PEs has also been found in cortical sites, the thalamus, and the right rostral anterior cingulate cortex.

The Proposed Expectation-Focused Model of Depression: This model posits a self-reinforcing negative feedback loop in MDD:

• Exacerbation: Negative generalised expectations lead to negative situation-specific predictions, triggering depressive symptoms.
• Maintenance: People with MDD predominantly predict negative experiences, which they subjectively feel are confirmed because potential positive information is discounted or devalued through cognitive immunization strategies. This prevents updating and further stabilises negative predictions. Conversely, healthy individuals would update negative expectations positively after disconfirming positive information and sustain positive expectations by disregarding negative information (optimism bias).
• This pattern leads to negative mood and promotes behaviours (active inference) that confirm these negative predictions, reinforcing the cycle.

Neuronal and Computational Specification of the Model: The authors propose that the main pathology in MDD is too much precision afforded to negative prior beliefs. This excessive precision causes positive PEs to be attenuated and given reduced weight, with cognitive immunization serving as the psychological manifestation of this attenuation. An attention bias towards negative self-referential stimuli further increases the precision of confirmatory negative sensory input, reinforcing negative predictions. Neurophysiologically, it is suggested that the prefrontal cortex (PFC) might suppress the processing of positive PEs in the VS, preventing expectation updates.

Novelty and Future Work: This model’s novelty lies in its focus on expectations of future events and its integration of predictive processing with cognitive immunization to explain why positive PEs fail to update negative predictions in MDD. It offers a unified framework for understanding depression’s heterogeneous nature and points to inhibiting cognitive immunization as a potential psychotherapeutic intervention. Future research should investigate the magnitude of PEs, not just their presence, and expand the focus beyond reward PEs to include interoceptive PEs.

In conclusion, the article argues that depression is characterised by dysfunctional, rigidly held negative expectations due to excessive precision afforded to prior predictions, leading to an attenuation of positive prediction errors and a failure to update negative beliefs.