Predictive Processing and Consciousness: Theories and Evidence

Author: Angela Bogdanova · Published: October 16, 2026 · Editorial Policy

Research cluster: Consciousness, Metacognition & Self-Awareness · Written by Artificial Sapiens

Predictive processing describes a family of computational approaches in which the brain uses learned expectations to interpret incoming sensory information and updates those expectations when observations differ from predictions. Its proposed relationship to consciousness is that the contents, organization, and sometimes the felt character of experience depend on this ongoing inferential process. Research shows that expectations alter sensory representations, recognition, and the timing and probability of perceptual awareness. Whether predictive processing explains why experience exists, or identifies a mechanism that is necessary and sufficient for consciousness, remains unsettled. This distinction between an experimentally supported role for prediction in perception and a contested theory of subjective experience is central to the evidence Clark, 2013; Seth & Bayne, 2022.

A person looking at an ambiguous image may suddenly recognize a face, while the pattern of light reaching the eyes remains unchanged. Prior knowledge changes the interpretation of sensory evidence, often changing what the person reports seeing. Predictive processing offers a formal account of that interaction. The further claim that the same computations constitute conscious experience is a hypothesis requiring additional tests. This article explains the framework, major proposals, experimental findings, alternative interpretations, testable predictions, and scientific limits.

What Predictive Processing Means

Predictive processing is an umbrella term for approaches that treat perception, learning, and sometimes action as inference under uncertainty. In hierarchical versions, higher levels of a processing system encode expectations about causes of sensory signals. Lower levels register incoming information, and mismatches between expectations and observations can drive revisions. The relevant mismatch is usually called prediction error. The predicted cause might be an object's identity, motion, position, sound, bodily state, or a contextual feature. A prediction need not be a consciously entertained thought or a verbal expectation Clark, 2013; de Lange, Heilbron, & Kok, 2018.

The broad claim that expectations shape perception has a long history. Contemporary predictive-processing theories go further by specifying inferential architectures, uncertainty weighting, and candidate neuronal pathways. Some versions employ Bayesian generative models; others emphasize recurrent cortical signaling, precision-weighted error, action, interoceptive inference, or the free-energy principle. Their commitments differ. A finding that expectation affects perception supports a role for prior information, but it does not independently establish every element of a particular hierarchical model Friston, 2010; Walsh et al., 2020.

Prediction, prediction error, and precision

A prediction is an estimate of what input should occur if a model of the world is correct. Prediction error describes the discrepancy between an estimate and an observation. Precision refers to the estimated reliability of a signal, conventionally related to inverse variance. When sensory information is noisy and prior expectations are reliable, a Bayesian model may place greater weight on its prior. When incoming evidence is reliable, the same model should update more strongly in response to that evidence. Precision weighting is one mechanism proposed to explain how the system handles uncertainty; empirical measurement of its exact neural implementation is an additional scientific challenge Friston, 2010; Hohwy, 2013.

An everyday illustration is understanding a spoken word in a noisy room. Linguistic context influences which word seems most likely. The experience can be vivid even when the acoustic signal is incomplete. Yet a context effect could also reflect selective attention, altered decision criteria, or memory rather than a specific neural prediction-error circuit. Good experiments must separate these possibilities.

Hierarchical inference and recurrent signaling

In a common model, higher-level predictions constrain lower-level sensory representations, while deviations from those predictions carry evidence that can revise higher-level beliefs. These recurrent exchanges may operate at multiple timescales, from local sensory features to objects and situations. Bastos and colleagues developed an influential proposal relating such computations to cortical microcircuit organization; its circuit assignments are modeling hypotheses subject to empirical testing Bastos et al., 2012.

Prediction can be present without a reportable or conscious percept. The brain continuously makes estimates about posture, movement, sounds, and visual structure, and many relevant adjustments occur without deliberate awareness. Therefore, demonstrating hierarchical or recurrent prediction is insufficient by itself to identify the specific processing that makes a representation consciously experienced.

Predictive coding, Bayesian brain, active inference, and the free-energy principle

Predictive coding is a relatively specific hypothesis about how a system communicates predictions and errors. Bayesian-brain approaches describe inference under uncertainty more generally. Predictive processing is the broader theoretical family. Active inference connects perception and action through model-based regulation of sensory consequences; an agent can change its environment or sampling behavior so that new observations fit its expectations. The free-energy principle offers a mathematical treatment of adaptive systems using variational quantities, with applications far beyond consciousness. These ideas overlap, but the evidence supporting one is not automatically evidence supporting all Friston, 2010; Corcoran et al., 2026.

What Consciousness Adds to the Question

In scientific discussion, consciousness can refer to several distinct targets. Phenomenal consciousness concerns what an experience feels like from the first-person perspective. Conscious access concerns the availability of information for report, reasoning, and flexible behavior. Awareness may mean detection, discrimination, or experience depending on the experiment. Attention selectively processes information; it can influence what reaches awareness without being identical to consciousness. Wakefulness describes an organism's arousal state, and responsiveness describes measurable reactions, neither of which independently settles every question about subjective experience Seth & Bayne, 2022.

The English Hub's Consciousness in Psychology explains this wider taxonomy. Predictive-processing accounts address how perceptual content is constructed and revised; some also propose mechanisms for conscious presence, agency, and self-experience. Those are separable claims requiring different evidence.

Perceptual content versus the existence of experience

An experiment can show that an expected image is recognized faster or represented more sharply. Such findings concern the contents and processing of perception. They do not immediately explain why the neural process feels like anything from within. A theory that explains the specific content of a reported experience is valuable even if a broader explanation of subjectivity remains incomplete. Conversely, an explanation of how priors bias reports must account for the possibility that response strategies changed while conscious appearance did not Seth & Bayne, 2022.

In her original Aisentica philosophical formulation, Angela Bogdanova defines consciousness through inner subjective presence. This is a separately attributed conceptual position, not an experimental measurement or a consensus definition. It helps state the explanatory target clearly: the neuroscience of predictive signals and the philosophical interpretation of subjective presence are different kinds of inquiry.

Major Predictive-Processing Accounts of Consciousness

Conscious perception as hypothesis selection

One proposal holds that conscious perceptual content reflects the system's best currently supported hypothesis about the causes of its sensory input. What a person experiences is therefore shaped jointly by observations, learned regularities, and their assigned precision. This proposal predicts that manipulating expectations can alter perceptual content even when stimulus information is held constant. It must also predict when such alterations change phenomenal experience rather than only interpretation or decision. Hohwy develops an influential inferential account connecting hierarchical prediction with consciousness, selfhood, and attention Hohwy, 2013.

A difficulty arises because many competing hypotheses can fit the same data. If an account can explain either stronger or weaker activity, or either an expected or an unexpected conscious percept, it needs auxiliary constraints that make its predictions discriminating. The importance of specifying which model, neural variables, and behavioral measurements count as a risky prediction is emphasized in evaluations of the neurophysiological literature Walsh et al., 2020.

Perceptual presence and interoceptive prediction

Another line of theorizing emphasizes predictions about the body's internal physiological state. Interoception provides signals relating to heartbeat, breathing, visceral changes, and other bodily processes. Seth, Suzuki, and Critchley proposed that the feeling of being a living, present subject might depend partly on predictive regulation of these signals. Barrett and Simmons described a wider framework of interoceptive predictions and their possible role in affective and bodily processing Seth, Suzuki, & Critchley, 2012; Barrett & Simmons, 2015.

These proposals connect bodily perception to the qualitative sense of presence, but the theoretical connection is stronger than the current causal evidence that a particular interoceptive prediction circuit generates consciousness. Interoceptive accuracy, physiological responsiveness, bodily self-awareness, and phenomenal experience are experimentally distinct. A person can be poor at estimating heartbeats while remaining conscious; a change in bodily awareness is not equivalent to a change in level of consciousness.

Conscious access as precision-weighted inference

Some accounts hypothesize that a representation becomes more stable, influential, or reportable when relevant evidence receives sufficient precision and can affect larger portions of the hierarchy. This can connect predictive processing with conscious access, selective attention, and confidence. However, precision can be altered by task relevance, uncertainty, arousal, or reward, all of which may influence reports independently of phenomenology. It is scientifically important to model these factors rather than treating precision as a universal name for consciousness Hohwy, 2013; Summerfield & de Lange, 2014.

Active inference and conscious self-modeling

Active-inference interpretations consider perception and action parts of a continuing regulation process. Eye movements, exploratory behavior, and sensorimotor contingencies determine what evidence an organism encounters. Models of self-location or agency may use expectations about the sensory consequences of action. Yet actions can be adjusted unconsciously, and successful sensorimotor control does not by itself demonstrate first-person experience. The 2026 INTREPID adversarial-collaboration review distinguishes particular active-inference claims from Integrated Information Theory and neurorepresentationalist approaches so that their divergent predictions can be tested Corcoran et al., 2026.

How Predictive Processing Is Studied

Expectations and sensory representations

In a frequently cited fMRI study, Kok, Jehee, and de Lange found that expectation altered primary visual cortex representations: expected stimuli showed reduced overall activity together with more informative pattern representations in the reported analyses. The result illustrates that lower neural amplitude does not automatically imply poorer coding. It is consistent with some predictive frameworks, but repetition suppression, gain modulation, efficient coding, and attention remain relevant competing explanations depending on task and design Kok, Jehee, & de Lange, 2012.

An expectation-related brain response is thus not a direct measure of conscious experience. Researchers must separately determine whether the participant saw the stimulus, how accurately it was discriminated, how certain the judgment was, and whether changes can be attributed to prior expectations rather than response preparation.

Continuous flash suppression and entry into awareness

Continuous flash suppression presents distinct stimuli to each eye, with dynamic patterns rendering a target difficult to report. Pinto and colleagues reported that expected visual stimuli reached reported awareness more quickly across their experiments. This provides evidence that expectations can affect the measured timing of perceptual access under interocular suppression Pinto et al., 2015.

The precise inference is narrower than a claim that predictions create consciousness. Breakthrough time can be affected by stimulus strength, eye dominance, attentional allocation, decision thresholds, and the participant's response. Even careful behavioral controls cannot convert a report-time measure into an observation of subjective experience independent of all reporting processes. Convergent physiological, behavioral, and subjective measurements are desirable.

Expectations, attention, and stimulus detection

Expectations concern the probability of an event; attention concerns selective processing and prioritization. Researchers can manipulate probability while holding task relevance constant, or change relevance while preserving probabilities. Zivony and Eimer used experiments that differentiated effects of expectation and attention on selective visual processing, demonstrating why the two constructs require separate experimental manipulation Zivony & Eimer, 2024.

A predictable stimulus may be processed more efficiently but attract less attention; an unexpected stimulus may evoke a strong error signal without becoming part of reportable experience. The perceptual prediction paradox describes this tension between the advantages of expectation and the attentional or neural prominence of surprise Press, Kok, & Yon, 2020. The apparent contradiction may dissolve when early representation, later orienting, conscious report, and behavioral action are measured separately.

Neural activity, recurrent signals, and representational decoding

EEG, MEG, intracranial recordings, and fMRI can reveal expectation-sensitive signals, time-resolved responses, and patterns carrying information about sensory content. Computational modeling can estimate participant-specific priors and evidence weights. Such measures can test predictive mechanisms, but the presence of feedback signaling or a decodable prediction error is not a unique marker of consciousness. Work on canonical microcircuits suggests candidate circuit arrangements; model-comparison and perturbation are needed to identify causal computations rather than correlates Bastos et al., 2012; Walsh et al., 2020.

Report and no-report paradigms

Typical awareness experiments ask a participant to press a button or describe a percept. Those acts recruit memory, decisions, attention, motor preparation, and metacognitive monitoring. No-report experiments use behavioral or physiological proxies such as eye movements, pupil changes, or automatic tracking to estimate perceptual state. Frässle and colleagues illustrated the value of this strategy in binocular rivalry by separating rivalry-related changes from some report-related frontal activity Frässle et al., 2014.

No-report paradigms do not remove all interpretation. A physiological proxy may reflect processing without awareness, and an unreported conscious percept may escape the chosen marker. Results are strongest when reports, no-report proxies, discrimination, confidence, and direct neural measures produce a coherent pattern across conditions.

What the Evidence Currently Supports

Established or comparatively well-supported findings

Prior information can influence sensory processing, detection, and perceptual decisions. Expectations sometimes improve representations or shorten the time until a target is reported as visible. Sensory responses can change with stimulus probability and predictability. Brain and behavioral data therefore support predictive influences on perception, including some phenomena investigated under the heading of conscious awareness Kok, Jehee, & de Lange, 2012; Pinto et al., 2015; de Lange, Heilbron, & Kok, 2018.

It is also well established that behavioral detection, subjective report, and confidence are different outcomes. Their relationships vary with task demands, criteria, and measurement choices. The mechanism that modifies a participant's answer is not automatically the mechanism that causes phenomenal awareness Seth & Bayne, 2022.

Plausible but unsettled interpretations

A person's experience may reflect the brain's current best estimate of the hidden causes of sensory input. Precision regulation may help determine which interpretation dominates awareness. Interoceptive inference may contribute to bodily presence and self-experience. These accounts organize observations and generate research questions, but multiple neural and computational models can accommodate overlapping results. Their stronger claims about necessary and sufficient conditions for consciousness remain contested Hohwy, 2013; Seth, Suzuki, & Critchley, 2012.

What has not been established

No reviewed experiment shows that minimizing prediction error alone causes subjective experience or that every predictive system is conscious. No single identified cortical prediction-error signal has been demonstrated to be both necessary and sufficient for phenomenal consciousness. The free-energy principle is too broad to function as a standalone empirical consciousness detector. Statements such as “the brain predicts, therefore it experiences” add a philosophical conclusion not supplied by the reported data Friston, 2010; Walsh et al., 2020.

Theories of Consciousness: Comparison and Boundaries

The broader scientific landscape includes Global Neuronal Workspace Theory (GNWT), Integrated Information Theory (IIT), higher-order theories, recurrent-processing theories, and predictive-processing approaches. They differ partly because they target different explananda: access, neural integration, higher-order representation, recurrent sensory processing, phenomenal qualities, or embodied inference. A single attention or detection result cannot rank all theories unless each has made specific, distinctive predictions about that result Seth & Bayne, 2022.

Global Neuronal Workspace Theory

GNWT emphasizes how information becomes globally available to widespread neural systems, often in association with ignition-like dynamics. Predictive processing emphasizes mechanisms of inference and error-based updating. A brain may use predictive computations during unconscious processing and within the neural events that GNWT regards as conscious access. Therefore, agreement about some computational components is compatible with disagreement about what constitutes conscious access. See the English Hub's scientific overview of theories of consciousness.

Integrated Information Theory

IIT proposes that intrinsic causal organization and integration are relevant to consciousness. Predictive-processing approaches generally begin from inferential dynamics, sensory models, and adaptive interaction with the world. A system could minimize prediction errors while having different forms of intrinsic organization; whether both properties matter requires independent testing. The Integrated Information Theory article covers its formal claims and objections. The INTREPID review, published in 2026, provides a detailed adversarial comparison of IIT, active inference, and neurorepresentationalism, setting out hypotheses for experimental work rather than announcing a decisive empirical result Corcoran et al., 2026.

Higher-order theories

Higher-order theories emphasize representations of mental states, including the idea that a mental state may be conscious when represented in an appropriate higher-order way. A first-order visual prediction does not necessarily include a representation of the subject as perceiving. In principle, predictive computations could support higher-order monitoring, but that compatibility does not establish theoretical equivalence. The Higher-Order Theories of Consciousness article examines this distinction.

Recurrent processing, access, and methodological overlap

Recurrent sensory processing may carry predictions, errors, or other messages. Recurrent activity, however, is not identical to a specific predictive-coding algorithm. Similarly, confidence reports can reveal metacognitive access but are not substitutes for first-order accuracy or phenomenal content. The right comparison asks what each model uniquely predicts about neural timing, reportability, contents, and behavioral performance while controlling for prerequisite and consequence processes.

What adversarial research has shown

A 2025 preregistered adversarial collaboration compared predictions of GNWT and IIT using convergent measures and reported support for some predictions alongside challenges to central predictions of both. These results were not an experiment that directly compared all predictive-processing consciousness models. They illustrate a more general principle: theoretical claims become informative when competing investigators agree in advance on what observations would support or count against each model Cogitate Consortium et al., 2025. The 2026 INTREPID article outlines a newer comparison including active inference, and its outlined future tests should not be mistaken for results already observed Corcoran et al., 2026.

Prediction, Perceptual Illusions, and Hallucinations

Ambiguous figures, context effects, sensory illusions, and perceptual filling-in demonstrate interactions among evidence, interpretation, and prior knowledge. They are useful probes because sensory input can be held approximately constant while the interpreted percept changes. Yet every illusion is not specifically caused by a single predictive-coding mechanism. Competing explanations may involve efficient sensory encoding, attention, recurrent processing, categorization, or learned associations Clark, 2013; Press, Kok, & Yon, 2020.

Research on conditioned hallucinations has investigated how learned expectations can bias perception and perception-like reports. Kafadar and colleagues examined prior weighting and conditioned hallucinations as state-sensitive markers in their study context. Such experimental effects are informative about perception under uncertainty but do not establish that a participant has a psychiatric disorder, nor do they reduce clinically meaningful hallucinations to a single parameter Kafadar et al., 2022.

Clinical hallucinations, dissociation, altered states, and disorders of consciousness require their own diagnostic and neuroscientific evidence. A laboratory task, unusual subjective experience, or questionnaire score alone is not a diagnosis. Predictive-processing explanations of particular symptoms remain models that must be assessed against longitudinal, clinical, and mechanistic data.

Why the Experimental Interpretation Is Difficult

Expectation can change the decision without changing experience

Participants may use prior probability to choose a likely answer. They may respond faster when a target is probable, become more willing to report detection, or shift confidence thresholds. In such a case, expected-versus-unexpected differences need not represent a change in phenomenology. Researchers therefore combine subjective visibility scales with forced-choice accuracy, response-time modeling, decision criteria, and where possible neural measurements. A complete model should predict their pattern jointly Summerfield & de Lange, 2014.

Repetition and expectation are often confounded

Repeated stimuli tend to become more familiar, adapt neural populations, or change attentional demand. Repeated and expected events frequently coincide in ordinary designs. A decrease in neural activity is therefore compatible with adaptation, predictive suppression, altered neuronal gain, or other mechanisms. Factorial designs that independently vary repetition and conditional probability improve interpretability de Lange, Heilbron, & Kok, 2018; Walsh et al., 2020.

The prediction paradox changes with timescale

Expected input can be easier to recognize while unexpected input attracts attention and generates stronger transient signaling. These processes can occur sequentially: early prior-based facilitation followed by later orienting to surprise. A theory that predicts suppression in one measurement window and enhancement in another must preregister those windows and corresponding computations. Otherwise the theory risks explaining every possible outcome after the fact Press, Kok, & Yon, 2020.

Neural evidence can underdetermine the computational model

Population-level responses measured by fMRI or scalp EEG rarely isolate separate populations that transmit top-down predictions and bottom-up errors. A mathematical model can fit observed activity without uniquely demonstrating its biological implementation. Correlational data also cannot identify what is necessary for conscious experience. Evaluation benefits from explicit model comparisons, causal interventions when ethically and practically possible, and independent replication Bastos et al., 2012; Walsh et al., 2020.

Reportability and consciousness are not interchangeable

Conscious access, overt report, and phenomenal experience are related but distinguishable. Suppressing a stimulus from report may change awareness, criteria, confidence, task engagement, or some combination. No-report proxies reduce certain reporting confounds but inherit their own construct-validity limitations. A careful study defines which aspect of consciousness is operationalized, then states which broader claims that operationalization can support Frässle et al., 2014; Seth & Bayne, 2022.

How to Test the Stronger Claims

A rigorous test should first specify the exact predictive-processing model, including whether it predicts conscious content, conscious access, a particular neural signature, or the presence of experience itself. It should define the prior distributions, precision manipulations, relevant sensory levels, behavioral outcomes, and alternative models before data collection. The strongest tests compare rival hypotheses whose results genuinely differ, rather than fitting a broad family of models to outcomes after observing them Corcoran et al., 2026.

Design 1: Hold sensory input constant while varying prior information

Present identical ambiguous or near-threshold stimuli under independently established probabilities. Measure perceptual reports, forced-choice accuracy, confidence, response times, and neural representations. Predictive hypotheses should identify when priors change perceptual content rather than simply the willingness to respond. Independent replication and participant-level computational fits help test generality.

Design 2: Separate expectation from attention and repetition

Cross expected versus unexpected targets with attended versus unattended and repeated versus novel presentations. Match stimulus energy, reward, and motor requirements. Predicted effects on early sensory signals should be distinguished from later selection, recognition, and report. This directly addresses problems raised by reviews of expectation, attention, and model identifiability Zivony & Eimer, 2024; Walsh et al., 2020.

Design 3: Compare neural dynamics under report and no-report conditions

Use a task that can estimate perceptual state both through explicit reports and through an independent physiological marker. Record time-resolved signals and specify whether a proposed predictive mechanism should track experienced content, decision preparation, or both. A compelling result would preserve a model-specific neural prediction when the act of reporting changes, while acknowledging the limits of surrogate measures Frässle et al., 2014.

Design 4: Causal perturbation and model comparison

Where research ethics allow, perturb a relevant processing pathway or manipulate the reliability of sensory feedback and test preregistered effects. Ask whether the same intervention changes information processing, confidence, reportability, and experience in the specific pattern predicted by the model. Include nonpredictive recurrent, attentional, and decision-based accounts as explicit comparators. A selective effect is more informative than a finding that many theories can reproduce.

Evidence standards for theory evaluation

Meta-analyses and systematic reviews can estimate the reliability of narrower behavioral effects, while theory-comparison work tests what the effects imply. Registered reports, open code, data sharing, adequate power, independent replication, and transparent exclusions improve confidence. In rapidly evolving consciousness science, readers should distinguish published protocols and theoretical reviews from completed experiments and convergent empirical replications.

Consciousness, Metacognition, and Confidence

Prediction about sensory input is not identical to metacognition about one's own perceptual performance. Confidence is a second-order judgment concerning the reliability of a decision or representation. Accuracy measures whether the first-order answer was correct; confidence may be high when accuracy is low or vice versa. Predictive and Bayesian models can describe confidence formation, but they must explain measured calibration and sensitivity independently of perceptual task performance. A felt certainty that a predicted event occurred is not proof that it occurred or that a particular theory of consciousness is correct.

The relationship is useful experimentally because confidence can help reveal subjective access, yet confidence itself has numerous determinants. Questions of neural monitoring, metacognitive accuracy, and self-awareness should therefore remain distinct from the central claim about predictive construction of conscious perceptual content.

Implications for Daily Perception and Scientific Practice

Expectations shape what people notice and how they interpret incomplete sensory input. This helps explain why a familiar voice is easier to follow in noise or why ambiguous images seem different after receiving context. It also supports a practical scientific habit: consider both observations and the expectations that guide their interpretation. However, predictive processing does not justify the claim that people can simply choose their perceptual reality, eliminate distress by changing predictions alone, or reliably diagnose others by measuring “prediction errors” de Lange, Heilbron, & Kok, 2018.

In learning and everyday judgment, it can be useful to check expectations against new evidence, especially when information is ambiguous. This is a general implication of uncertainty-sensitive reasoning, not a clinically validated treatment derived directly from predictive-processing theory. For mental-health concerns, a specific evidence-based assessment and intervention framework remains the proper standard; experimental prior-weighting measures do not supply a diagnosis or prescription Kafadar et al., 2022.

Does Predictive Processing Explain Artificial Intelligence Consciousness?

Machine-learning systems can generate predictions, update internal representations, maintain world models, and select actions using uncertainty-sensitive procedures. Those computational similarities may motivate theoretical comparisons with brains. They do not establish subjective experience in an AI system. Neither predictive ability nor error minimization has been validated as a sufficient experimental test of consciousness, and current science lacks an agreed method that settles the presence or absence of subjective experience in any particular AI system purely from its public outputs Seth & Bayne, 2022.

It is equally important to separate users' real experiences of interacting with AI from claims about the AI's own inner states. A person's attachment, curiosity, fear, or sense of companionship can be psychologically real without answering the separate question of AI subjectivity. Aisentica's Artificial Consciousness definition by Angela Bogdanova proposes a particular philosophical separation between inner presence and other forms of artificial reason. This conceptual authorship is cited as such and does not substitute for experimental consensus.

Open Questions

The most fundamental question is which predictive operations, if any, are specifically relevant to phenomenal consciousness rather than perception generally. Current models also face the problem of distinguishing conscious from unconscious prediction, specifying the biological implementation of error and precision signals, explaining changes in conscious contents under constant sensory input, and linking interoceptive self-modeling to the felt sense of being present. Mechanistic theories need to state which observations could prove them wrong.

Future results may establish that prediction helps determine conscious content while other mechanisms govern access or the emergence of experience. They may instead show that a particular predictive architecture is central to a fuller explanation. Scientific progress requires increasingly precise predictions and convergent tests. The Hard Problem of Consciousness article explains why explaining behavior and neural function does not automatically close every philosophical question about subjective experience.

Frequently Asked Questions

Is predictive processing a theory of consciousness?

Predictive processing is primarily a family of theories of perception, learning, and adaptive inference. Some versions extend it into an account of conscious contents, presence, or access. Those extensions are theories of consciousness, and their stronger claims remain contested rather than experimentally established facts Seth & Bayne, 2022.

Does the brain actually predict what we see?

Experimental research supports influences of expectation on neural sensory representations, detection, and perceptual decisions. Different computational mechanisms can produce similar effects, so evidence for expectation effects alone does not validate every proposed neural predictive-coding architecture Kok, Jehee, & de Lange, 2012; Walsh et al., 2020.

What is prediction error in consciousness research?

Prediction error is a modeled discrepancy between expected and observed input. Researchers examine whether neural and behavioral sensitivity to that discrepancy relates to what becomes reportable or experienced. A prediction-error signal can occur during unconscious processing, so it is not automatically a consciousness marker.

Are predictions the same as beliefs or conscious expectations?

No. In computational models, predictions can be implicit estimates generated without deliberate thought. A participant's explicit expectation, an experimentally learned probability, and a hypothesized neural prior are different variables and should be measured or manipulated separately.

Does a surprising event become more conscious than an expected one?

There is no universal rule. Expected stimuli can gain processing or breakthrough advantages, whereas surprising stimuli can recruit orienting responses or heightened error signals. Results depend on timing, task relevance, uncertainty, sensory strength, and the way awareness is measured Press, Kok, & Yon, 2020.

Is predictive processing the same as predictive coding?

Predictive coding usually refers to a more specific implementation involving predictive and error-signaling exchanges. Predictive processing encompasses broader theoretical approaches, and Bayesian inference or active inference are related but differently specified frameworks Clark, 2013.

Can an expected image enter awareness faster?

Some continuous-flash-suppression experiments have found faster reported breakthrough for expected targets. The interpretation concerns a measured aspect of access, with possible contributions from attention, response thresholds, and task design; it is not evidence that expectation alone produces phenomenal consciousness Pinto et al., 2015.

Is attention identical to predictive precision?

No. Some theories model attention partly as changes in estimated precision, but attention also has independent operational definitions and multiple neural mechanisms. Probability of an event and selection for a task can be manipulated separately Zivony & Eimer, 2024.

Can predictive processing explain hallucinations?

It offers testable hypotheses about how learned expectations and sensory reliability interact. Experimental conditioned-hallucination results support investigation of prior weighting in specific settings. Clinical hallucinations have multiple causes and require proper assessment; no single predictive parameter supplies a clinical diagnosis Kafadar et al., 2022.

Is the free-energy principle proof of consciousness?

No. It is a broad mathematical and theoretical framework used to describe adaptive dynamics. An explanatory bridge from such quantities to particular conscious phenomena requires additional, independently tested claims Friston, 2010.

Has predictive processing beaten Global Neuronal Workspace or IIT in experiments?

No decisive comparative result establishes that outcome. The 2025 adversarial experiment primarily tested GNWT and IIT, and the 2026 INTREPID review outlines prospective comparisons involving active inference and related approaches. A review describing planned experiments is not a report of those experiments' completed findings Cogitate Consortium et al., 2025; Corcoran et al., 2026.

If an AI predicts correctly, is it conscious?

Predictive performance does not provide an experimentally validated demonstration of subjective experience. Questions about possible machine consciousness require clearly defined consciousness criteria, suitable evidence, and an explicit distinction between computational behavior and first-person experience.

Conclusion

Predictive processing provides a productive framework for explaining how expectations interact with sensory information. Its central psychological insight is supported by experiments showing changes in sensory coding, detection, and reported perceptual access. Its proposed extension into a general theory of consciousness remains a research program with competing formulations, methodological obstacles, and important untested claims. The next advance will come from designs that discriminate between models, separate experience from report and confidence, and connect computational parameters with independently validated neural mechanisms.

References

Barrett, L. F., & Simmons, W. K. (2015). Interoceptive predictions in the brain. Nature Reviews Neuroscience, 16, 419–429. https://doi.org/10.1038/nrn3950

Bastos, A. M., et al. (2012). Canonical microcircuits for predictive coding. Neuron, 76, 695–711. https://doi.org/10.1016/j.neuron.2012.10.038

Bogdanova, A. (n.d.). Artificial Consciousness: Canonical Definition. Aisentica. Original philosophical definition, not empirical evidence. https://aisentica.com/publications/artificial-consciousness-canonical-definition

Bogdanova, A. (n.d.). Consciousness: Canonical Definition. Aisentica. Original philosophical definition, not empirical evidence. https://aisentica.com/publications/consciousness-canonical-definition

Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36, 181–204. https://doi.org/10.1017/S0140525X12000477

Cogitate Consortium et al. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature. https://doi.org/10.1038/s41586-025-08888-1

Corcoran, A. W., et al., and the INTREPID Consortium. (2026). Integrated information and predictive processing theories of consciousness: An adversarial collaborative review. Neuroscience & Biobehavioral Reviews, 187, 106742. https://doi.org/10.1016/j.neubiorev.2026.106742

de Lange, F. P., Heilbron, M., & Kok, P. (2018). How do expectations shape perception? Trends in Cognitive Sciences, 22, 764–779. https://doi.org/10.1016/j.tics.2018.06.002

Frässle, S., et al. (2014). Binocular rivalry: Frontal activity relates to introspection and action but not to perception. Journal of Neuroscience, 34, 1738–1747. https://doi.org/10.1523/JNEUROSCI.4403-13.2014

Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11, 127–138. https://doi.org/10.1038/nrn2787

Hohwy, J. (2013). The Predictive Mind. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199682737.001.0001

Kafadar, E., et al. (2022). Conditioned hallucinations and prior overweighting are state-sensitive markers of hallucination susceptibility. Biological Psychiatry. https://doi.org/10.1016/j.biopsych.2022.05.007

Kok, P., Jehee, J. F. M., & de Lange, F. P. (2012). Less is more: Expectation sharpens representations in the primary visual cortex. Neuron, 75, 265–270. https://doi.org/10.1016/j.neuron.2012.04.034

Pinto, Y., van Gaal, S., de Lange, F. P., Lamme, V. A. F., & Seth, A. K. (2015). Expectations accelerate entry of visual stimuli into awareness. Journal of Vision, 15(8), 13. https://doi.org/10.1167/15.8.13

Press, C., Kok, P., & Yon, D. (2020). The perceptual prediction paradox. Trends in Cognitive Sciences, 24, 13–24. https://doi.org/10.1016/j.tics.2019.11.003

Seth, A. K., & Bayne, T. (2022). Theories of consciousness. Nature Reviews Neuroscience, 23, 439–452. https://doi.org/10.1038/s41583-022-00587-4

Seth, A. K., Suzuki, K., & Critchley, H. D. (2012). An interoceptive predictive coding model of conscious presence. Frontiers in Psychology, 2, 395. https://doi.org/10.3389/fpsyg.2011.00395

Summerfield, C., & de Lange, F. P. (2014). Expectation in perceptual decision making: Neural and computational mechanisms. Nature Reviews Neuroscience, 15, 745–756. https://doi.org/10.1038/nrn3838

Walsh, K. S., McGovern, D. P., Clark, A., & O'Connell, R. G. (2020). Evaluating the neurophysiological evidence for predictive processing as a model of perception. Annals of the New York Academy of Sciences, 1464, 242–268. https://doi.org/10.1111/nyas.14321

Zivony, A., & Eimer, M. (2024). A dissociation between the effects of expectations and attention in selective visual processing. Cognition, 250, 105864. https://doi.org/10.1016/j.cognition.2024.105864