A Theory of Consciousness

Imagine noticing a spider crawling across your hand. Before you have consciously considered what it is, much of the relevant processing may already have happened. Visual systems have detected and classified features of the animal, learned associations may have activated a threat response, muscles may tense, autonomic state may change, and an impulse to withdraw may already be developing. A moment later, however, the situation can change. You recognise the spider, recall that it is harmless, notice your own fear, remember previous encounters and decide not to follow the immediate impulse. With repeated safe exposure, even that first automatic reaction may eventually diminish.

This small example contains much of the puzzle of consciousness. The brain can perceive, classify, predict, evaluate and prepare actions without waiting for conscious deliberation. A substantial amount of sophisticated processing can occur outside awareness.1–3 Yet there are also situations in which leaving those specialised processes to continue on their own appears insufficient. Their outputs may conflict, circumstances may be unfamiliar, a habitual solution may suddenly fail, or information from several domains may have to be combined before the organism can determine what to do. At such moments, something like a higher level of arbitration seems functionally useful.

This suggests a different way of asking what consciousness is for. Instead of beginning with the mystery of subjective experience, we can begin with a more biological problem: what does a complex organism gain from having a process that can interrupt, compare, coordinate and sometimes override the outputs of systems that normally operate independently?

The hypothesis developed here is that consciousness is closely associated with precisely that function. It may be the brain’s mechanism for arbitrating situations that cannot adequately be left to automatic processing: novel, conflicting, uncertain or consequential problems that require relevant information from across the brain and body to be brought together into an informed determination of what happens next.

That arbitration need not depend on a literal central executive somewhere in the brain. It can emerge from distributed processes whose combined activity becomes functionally central because it determines which trajectory governs the organism’s behaviour. Nor does every conscious moment need to culminate in an overt action. Continuing, waiting, redirecting attention, searching memory, reconsidering a goal or maintaining a problem in focus can all count as outcomes of the same broader function.

What matters is that specialised processes which normally operate with considerable autonomy can, when circumstances demand it, be coordinated around a common focal problem.

A brain that already knows how to do things

It is easy to think about consciousness as the place where cognition happens. Mechanistically, that is almost certainly wrong. The nervous system is massively parallel. Specialised cortical and subcortical systems process visual features, sounds, bodily signals, memories, rewards, threats and possible actions simultaneously, and most of this activity never becomes part of what we can report experiencing.

Research on unconscious cognition has repeatedly shown that the relevant boundary is not a simple divide between elementary unconscious processing and sophisticated conscious processing. Information outside reported awareness can influence behaviour, response inhibition, conflict processing and some forms of cognitive control.2,3 Likewise, attention, decision-making and integration cannot individually be identified with consciousness, because versions of each can occur without awareness.4

This creates an important constraint on any functional account. Consciousness cannot simply be what happens when the brain performs a difficult calculation. The more promising distinction concerns how flexibly information can be maintained, combined and reused across contexts. Dehaene and Naccache argued early on that conscious processing is particularly important for durable information maintenance, novel combinations of operations and intentional behaviour.1 Later work has complicated that picture, but the broader contrast remains useful: unconscious processing can be remarkably powerful while tending to be more constrained by the pathways, associations and task structures through which it operates.

That division of labour is also what an efficient biological system should look like. Most problems ought to be solved automatically whenever possible. A system that consciously reconsidered every muscle movement, every familiar percept and every routine action would be hopelessly inefficient. Specialised processing is fast and parallel precisely because it does not require broad coordination across the entire organism.

The interesting question begins when that arrangement stops being sufficient.

The arbitration problem

A flexible organism constantly encounters situations in which different systems favour incompatible outcomes. A threat response promotes withdrawal while explicit knowledge says there is no danger. Immediate reward favours one choice while a long-term objective favours another. A well-learned habit produces a response that was appropriate yesterday but is wrong today. Visual evidence suggests one interpretation while remembered context suggests another.

Many such conflicts can be resolved locally and automatically. The cases of interest are those in which no existing specialised solution is reliable enough. Relevant information may then have to be drawn from processes that ordinarily serve very different functions: current perception, memory, bodily state, social knowledge, learned values, long-term goals and predictions of future consequences.

Ultimately, incompatible possibilities must be resolved into coherent behaviour by one organism. This is where the idea of a functional “final say” becomes useful. It does not imply a single anatomical command centre or a hidden observer issuing instructions. It refers to the fact that distributed competition and coordination must eventually produce one effective trajectory: act, wait, continue, stop, investigate, redirect attention or construct another plan.

The present hypothesis locates the functional value of consciousness close to this process of flexible arbitration. Consciousness becomes especially useful when the organism cannot safely or effectively accept the first answer produced by its specialised machinery.

Why consciousness has a focus

This gives attention a central role, though the term must be used carefully. Attention and consciousness are closely related but not identical. Experimental studies show that some attentional selection can occur without awareness, and neural effects of attention and awareness can sometimes be dissociated.4 The broader word “attention” also covers a range of mechanisms, from local enhancement of sensory processing to deliberate maintenance of a problem in working focus.

For the present model, the crucial concept is priority. At any moment, enormous numbers of processes could potentially influence the organism, while flexible coordination is limited and comparatively expensive. Some mechanism must determine which problem deserves access to that capacity.

Physical intensity is one way to win that competition, but relevance can matter more. A quiet mention of your name can become more important than a loud but irrelevant sound. A stimulus previously associated with reward can continue to capture attention after the reward contingency has disappeared.5 Experience therefore changes what acquires priority.

This suggests a functional progression in which relevance influences priority, priority establishes focus, focus allows broader integration, and integration enables arbitration. The brain does not implement this as a neat one-directional pipeline; recurrent loops and feedback dominate neural processing. Still, the sequence captures an important logic. Something becomes sufficiently important to interrupt or modify ongoing activity. Attention concentrates limited flexible-processing capacity around the problem. Information relevant to resolving that problem can then constrain the same coordination process, allowing competing possibilities to be evaluated until a course of action emerges.

From this perspective, the focus of consciousness is not merely a processing limitation. It is part of what makes arbitration possible. Thousands of processes can proceed in parallel, while flexible coordination requires some problem to become temporarily privileged.

The motive matters

That immediately raises a deeper question: priority according to what?

A living system is not a neutral information processor. It has a body whose internal variables must remain within viable ranges. It has needs, capabilities and vulnerabilities. It learns which events predict reward or harm. It develops goals through biology, experience, social learning and culture. In humans, it can eventually represent consequences extending decades into the future.

Motivational relevance therefore becomes part of the explanation for why some information acquires attention. Research on value-driven attention provides a clear example. Stimuli associated with reward can later attract attention even when they are no longer useful for the current task.5 Learning has changed the competitive weight of the representation.

The same principle extends beyond reward. Organisms learn what predicts danger, food, pain, social approval, rejection, opportunity and loss. Humans extend this mechanism still further through abstraction. Money, reputation, examination results, professional status, moral commitments and scientific ideas can all become motivationally significant despite having no innate sensory importance.

This matters because a system capable of arbitration needs criteria for deciding which outcome is preferable. A perceptual signal does not come with its significance built in. Significance is created by the relationship between the signal, the organism’s history, its current state, its goals and its possible futures.

The detective’s question—what was the motive?—has a biological analogue. Behaviour becomes intelligible only when we understand why one possible outcome mattered more than another.

Feeling is part of the calculation

The state of the body is an important part of this evaluation. Interoception refers broadly to the sensing and integration of signals originating within the body, involving distributed systems that include brainstem, hypothalamic, insular and cingulate regions. These systems participate in homeostatic regulation, emotion and cognition.6,7

Different theories disagree about exactly how bodily information contributes to conscious feeling. Damasio and colleagues emphasise the representation of bodily and homeostatic states as foundational to feelings, while predictive-processing approaches describe emotion partly in terms of inference about the causes of interoceptive signals.6–8 The present argument does not depend on choosing one account as complete.

The important point is that a nervous system continually represents both the external world and the condition of the organism encountering that world. The spider is therefore not processed merely as a visual object. Its representation can interact with learned associations, changes in heart rate, muscle tension, remembered experiences, predicted harm and possible actions.

This makes feeling functionally relevant to arbitration. Affect can provide information about how strongly a state matters and in what direction. Attention helps determine what deserves priority, while affective and interoceptive processes help determine why it deserves priority for this particular organism.

That relationship offers a potentially important bridge between function and experience. Conscious states feel significant because the information being coordinated is not abstract in the biological sense. It is coupled to a body, a history of reward and punishment, current needs and predicted consequences.

Consciousness works with summaries

Despite this sophistication, consciousness has remarkably little access to the computations that generate its contents. When you recognise a familiar person, you do not consciously experience the successive visual operations required to make recognition possible. You simply see Sarah. When unease appears, you may become aware that something is wrong without knowing which cues produced that judgement.

This is exactly what an efficient coordination architecture would need. Specialised processors do not have to expose their implementation details to every other system. They need to make useful outputs available.

Conscious processing can therefore operate on compressed representations such as “dangerous,” “familiar,” “I know this,” “I am afraid” or “this matters.” Each may summarise an enormous amount of inaccessible processing. This also helps explain why introspection is often unreliable. Conscious access to a conclusion does not guarantee conscious access to its causal history. The system may receive the outcome and then construct an explanation from whatever information is currently available.

Sometimes that explanation is accurate. Sometimes it is only plausible.

This mixture of power and ignorance is a recurring feature of consciousness. We can reason about highly abstract information while remaining almost completely blind to the machinery that produced the representations we are reasoning with.

Passenger and participant

This architecture also makes sense of a familiar duality in conscious experience. Sometimes consciousness appears to be a passenger. An action has begun before deliberate thought, a judgement simply arrives, or a feeling appears without invitation. At other times, conscious processing clearly participates in what happens next. Explicit information changes an action, a remembered rule suppresses a habitual response, deliberation alters a plan, and repeated consciously guided behaviour contributes to later learning.

These are not necessarily contradictory observations. Consciousness does not need to originate every process in order to influence arbitration among processes. Indeed, if it had to generate every percept, impulse and proposal itself, the efficiency of specialised parallel processing would be lost.

The spider example again illustrates the point. Automatic threat processing may already have produced a strong withdrawal tendency. Explicit knowledge that the spider is harmless can nevertheless alter what the organism does. Repeated outcomes of that kind can later change the automatic response itself.

A useful functional cycle therefore emerges: novel or conflicting situations recruit focal coordination; successful resolutions influence behaviour and learning; repeated success allows parts of the solution to become increasingly automatic. When circumstances unexpectedly change, the automated solution can fail and the problem returns to conscious focus.

Learning, in that sense, often teaches the organism how not to need consciousness for the same problem again.

Arbitration needs a reference point

The argument becomes more interesting when we ask what arbitration is actually for. Possible outcomes can only be evaluated in relation to the organism whose state they affect. The nervous system therefore benefits from maintaining information about its own body, capabilities, current needs, previous experiences, social commitments, goals and possible future states.

At a primitive level, this need does not require anything resembling the adult narrative self. Sensorimotor control already depends on distinctions between self-produced and externally produced events, and research on agency and minimal selfhood suggests roles for multisensory integration, bodily prediction and interoception in generating elementary forms of ownership and agency.9,10

As behaviour becomes more flexible, however, a richer model of the organism becomes useful. Can I reach this object? Am I injured? Have I encountered this before? What am I trying to achieve? What happened the last time I did this? What will happen to me if I choose one option over another?

The organism becomes one of the most important models used by its own control architecture.

This offers a functional route toward the self-awareness component of consciousness. Arbitration needs an organism-relative frame of reference. The self-model supplies that frame.

The self is not the observer

Subjectively, it often feels as though there is an “I” inside the brain observing experiences and making decisions. Mechanistically, that intuition creates an obvious problem: if an inner observer receives conscious information, what observes the observer?

A self-model offers a different possibility. The self need not be the observer of the coordination process. It can instead be one of the most important representations within it.

Information is continually evaluated in relation to the organism represented by that model. What is happening to me? What do I know about this? How do I feel? What can I do? What do I want? What will happen if I choose this?

The first-person perspective of consciousness may partly arise because flexible coordination is organised around this organismic reference point. The system does not first require an independently existing conscious self and then ask what matters to it. Increasingly sophisticated control creates increasing functional value in modelling the organism for which control is being exercised.

The “I” may therefore be part of the solution to the arbitration problem.

From organism to self

This distinction also prevents us from identifying basic consciousness with mature human self-awareness. Infants almost certainly have conscious experiences before possessing anything resembling an adult autobiographical self. Neural studies have identified candidate signatures of conscious perceptual access in infants as young as five months, although transferring adult markers of consciousness to infants remains methodologically difficult and conclusions are strongest when multiple indicators converge.11,12

Self-representation appears to develop in layers rather than appearing all at once. A useful conceptual progression might run from organismic reference to embodied self, agentive self, social self, reflective self and narrative self. These are not meant as rigid developmental stages with precise age boundaries. They distinguish functions that are often collapsed into the single term “self-awareness.”

The embodied organism can distinguish its own state and actions from the environment. The agentive self represents itself as a source of actions and consequences. The social self represents itself as one agent among others. The reflective self can direct cognition toward its own beliefs, motives and mental states. The narrative self organises these representations across an extended personal history and imagined future.

Each layer expands what can be included in arbitration.

Other minds enlarge the problem

Social cognition creates one of the most dramatic expansions. A child gradually learns that other people possess information, beliefs, goals and perspectives that can differ from its own. Classic false-belief research shows a robust developmental progression in explicit understanding that another person can believe something the child knows to be false.13 More complex recursive reasoning continues to develop: I think that she believes that he knows.

Once other minds become part of the model, the space of possible consequences expands enormously. A decision can depend on what another person believes, what they believe about us, and what we predict they will do as a result. Reputation, trust, deception, obligation, loyalty and cooperation become relevant to behaviour.

The self-model consequently acquires a social position. The organism no longer represents only “me,” but “me among other selves.”

That social recursion is likely to be an important part of the unusually elaborate self-awareness characteristic of adult humans.

Language changes the scale again.

Language makes the self explicit

Basic consciousness does not require language. Infant research alone makes a simple identity between language and consciousness implausible, and neurological evidence shows that aspects of awareness and selfhood can persist despite severe disturbances of language.11,12,14

Language nevertheless gives the arbitration system an extraordinary new capacity. Internal states can become symbolic objects.

The organism can experience fear. It can then represent that state as “I am afraid.” That representation can itself become the object of attention. The system can ask why it is afraid, whether the fear is justified, whether it wants the fear to control behaviour, and eventually what it means about the kind of person it believes itself to be.

The output of one round of conscious processing can therefore become the input to another. Experience becomes representation; representation can become linguistic abstraction; abstraction can become a new focus of attention; and the resulting reflection can feed back into further arbitration.

This recursive loop gives language an important role in human self-awareness without making language the origin of consciousness. Research on inner speech supports roles in planning, self-regulation and self-reflection, while substantial individual variation and evidence of self-awareness without normal inner speech argue strongly against equating the two.14–17 Recent work on language and consciousness likewise suggests that lower-level linguistic processing can occur without awareness, while higher-order semantic integration and reflective thought are more tightly linked to conscious processing.18

Language is therefore better understood as an amplifier of reflective consciousness.

The self becomes an object in its own world

Language also allows the self-model to escape the immediate present. “I” can refer to the organism sitting here now, but also to the child I remember being, the person I expect to become, the person another individual believes me to be, or a hypothetical future self decades away.

This allows present behaviour to be evaluated in relation to a symbolic future organism that does not yet exist. Social language adds roles, reputations, identities and descriptions acquired from other people. The self becomes temporally extended, socially situated and increasingly abstract.

At sufficient levels of recursion, propositions about the self become ordinary objects of arbitration. I believe this. My belief may be wrong. I want this. Why do I want it? My immediate desire conflicts with another goal. Perhaps I should change the goal. What kind of person do I want to become?

At this point, the architecture has acquired a striking new property. It was useful to model the organism in order to arbitrate effectively on its behalf. Increasing representational sophistication produced a richer model of that organism. Language made parts of the model explicit and manipulable. Recursive cognition then allowed the model itself to become a problem presented to the same arbitration machinery.

The “I” became both the reference point of the decision and one of the things about which decisions could be made.

This may be one reason human consciousness contains such a powerful sense of self-awareness.

Why it feels like it happens to me

This offers a possible bridge between function and phenomenology. Consciousness does not feel like a neutral stream of data. Things happen to me. They matter to me. I experience fear, curiosity, embarrassment, desire, doubt, pain and relief.

If conscious arbitration integrates information according to its relevance to a model of the organism, that first-person structure is exactly what we should expect. External information is combined with bodily state, memory, motivation and prediction. Possible consequences are evaluated in relation to an organism represented within the system. The self-model supplies the persistent reference around which the problem is organised.

The experienced “I” therefore need not be an additional entity watching consciousness. It may be the form taken by the organism’s model when that model participates in conscious coordination.

This does not solve the philosophical hard problem by definition. There remains a legitimate question about why any physical process should possess phenomenal character at all. It does, however, reduce the number of independent mysteries we need to posit. The focus, relative unity, emotional significance, agency and first-person structure of ordinary human consciousness may be mutually related consequences of the same functional architecture.

Report comes later

Language also makes it important to distinguish experience from reporting experience. Traditional consciousness experiments often ask participants to say what they saw, but report requires additional processing: decision, memory, action preparation and often language. No-report paradigms were developed precisely to separate neural processes associated with conscious contents from processes required to describe those contents.19

These paradigms have limitations of their own. Eliminating an overt response does not guarantee that internal reflection or task-related cognition has disappeared.20 The debate nonetheless reinforces an important point: experience, cognitive access and explicit report are separable enough that they must be distinguished experimentally.

Within the present model, external report is a downstream capability. Information involved in conscious arbitration can become available to language and deliberate communication without linguistic report being what originally makes the information conscious.

Human language makes consciousness extraordinarily reportable. That does not mean report creates consciousness.

A working model

The proposed architecture can now be stated more precisely.

The brain contains many specialised processes capable of sophisticated operation largely autonomously and in parallel. As long as their outputs are reliable and mutually compatible, extensive flexible coordination adds little value. Learned routines can govern behaviour efficiently.

Certain situations create a different requirement. Something becomes sufficiently novel, conflicting, uncertain or consequential that the existing automatic response is inadequate or should not simply be trusted. Relevance gives the problem priority. Attention establishes a temporary focus. Information needed to address the problem becomes available across otherwise partly independent systems. Perception, memory, bodily state, learned value, current goals, possible actions, social knowledge and predicted consequences can constrain one another.

The system then arbitrates among the resulting possibilities until a temporary distributed state acquires a functional final say. The phrase does not imply a final neural executive. It means that competing influences have been resolved sufficiently for one trajectory to govern what the organism does next.

The result can then modify action and learning. Repeated successful solutions can become increasingly automatic, reducing their future requirement for conscious arbitration. Novelty, error or conflict can return them to focal processing.

A simplified functional sequence therefore looks like this:

automatic processing → relevance → priority → focus → integration → arbitration → action → learning → automation

The self-model occupies a special position because arbitration requires information about the organism for which consequences matter. In humans, social cognition greatly enriches that model. Language then makes parts of it explicit, symbolic and recursively available to the arbitration process itself.

A second loop emerges:

experience → self-relevant representation → linguistic abstraction → renewed attention → reflection → revised judgement or goal

Together, these loops provide a plausible functional account of several properties usually grouped under consciousness. Its focus reflects the need to privilege a problem for flexible arbitration. Its limited capacity reflects the cost of broad coordination compared with parallel specialised processing. Its relative unity reflects the requirement that incompatible possibilities ultimately produce coherent behaviour in one organism. Its emotional character reflects the integration of information with bodily state and learned value. Its agency reflects the ability of coordinated processing to alter which proposal governs behaviour. Its first-person perspective reflects the self-model around which many consequences are organised, while reflective self-awareness is amplified when that model itself becomes available for attention and arbitration.

The curious mixture of causal influence and introspective ignorance follows from the same architecture. Conscious processing can use the outputs of specialised systems without possessing access to the computations that generated them.

How this relates to existing theories

None of the constituent mechanisms is new. Global Workspace and Global Neuronal Workspace theories emphasise broad availability of conscious information to specialised processors.1,21 Higher-order theories emphasise representations of mental states. Recurrent-processing theories emphasise recurrent neural interactions. Predictive approaches describe perception and action in terms of inference. Attention Schema Theory proposes that the brain constructs simplified models of attention that contribute to awareness and control.22

Contemporary consciousness science has not established a single winner among these theories. A major review in 2022 emphasised both their differences and the difficulty of translating many of their claims into decisive experimental contrasts.23 A large preregistered adversarial collaboration published in 2025 directly tested predictions of Global Neuronal Workspace Theory and Integrated Information Theory and found results that challenged important predictions of both rather than producing a simple victory for either.24

The present proposal can therefore be understood as a functional synthesis rather than a rival claim that one existing theory should be discarded. Global availability may describe how information escapes local specialisation. Attention research helps explain how information acquires priority. Cognitive-control research examines aspects of arbitration. Affective and interoceptive neuroscience investigates how consequences acquire organismic significance. Predictive processing addresses how current information is evaluated against models and possible futures. Research on agency and selfhood examines representations of the organism performing and experiencing actions. Developmental psychology shows how increasingly sophisticated models of other minds and the self emerge. Language research shows how symbolic representation adds recursive and temporally extended forms of self-reflection.

These may be related because they contribute to a common functional problem: how a living organism determines what matters now and what to do about it when its specialised automatic machinery does not provide a sufficient answer.

A direct test

The central hypothesis suggests a relatively straightforward experimental strategy. Instead of searching only for neural signatures associated with conscious report, we can manipulate the degree to which a task requires genuine arbitration.

Participants could receive visual, auditory and contextual information under four conditions. In the simplest condition, one signal directly determines the correct response. A second requires two signals to be integrated. A third creates conflict between signals. In the strongest condition, no source has fixed priority: a changing contextual rule determines which competing source should control behaviour, and novel combinations require application of the rule rather than retrieval of a memorised response.

The critical information can be manipulated around the threshold of conscious visibility using established masking techniques.

A second dimension would vary relevance independently of sensory input. Some decisions would have trivial consequences, while others would affect a meaningful reward, loss, current goal or socially relevant outcome.

The result would be a matrix:

For each condition, researchers could measure objective performance, subjective visibility, confidence, autonomic responses and neural indicators of local versus broader information availability.

The hypothesis does not predict that unconscious processing should collapse as soon as information must be integrated. Existing evidence makes that unlikely. The more interesting prediction is that unconscious processing should reach its clearest limits when a task demands genuinely novel, flexible and context-sensitive arbitration among competing information sources.

Relevance should alter this relationship. Information associated with larger consequences should acquire priority more readily, potentially changing the threshold at which it becomes available for sustained coordination.

Training provides a further test. With repeated practice, parts of the task should become more automatic and the demand for conscious arbitration should decline. If the rule is suddenly changed, the automated solution should fail and the problem should return to focal conscious processing until a new reliable solution is learned.

The strongest result against the model would be equally informative. If participants can use genuinely unconscious critical information to identify competing proposals, interpret a novel rule, determine which source should have priority and generalise that arbitration to combinations they have never encountered, then the proposed functional boundary is in the wrong place.

That would force the model to become more precise.

The puzzle may already be partly assembled

An important feature of this hypothesis is that it does not require the discovery of an entirely unknown brain function. Much of the relevant machinery is already being studied in separate scientific traditions.

Attention researchers ask how information gains priority. Cognitive-control researchers investigate competition and response regulation. Learning researchers ask how effortful behaviour becomes automatic. Affective and interoceptive neuroscience examine how bodily state and value influence cognition. Memory and predictive-processing research investigate how absent past and future events influence present behaviour. Agency research asks how the nervous system represents its own actions and consequences. Developmental psychology studies the construction of models of self and other minds. Language research explores how symbolic representations transform planning and self-regulation. Consciousness research asks which processes distinguish conscious from unconscious information.

The possibility worth considering is that many of these are pieces of the same architecture viewed from different directions.

That would change the aim of a useful consciousness theory. The challenge would no longer be only to identify another neural correlate of awareness, but to determine whether established mechanisms can be assembled into a causal explanation of why consciousness has the properties it does.

Why is conscious processing focused and limited? Why does novelty recruit it? Why does practice make it recede? Why can it override some automatic responses while remaining powerless over others? Why does information in consciousness seem to matter rather than simply exist? Why is experience organised around a self, and why can that self itself become an object of thought? Why does language amplify the phenomenon so dramatically?

The arbitration hypothesis offers one possible answer. A living system contains many specialised processes that normally function without broad supervision. When their outputs become insufficient, incompatible or important enough to reconsider, limited flexible capacity is concentrated around the problem. Relevant information from across brain and body becomes mutually influential. Possible outcomes are evaluated relative to an organism represented through an increasingly sophisticated self-model. The resulting coordination determines what happens next and can change future automatic behaviour through learning.

In humans, language transforms the architecture further. The organism can symbolically represent its own states, beliefs, motives, history and possible futures. Those representations can themselves become objects of attention and arbitration. The system can reconsider not only what it should do, but why it wants to do it and whether the goals directing the decision should themselves be changed.

From the outside, we describe attention, salience, cognitive control, interoception, affect, memory, prediction, learning, agency, language and self-representation. From the inside, something has become important to me, different possibilities compete, relevant knowledge and feelings are brought to bear, and eventually one course of action gains control.

Perhaps the apparent observer at the centre of this process is not an additional entity waiting to be found. The “I” may be the increasingly sophisticated model of the organism for whose benefit the arbitration is taking place. Our unusually strong human self-awareness may then arise when language allows that model to become explicit, recursive, social and extended through time.

On this view, consciousness is not merely information becoming globally available. Its functional significance lies in what broad availability makes possible: the capacity to focus the organism’s limited flexible processing on what matters, bring otherwise separate sources of information into a common decision context, and resolve competing possibilities into a coherent response.

That is a plausible reason for such a mechanism to have enormous evolutionary value. It also offers a plausible explanation for why consciousness has the form we experience: focused, limited, affective, first-personal, sometimes decisive, sometimes merely observant, and capable of turning back upon itself.

The scientific task is now to see how well the pieces already known from neuroscience and psychology fit this architecture, where the gaps remain, and whether experiments aimed specifically at those gaps support or undermine the model.

That is a more interesting question than asking where consciousness sits in the brain.

It asks what consciousness is doing there in the first place.

Notes

1. “Arbitration” is a functional term. It does not imply a single anatomical decision centre. The hypothesis is compatible with distributed competition, recurrent processing and large-scale network dynamics.

2. “Final say” refers to behavioural outcome, not a literal neural executive. Competing processes eventually have to produce one effective trajectory for the organism, even if that trajectory is to wait, continue, redirect attention or gather more information.

3. “Automatic” does not mean simple. Unconscious and automated processes can be highly sophisticated. The proposed distinction concerns flexible, context-sensitive use of information across otherwise partly independent systems.

4. The self-model is not proposed as a prerequisite for all consciousness. The model distinguishes basic conscious experience from increasingly sophisticated embodied, agentive, social, reflective and narrative forms of self-awareness.

5. Language is treated as an amplifier of reflective self-awareness rather than the origin of consciousness. It allows internal states and self-representations to become explicit symbolic objects that can be returned to the same process of attention and arbitration.

References

1. Dehaene, S., & Naccache, L. (2001). Towards a cognitive neuroscience of consciousness: Basic evidence and a workspace framework. Cognition, 79(1-2), 1-37. https://doi.org/10.1016/S0010-0277(00)00123-2.
2. van Gaal, S., de Lange, F. P., & Cohen, M. X. (2012). The role of consciousness in cognitive control and decision making. Frontiers in Human Neuroscience, 6, 121. https://doi.org/10.3389/fnhum.2012.00121.
3. van Gaal, S., Lamme, V. A. F., Fahrenfort, J. J., & Ridderinkhof, K. R. (2011). Dissociable brain mechanisms underlying the conscious and unconscious control of behavior. Journal of Cognitive Neuroscience, 23(1), 91-105. https://doi.org/10.1162/jocn.2010.21431.
4. Koch, C., & Tsuchiya, N. (2007). Attention and consciousness: Two distinct brain processes. Trends in Cognitive Sciences, 11(1), 16-22. https://doi.org/10.1016/j.tics.2006.10.012.
5. Anderson, B. A., Laurent, P. A., & Yantis, S. (2011). Value-driven attentional capture. Proceedings of the National Academy of Sciences, 108(25), 10367-10371. https://doi.org/10.1073/pnas.1104047108.
6. Damasio, A., & Carvalho, G. B. (2013). The nature of feelings: Evolutionary and neurobiological origins. Nature Reviews Neuroscience, 14, 143-152. https://doi.org/10.1038/nrn3403.
7. Carvalho, G. B., & Damasio, A. (2021). Interoception and the origin of feelings: A new synthesis. BioEssays, 43(6), e2000261. https://doi.org/10.1002/bies.202000261.
8. Seth, A. K., & Critchley, H. D. (2013). Extending predictive processing to the body: Emotion as interoceptive inference. Behavioral and Brain Sciences, 36(3), 227-228. https://doi.org/10.1017/S0140525X12002270.
9. Limanowski, J., & Blankenburg, F. (2013). Minimal self-models and the free energy principle. Frontiers in Human Neuroscience, 7, 547. https://doi.org/10.3389/fnhum.2013.00547.
10. Seth, A. K., & Tsakiris, M. (2018). Being a beast machine: The somatic basis of selfhood. Trends in Cognitive Sciences, 22(11), 969-981. https://doi.org/10.1016/j.tics.2018.08.008.
11. Kouider, S., Stahlhut, C., Gelskov, S. V., Barbosa, L. S., Dutat, M., de Gardelle, V., Christophe, A., Dehaene, S., & Dehaene-Lambertz, G. (2013). A neural marker of perceptual consciousness in infants. Science, 340(6130), 376-380. https://doi.org/10.1126/science.1232509.
12. Bayne, T., Seth, A. K., Massimini, M., Shepherd, J., Cleeremans, A., Fleming, S. M., Malach, R., Mattingley, J. B., Menon, D. K., Owen, A. M., Peters, M. A. K., Razi, A., & Mudrik, L. (2024). Tests for consciousness in humans and beyond. Trends in Cognitive Sciences. See also the 2024 literature on cluster-based approaches to infant consciousness.
13. Wellman, H. M., Cross, D., & Watson, J. (2001). Meta-analysis of theory-of-mind development: The truth about false belief. Child Development, 72(3), 655-684. https://doi.org/10.1111/1467-8624.00304.
14. Mitchell, R. W. (2009). Self-awareness without inner speech: A commentary on Morin. Consciousness and Cognition, 18(2), 532-534. https://doi.org/10.1016/j.concog.2008.12.003.
15. Perrone-Bertolotti, M., Rapin, L., Lachaux, J.-P., Baciu, M., & Lœvenbruck, H. (2014). What is that little voice inside my head? Inner speech phenomenology, its role in cognitive performance, and its relation to self-monitoring. Behavioural Brain Research, 261, 220-239. https://doi.org/10.1016/j.bbr.2013.12.034.
16. Alderson-Day, B., & Fernyhough, C. (2015). Inner speech: Development, cognitive functions, phenomenology, and neurobiology. Psychological Bulletin, 141(5), 931-965. https://doi.org/10.1037/bul0000021.
17. Fernyhough, C., & Borghi, A. M. (2023). Inner speech as language process and cognitive tool. Trends in Cognitive Sciences, 27(12), 1180-1193. https://doi.org/10.1016/j.tics.2023.08.014.
18. Aubinet, C., Gosseries, O., & Majerus, S. (2026). The interaction between language and consciousness. Neuroscience & Biobehavioral Reviews, 180, 106498. https://doi.org/10.1016/j.neubiorev.2025.106498.
19. Tsuchiya, N., Wilke, M., Frässle, S., & Lamme, V. A. F. (2015). No-report paradigms: Extracting the true neural correlates of consciousness. Trends in Cognitive Sciences, 19(12), 757-770. https://doi.org/10.1016/j.tics.2015.10.002.
20. Duman, I., et al. (2022). The limits of no-report paradigms in consciousness research. PMID 35634201.
21. Mashour, G. A., Roelfsema, P., Changeux, J.-P., & Dehaene, S. (2020). Conscious processing and the Global Neuronal Workspace hypothesis. Neuron, 105(5), 776-798. https://doi.org/10.1016/j.neuron.2020.01.026.
22. Graziano, M. S. A., & Webb, T. W. (2015). The attention schema theory: A mechanistic account of subjective awareness. Frontiers in Psychology, 6, 500. https://doi.org/10.3389/fpsyg.2015.00500.
23. Seth, A. K., & Bayne, T. (2022). Theories of consciousness. Nature Reviews Neuroscience, 23, 439-452. https://doi.org/10.1038/s41583-022-00587-4.
24. 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.