Iterated Insights

Iterative Updating, Diachronic Unity, and the Progressive Evolution of Experience

Jared Edward Reser, Ph.D.

Theory article

Abstract

Most neurocognitive theories characterize the conditions under which a representation becomes conscious, but provide less detail about the rule by which one conscious state becomes the next. State-Spanning Workspace Theory of Consciousness proposes that the conscious stream is a globally available, recurrently grounded, self-conditioning trajectory of partially overlapping representational states. State-spanning coactivity (SSC) refers to content that remains causally available across consecutive states. Incremental change in state-spanning coactivity (icSSC) refers to its selective turnover as some representations persist, others are removed, and new representations are admitted. This iterative updating gives each state an inherited context and makes it both an interpretation of the preceding state and a constraint on its successor. Multiassociative search selects additions in relation to the whole retained configuration, while progressive imagery modification links abstract workspace contents to modality-specific, interoceptive, and motor maps. The theory is offered primarily as an account of stream consciousness: the diachronic unity, temporal thickness, contextual inheritance, and progressive evolution of conscious experience. It does not identify SSC with phenomenal existence. Basal phenomenality, conscious access, stream consciousness, reflective consciousness, and self-consciousness are treated as distinguishable levels that require partially different mechanisms. The account complements global neuronal workspace theory by adding a transition law, recurrent processing theory by adding selective persistence across recurrent cycles, higher-order approaches by specifying a temporal substrate for metarepresentation, and integrated information theory by emphasizing inter-state causal organization. Distinctive predictions concern causal inheritance, joint-context effects, thought boundaries, self-anchoring, adaptive turnover, and transitions in level of consciousness.

Keywords: consciousness; diachronic unity; global workspace; iterative updating; phenomenal continuity; state-spanning coactivity; stream consciousness; working memory

1. Introduction: consciousness needs a transition law

Consciousness is usually investigated one content at a time. A stimulus is presented, a report is collected, and neural activity associated with conscious access is contrasted with activity associated with nonconscious processing. This strategy has generated a substantial body of evidence, but it favors an instantaneous conception of experience. Ordinary consciousness is not encountered as an inventory of isolated contents. It is encountered as a stream in which a perception inherits a mood, a thought carries forward a question, a remembered event changes the meaning of what is currently seen, and an enduring sense of self accompanies continual change. A theory that explains only why a representation is conscious at a single moment leaves open the question of how conscious moments become one temporally organized experience.

This omission is increasingly visible within consciousness science. Global workspace theories explain how selected information becomes widely available. Recurrent processing theories emphasize feedback interactions that support perceptual organization. Higher-order theories identify metarepresentation as a condition of conscious awareness. Integrated information theory attempts to identify the intrinsic causal structure corresponding to phenomenal existence. These approaches differ substantially, yet most of their central explanatory units are states, episodes of ignition, recurrent exchanges, or cause-effect structures considered at a particular temporal grain (Seth & Bayne, 2022). Time-consciousness researchers have therefore argued that dominant theories attend more readily to brief functional moments than to the extended, dynamic, and continuous character of the experienced present (Kent & Wittmann, 2021; Singhal & Srinivasan, 2021).

The proposal developed here begins with a different explanatory target. It asks how one conscious state becomes the next without the stream dissolving into unrelated snapshots. The central claim is that a conscious moment is not an isolated neural state. It is a selectively retained and revised continuation of the immediately preceding conscious state. A subset of represented content remains causally available, newly selected content enters, other content loses priority, and the resulting configuration becomes the context from which the next transition is generated. The transition itself is therefore part of the mechanism of consciousness rather than a neutral interval between conscious events.

This framework is called State-Spanning Workspace Theory of Consciousness (SSWTC). It builds on the concepts of state-spanning coactivity (SSC), incremental change in state-spanning coactivity (icSSC), iterative updating, multiassociative search, and progressive imagery modification (Reser, 2016, 2024). Its strongest claim is not that SSC explains why phenomenal experience exists. Its strongest claim is that SSC and icSSC explain the diachronic unity, temporal organization, contextual inheritance, and progressive evolution of conscious experience. In this sense, the theory addresses stream consciousness. It specifies a transition law for consciousness and identifies testable neural and computational properties that distinguish a conscious stream from a mere succession of states.

2. The explanandum: diachronic unity

Several forms of unity are commonly grouped together under the heading of consciousness. Synchronic unity concerns why color, shape, sound, affect, and bodily sensation are experienced together at a given time. Diachronic unity concerns why the current experience belongs to the same stream as the experience immediately before it. Subject unity concerns why diverse contents are experienced from one perspective. Narrative unity concerns how experiences accumulate into a coherent thought, episode, or autobiographical history. These problems interact, but they are not identical. State-Spanning Workspace Theory is directed most specifically at diachronic unity and at the way diachronic unity supports subject and narrative unity.

Diachronic unity cannot be reduced to temporal succession. Frames in a film are successive, but succession alone does not explain why a system experiences one frame as the continuation of another. Nor is similarity sufficient. Two independently generated states could resemble one another without sharing a causal history. The relevant relation is causal state-spanning continuity: some content in the earlier state remains available in the later state and contributes to the selection, interpretation, or construction of what appears next. A state belongs to a conscious stream because it inherits part of its representational organization from its predecessors and actively uses that inheritance.

This proposal gives the experienced present temporal thickness. The present is not treated as a durationless point. It contains newly arriving information, still-active information from the immediate past, slower traces that remain available for reinstatement, and anticipations of likely next states. William James described experience as a stream and used overlapping units to explain how one phase shades into another (James, 1890). Time-consciousness research similarly distinguishes brief functional events from the longer experienced moment, which may integrate multiple events over seconds (Kent & Wittmann, 2021). State-Spanning Workspace Theory contributes a candidate neural and computational mechanism for that integration: staggered persistence, selective turnover, recurrent grounding, and context-dependent updating.

The theory does not require the conscious stream to be perfectly smooth. Attention can be captured, a train of thought can be interrupted, and a new task can trigger extensive replacement of focal contents. The claim is instead that normal waking experience contains a graded pattern of overlap across several timescales. Local continuity can be weak or strong, and thought boundaries can be abrupt or gradual. Even when the focus of attention is largely reset, slower short-term traces, bodily variables, goals, and autobiographical structures may preserve continuity at broader scales.

3. The core proposal: consciousness as a trajectory

Let W_t denote the globally available representational organization at time t. W_t should not be understood as a literal box in one cortical location. It is a functional coalition whose selected contents can influence multiple specialist systems, including sensory, interoceptive, mnemonic, evaluative, language, and motor processes. A minimal update can be expressed as follows:

W(t+1) = G[ K(t) ⊙ W(t), A(t+1), I(t) ]

K(t) is a retention gate that preserves selected components of W_t, A(t+1) is newly admitted content, I(t) represents inhibition or removal, and G is a binding and normalization operation that produces a coherent, capacity-limited next state. The notation is intentionally schematic. Its purpose is to express the transition law: the next conscious configuration is generated from a retained portion of the current configuration plus contextually selected change. Complete replacement is possible, but it is a special case rather than the default.

The primary explanatory object is therefore not an individual state but a trajectory:

Γ = { W(1), W(2), W(3), … , W(T) }

A conscious episode is a trajectory in which successive states share causally effective content, novel information differentiates each state from its predecessors, and the changing configuration remains broadly available to the systems that guide report, reasoning, memory formation, imagery, and action. The trajectory is self-conditioning because each workspace state changes the conditions under which the next workspace state will be selected. Every state is both an interpretation of the preceding state and a set of constraints on the state that follows.

Four properties are central. First, global availability allows selected content to affect multiple specialist systems. Second, recurrent grounding links high-level workspace contents to sensory, interoceptive, and motor representations rather than leaving them as disembodied symbols. Third, causal inheritance ensures that retained content helps determine subsequent content. Fourth, differentiationintroduces enough change for the stream to develop rather than remain frozen. Conscious flow requires continuity and difference together. Continuity without change is stasis; change without continuity is fragmentation.

Figure 1. The state-spanning workspace trajectory. Successive workspace states partially overlap because some representations remain causally available while others are removed and added. Recurrent grounding, global availability, and multiassociative search operate throughout the trajectory. The conscious stream is identified with the organized trajectory rather than with any isolated snapshot.

4. State-spanning coactivity and iterative updating

4.1 State-spanning coactivity as continuous causal availability

State-spanning coactivity occurs when two or more represented contents remain jointly available across consecutive processing states. In the original formulation, SSC was described in terms of sustained firing among cortical assemblies, while icSSC described the gradual change in group membership as some assemblies remained active, others deactivated, and new assemblies became active (Reser, 2016). That proposal captures a biologically plausible pattern of staggered persistence, but the theory should not be tied to uninterrupted firing by exactly the same neurons.

Working-memory representations can be maintained through several mechanisms. These include persistent population activity, transient bursts, dynamic population codes, short-term changes in synaptic efficacy, and states that become decodable again when the network is perturbed or cued (Mongillo et al., 2008; Rose et al., 2016; Stokes, 2015; Wolff et al., 2017). Distributed working-memory representations can also depend on reciprocal interactions between cortical areas rather than on a single local store (Voitov & Mrsic-Flogel, 2022). For this reason, SSC is best operationalized as the continuous causal availability of represented information across states. The neuronal implementation may be stable, dynamic, intermittent, or partly activity-silent, provided the content remains available to influence the trajectory.

This representational definition avoids a false choice between persistent and dynamic coding. A representation may change its neural format while preserving its functional role. The relevant question is whether an experimentally identifiable content, relation, goal, or self-variable remains causally available and continues to constrain processing. SSC can therefore be measured at multiple levels, including individual neurons, population subspaces, inter-areal communication patterns, recurrent loops, and behaviorally inferred latent states.

4.2 Iterative updating makes the workspace self-conditioning

Iterative updating is the selective modification of workspace contents as processing proceeds. Some representations are retained, some are suppressed or allowed to decay, and some are added. This pattern differs from complete updating, in which one configuration is discarded before the next is installed. It also differs from passive accumulation, in which new contents are added without active replacement. Iterative updating is a controlled turnover process that allows a limited-capacity workspace to preserve a task-relevant core while revising its context.

The retained core can include objects, relations, goals, emotional valuations, bodily variables, rules, imagined events, and intermediate results. Its composition is not determined solely by recency. A recently admitted item may be removed quickly if it proves uninformative, while an older item may remain because it specifies the topic, goal, agent, location, or unresolved constraint. Persistence is therefore selective and functional. The content that lasts longest often supplies the through line of the episode.

The rate of iterative updating should vary with task demands. Fast turnover increases sensitivity to new input and facilitates rapid reorientation. Slow turnover protects a structured context, supports multistep reasoning, and allows intermediate products to accumulate. Neither extreme is optimal in all circumstances. Excessive turnover produces distractibility and fragmentation; excessive persistence produces perseveration, proactive interference, and entrapment in stale attractor states. The relevant capacity is not maximal persistence but adaptive control over when to maintain, elaborate, branch, compress, or reset.

4.3 Multiassociative search selects context-sensitive additions

The workspace does more than preserve context. Its coactive contents jointly search associative memory for the next addition. In multiassociative search, the current configuration acts as a composite query. A candidate may be weakly associated with each item considered separately but strongly supported by their conjunction, temporal order, and relational structure. The next state is therefore not selected by a simple chain in which one idea activates one successor. It is selected by the whole configuration of retained contents, current sensory input, bodily state, goals, and recently potentiated traces.

This mechanism explains why the same incoming representation can enter different conscious experiences. A word, face, sound, or bodily sensation takes on a different meaning when it is integrated with a different retained context. New content is not appended to an otherwise neutral stage. It is interpreted inside a contextualized configuration, and its entry redistributes the relevance of what is already present. The configuration that results then becomes a revised search query. Conscious succession becomes a sequence of context-dependent completion operations.

Associative proposal must be distinguished from verification. The candidate receiving the strongest support from prior learning may be plausible without being true. Logical, causal, perceptual, and instrumental evaluators should therefore test candidate additions before they guide consequential action or become consolidated. This distinction allows iterative thought to remain generative without equating associative fluency with accuracy.

4.4 Progressive imagery modification recurrently grounds the stream

The workspace is not proposed as a detached symbolic register. High-level representations repeatedly interact with modality-specific systems. Association areas can preserve abstract constraints across several cycles, while sensory, interoceptive, language, and motor systems construct more rapidly changing maps that instantiate those constraints in a modality-specific format. Salient features from each constructed map can return to the workspace and alter the next iteration. This reciprocal process has been termed progressive imagery modification (Reser, 2016, 2024).

Progressive imagery modification provides a mechanistic account of how an imagined scene, inner sentence, bodily expectation, or action plan can change in a continuous yet cumulative way. A partially stable high-level configuration guides the construction of a sensory or motor state. The constructed state introduces features that were not explicitly specified at the higher level. Some of those features are then selected into the next workspace configuration, which guides another round of construction. The interaction is informative because the participating systems encode different kinds of structure.

This arrangement also addresses the homunculus problem. No internal observer needs to inspect a sensory screen. The workspace and modality-specific systems causally transform one another. The resulting experience is identified with the organized recurrent activity itself, not with an additional subject who watches it. The theory remains neutral about whether recurrent grounding is sufficient for basal phenomenality, but it treats such grounding as necessary for the rich, context-sensitive contents of ordinary stream consciousness.

5. What the theory explains

5.1 Diachronic unity and the temporal thickness of the present

The first explanatory contribution is a process account of diachronic unity. Consecutive states belong to one stream because they share a causally active representational core. The earlier state is not merely remembered by the later state after the fact. Some of its contents remain present in the process that constructs the later state. This produces a temporally thick present in which immediate past, current input, and near-future expectation coexist in an ordered relationship.

The amount of overlap need not be constant. A stable perception may preserve a large relational core while changing only a few features. A topic shift may preserve a goal and self-model while replacing most object-level contents. A sudden threat may replace nearly everything in focal attention while leaving slower affective and autobiographical variables intact. Diachronic unity is therefore nested across timescales rather than governed by one universal integration window.

5.2 Contextual inheritance and the meaning of experience

The second contribution is an account of contextual inheritance. Retained content determines how new content is interpreted. The same sensory event can feel expected, surprising, threatening, humorous, or irrelevant depending on the configuration into which it enters. The theory locates this contextual modulation inside the transition law of the conscious stream. Context is not a static background. It is a partially inherited set of active constraints that is continuously revised by what it helps to interpret.

Contextual inheritance also explains why later information can reframe what has just been experienced. Because representations from the recent past remain available, a new item can alter their relations, significance, or source attribution. An ambiguous sound can become speech, an expression can become sarcastic, and a bodily sensation can become alarming when a new interpretation enters the workspace. The stream does not simply add contents. It reorganizes them.

5.3 The continuity of the subject

The subject of experience need not be represented by one immutable symbol. It can be carried by a slowly changing subset of the trajectory. Let S_t denote self-related variables within W_t, including bodily location, interoceptive condition, agency, current goals, autobiographical identity, and social role. If components of S_t persist longer than many object-level contents, they can anchor the stream while the experienced world changes around them.

S(t) ≈ S(t+1), while W(t) ≠ W(t+1)

This formulation distinguishes the minimal self from the narrative self. The minimal self is grounded in bodily perspective, interoception, proprioception, affect, and agency. The narrative self depends more heavily on autobiographical and conceptual representations. The minimal self may provide a nearly continuous embodied orientation, while narrative self-representations move in and out of focal awareness and remain available in slower memory traces. Subject unity is strengthened when these self-specifying variables remain causally integrated with the changing contents of the workspace.

5.4 Cumulative thought and reflective experience

Iterative updating allows conscious cognition to accumulate. Intermediate results can remain active long enough to shape the next inference, be suspended while a subproblem is addressed, and later be reinstated or merged with another line of thought. This produces a chain of interdependent states rather than a list of independent associations. Reasoning, planning, narrative comprehension, and imaginative simulation can then proceed by progressive modification of a partially conserved structure.

This cumulative organization also clarifies the relation between fast and slow cognition. Slow, reflective thought need not rely on a wholly separate processing architecture. It may emerge when ordinarily rapid associative processes are constrained by a more persistent, relationally organized workspace. Automatic proposals are generated, retained context evaluates them, and useful products become inputs to further cycles. The difference lies partly in the duration, stability, and recursive use of intermediate states.

6. Relation to major theories of consciousness

6.1 Global neuronal workspace theory

Global neuronal workspace theory proposes that recurrent amplification and network ignition make selected information globally accessible to otherwise specialized processors (Dehaene & Naccache, 2001; Mashour et al., 2020). State-Spanning Workspace Theory accepts global availability as an important property of conscious access, but asks a different question. Global workspace theory asks what gains access. State-Spanning Workspace Theory asks what remains, what leaves, what enters, and how the retained configuration determines the next access event.

The proposed addition is a transition law. A workspace can broadcast a succession of representations without specifying how one broadcast inherits organization from the previous broadcast. In the state-spanning account, each globally available state alters the competitive landscape for the next state. Broadcasting becomes self-conditioning. This extension is compatible with a global workspace but is not entailed by global availability alone.

The 2025 adversarial collaboration comparing global neuronal workspace theory and integrated information theory found evidence consistent with some predictions of both theories while challenging central predictions of both (Cogitate Consortium et al., 2025). That result underscores the value of hypotheses that cut across their principal localization and timing disputes. Cross-state causal inheritance, relational overlap, and adaptive turnover can be tested independently of whether conscious content is centered in posterior cortex, frontoparietal networks, or a more distributed recurrent coalition.

6.2 Recurrent processing theory

Recurrent processing theory distinguishes feedforward feature extraction from recurrent interactions that support perceptual organization and conscious seeing (Lamme, 2020; Lamme & Roelfsema, 2000). State-Spanning Workspace Theory is compatible with this emphasis on feedback but adds persistence and selective turnover across recurrent cycles. Recurrence can occur within a single perceptual episode without explaining why the next episode inherits some of the same high-level constraints. SSC and icSSC specify how recurrent processing is linked across a sequence of states.

The theories also differ in scope. Recurrent processing theory is primarily directed at the transition from unconscious to conscious perception. State-Spanning Workspace Theory is directed at the temporal organization of already conscious contents across perception, imagery, memory, thought, and action. Recurrent processing is treated as a grounding mechanism within the broader state-spanning trajectory.

6.3 Integrated information theory

Integrated information theory attempts to explain both why experience exists and why a particular experience has its specific phenomenal structure by identifying an intrinsic cause-effect structure (Albantakis et al., 2023). State-Spanning Workspace Theory does not presently make an equivalent claim about phenomenal existence. Its explanatory target is narrower: how conscious contents remain unified through change.

The two approaches may nevertheless be complementary. Integrated information is typically evaluated for a system in a state at a chosen temporal grain. State-Spanning Workspace Theory emphasizes relations between states and the causal organization of a trajectory. A future synthesis could ask whether the intrinsic structure associated with a conscious state must itself be embedded in a temporally extended pattern of causal inheritance. The relevant quantity would not be overlap alone, but the degree to which preserved distinctions and relations continue to make a difference to subsequent states.

6.4 Higher-order theories and metacognition

Higher-order theories hold that a mental state becomes conscious when the system represents itself as being in that state, or when a first-order state is situated within an appropriate higher-order state space (Brown et al., 2019; Fleming, 2020). State-spanning persistence can support higher-order representation because a first-order state must remain available long enough to be classified, evaluated, or represented by another process. Yet causal inheritance is not itself metarepresentation.

A later state may be caused by an earlier state without representing the earlier state as a mental state. Reflective consciousness therefore requires additional information. The workspace should encode source, confidence, temporal index, ownership, evidential basis, and consistency with goals or beliefs. A perceived object, an imagined object, and a remembered object may have similar first-order content while differing in source. State-Spanning Workspace Theory supplies the temporal substrate for this distinction; explicit metacognitive tagging makes the state genuinely higher-order.

6.5 Time-consciousness and temporo-spatial approaches

Time-consciousness theories emphasize that experience is extended, ordered, and dynamically nested across several timescales (Kent & Wittmann, 2021; Singhal & Srinivasan, 2021). State-Spanning Workspace Theory agrees with this starting point and offers an item-level and relation-level mechanism. The theory is extensional because represented contents physically span intervals. It is retentional because earlier content remains represented within later processing. It can also accommodate brief functional moments because the state-spanning trajectory may be implemented through discrete recurrent updates at a finer grain.

The distinctive contribution is not the general assertion that consciousness is temporal. It is the proposal that temporal continuity is generated by selective representational turnover within a globally available, recurrently grounded workspace. This yields concrete variables: what persists, how long it persists, which relations are preserved, how retained content changes the next-state distribution, and where abrupt decreases in overlap mark transitions between thoughts.

7. A layered account of consciousness

The term consciousness covers phenomena that need not depend on the same mechanism. A layered account prevents the state-spanning proposal from excluding simple or fleeting experiences merely because they lack prolonged working-memory continuity. It also prevents working-memory persistence from being treated as sufficient for phenomenality.

Level

Defining property

Proposed mechanism

Basal phenomenality

Modality-specific or bodily experience is present.

Recurrent sensory and interoceptive organization; not explained by SSC alone.

Conscious access

Selected content is flexibly available for report, reasoning, memory, and action.

Global availability or broadcast across specialist systems.

Stream consciousness

Successive contents are experienced as one temporally continuous, contextually inherited episode.

SSC, icSSC, iterative updating, and recurrent grounding.

Reflective consciousness

The system represents its own states, sources, confidence, and processing status.

Metarepresentational tagging of state-spanning contents.

Self-consciousness

Experience is organized around an embodied and autobiographical subject.

Persistent self-model variables integrated across several timescales.

Table 1. Levels of consciousness distinguished within State-Spanning Workspace Theory. The theory contributes most directly to stream consciousness, while treating basal phenomenality, conscious access, reflection, and self-consciousness as related but separable explanatory targets.

On this layered view, SSC and icSSC may be unnecessary for the bare existence of every fleeting sensation, but they are central to the richer form of consciousness in which experience is temporally extended, context-sensitive, reportable, reflectively accessible, and organized around an enduring subject. The model can therefore accommodate consciousness in infants and nonhuman animals without requiring human-like working-memory capacity, while predicting that differences in persistence, globality, self-modeling, and metacognition will alter the structure of their conscious streams.

8. Formalization and measurement

8.1 A structured workspace state

A useful formalization should represent more than an unordered set of active items. Conscious meaning depends on roles, relations, temporal order, confidence, and source. The workspace state can be expressed as:

W(t) = [ C(t), R(t), T(t), S(t), G(t) ]

C_t contains active contents; R_t contains relations among those contents; T_t contains slower short-term traces; S_t contains bodily and self-related variables; and G_t contains goals, values, and control settings. New candidates are retrieved through a joint query:

q(t) = f[ C(t), R(t), T(t), S(t), G(t), x(t) ]

where x_t is current sensory and interoceptive input. The next state is produced by retaining selected components, retrieving candidate additions, evaluating them, and enforcing capacity and coherence constraints. A process can be first-order Markov in this expanded state while displaying long behavioral memory because the state itself carries temporally extended information.

8.2 Weighted representational continuity

A simple item count is not an adequate measure of continuity. Preserving the current goal or embodied perspective may contribute more to experienced unity than preserving several incidental visual features. Relations may also matter more than membership. The same contents can form different experiences when their roles or order change. A weighted continuity score can therefore combine item, relational, and self-goal overlap:

O(t) = α O_items(t) + β O_relations(t) + γ O_self-goal(t)

The coefficients can be estimated from behavioral relevance, causal interventions, or predictive value. O_t should be evaluated across multiple lags to produce a continuity curve. The lag at which weighted overlap falls to half its initial value defines a working-memory or workspace half-life. This half-life is expected to vary across tasks, contents, arousal levels, and individuals rather than serving as one fixed trait.

8.3 Multiassociative synergy

The theory makes a stronger claim than ordinary spreading activation. The joint configuration should predict the next conscious addition better than the sum of its components considered independently. This can be measured by comparing a model that uses the complete structured state with an additive model that combines separate item effects:

Δ_synergy = L_additive – L_joint

L is held-out prediction loss. A positive synergy score indicates that relational or conjunctive information in the whole configuration contributes to next-state selection. This measure can be applied to behavioral reports, decoded neural states, memory retrievals, or transitions in an artificial agent.

8.4 Causal inheritance

Representational overlap becomes theoretically important only when retained content makes a causal difference. Let W(t+1)^(-R) denote the next state produced after experimentally disrupting a retained representation or relation R. A causal inheritance index can be defined as the expected distance between the unperturbed and perturbed next-state distributions:

CI(t,R) = E[ d( W(t+1), W(t+1)^(-R) ) ]

This index distinguishes genuine state-spanning organization from passive similarity. If removing a retained element does not change subsequent processing, its persistence may be epiphenomenal. Causal inheritance can be tested with transcranial magnetic stimulation, intracranial stimulation, optogenetic perturbation, targeted distraction, or counterfactual intervention in computational models.

8.5 Thought boundaries and adaptive turnover

A thought boundary can be operationalized as a sharp decrease in weighted overlap accompanied by a reorganization of the relational structure or goal state. No universal numerical threshold is assumed. Boundaries are graded and task-dependent. A topic shift may occur when item overlap falls while the self-model remains stable; a more profound discontinuity may involve simultaneous disruption of contents, goals, and self-specifying variables.

The rate of turnover should also be related nonlinearly to performance. Tasks requiring sustained relational integration should benefit from lower turnover until irrelevant contents begin to interfere. Tasks requiring rapid environmental tracking should benefit from higher turnover. The predicted optimum is therefore an adaptive region rather than a fixed percentage. Neuromodulatory and control systems may alter retention gates in response to novelty, uncertainty, reward, error, fatigue, and threat.

9. Distinctive predictions and experimental tests

A useful theory must predict findings that do not follow from working-memory maintenance, recurrence, or global broadcasting alone. The following predictions are intended to make State-Spanning Workspace Theory contrastive and falsifiable.

Prediction

Operational test

Expected result

Causal inheritance

Hold current sensory input constant while perturbing one retained representation or relation.

The identity, interpretation, or probability of the next conscious content changes selectively.

Joint-context prediction

Compare whole-state models with additive models based on individual active contents.

The structured joint configuration predicts the next decoded or reported content better than additive cues.

Continuity-disruption prediction

Disrupt cross-state carryover without preventing initial stimulus detection.

Temporal coherence, retrospective interpretation, confidence, and narrative integration decline more than immediate detection.

Thought-boundary prediction

Collect continuous topic-change reports and estimate multivariate overlap through time.

Reported boundaries coincide with sharp drops in weighted item and relational overlap.

Self-anchor prediction

Decode bodily, agency, goal, and autobiographical variables across changing contents.

Self-specifying variables have longer effective time constants and mediate experienced unity.

Level-of-consciousness prediction

Compare wakefulness, sleep, anesthesia, and disorders of consciousness using trajectory measures.

Conscious states show a characteristic combination of causal overlap, differentiation, recurrence, and global availability.

Table 2. Core predictions. Each prediction concerns the organization of transitions rather than the presence of a representation in one state alone.

A direct behavioral-neurophysiological experiment could manipulate required overlap between successive task states. Participants would maintain a structured scene or problem while transitions preserve 0, 25, 50, 75, or 100 percent of its task-relevant features. Sensory load, response requirements, elapsed time, and reward could be matched. Multivariate EEG, MEG, intracranial recordings, or fMRI could estimate item identity, relational structure, representational similarity, and state-transition boundaries. Performance should be best at a task-dependent intermediate level of turnover, with distinct errors under excessive replacement and excessive persistence.

The joint-context prediction is particularly discriminating. Suppose several retained clues jointly determine the next association. A model trained on the entire configuration should outperform models trained on single clues or their additive combination. Neural evidence for multiassociative search would be stronger if perturbing one retained clue changed the next-state distribution in a manner dependent on the other retained clues. The predicted effect is an interaction, not merely a reduction in overall activation.

Thought boundaries can be studied during spontaneous speech, movie viewing, problem solving, or internally directed thought. Participants can indicate perceived topic shifts, while representational analyses estimate overlap in contents, relations, goals, and self-variables. The theory predicts that within-thought transitions preserve a weighted relational core, whereas between-thought transitions show a more extensive reorganization. Thread resumption after interruption should return the neural state closer to the saved endpoint of the suspended thought than to states experienced during the interruption.

Level-of-consciousness studies should measure trajectories rather than static connectivity alone. Recent work indicates that conscious wakefulness is associated with a dynamic balance of integration and segregation, while anesthesia shifts this balance and alters network transitions (Jang et al., 2024). State-Spanning Workspace Theory predicts that conscious conditions will also preserve causally effective content across differentiated states. Static activation, simple autocorrelation, or persistent storage without global and recurrent influence should not be sufficient.

10. Scope, boundary conditions, and potential objections

10.1 Persistence is not sufficient for consciousness

Many nonconscious systems retain information. A cache, register, recurrent controller, or homeostatic process can preserve state across time. Persistence becomes relevant to stream consciousness only when the retained information participates in a globally available, recurrently grounded, differentiated trajectory. Even that conjunction may explain the organization of experience without explaining why the organization is phenomenal. State-Spanning Workspace Theory therefore rejects the inference that any system with memory or recurrence is conscious.

10.2 The theory is not restricted to prefrontal content

The workspace is a functional organization rather than a claim that all phenomenal content is stored in a small prefrontal buffer. Modality-specific regions may carry detailed perceptual structure, while association and control systems preserve abstract constraints, relations, goals, and metacognitive variables. The phenomenal field may therefore be richer than the focal set of items explicitly maintained for reasoning. Recurrent exchange between content-specific and supramodal networks is more consistent with the theory than a one-way hierarchy in which a frontal observer reads posterior representations.

10.3 Overlap is not enough without causal and relational structure

Two states can overlap because of slow physiological drift, repeated stimulation, or redundant coding. Such overlap does not establish diachronic unity. The theory requires content-sensitive causal inheritance. Preserved relations, role assignments, temporal order, and self-goal variables may be more important than raw similarity. Experimental tests should therefore combine representational similarity with intervention, decoding, and next-state prediction.

10.4 Basal phenomenality remains an open explanatory target

The theory does not yet explain why a recurrently organized sensory state feels like color, pain, sound, or affect. It can explain why a quality remains present, is contextualized by other contents, changes progressively, and belongs to a temporally extended episode. A complete theory of phenomenal existence would need an additional bridge principle linking physical or computational organization to the presence and quality of experience. Integrated information theory provides one explicit attempt at such a bridge; other approaches locate the relevant organization in recurrent sensory processing, predictive regulation, embodiment, or biological properties.

This limitation is a matter of scope rather than a reason to discard the transition problem. A theory of combustion can explain flame propagation without being a theory of every molecular property of fuel. Likewise, a theory of stream consciousness can identify the law governing the continuation of experience while leaving the ultimate basis of phenomenality unresolved. The explanatory target should be stated clearly so that success on diachronic unity is not mistaken for a complete solution to consciousness.

10.5 Toward universality

A theory intended to assess nonbiological systems must define its constructs independently of human anatomical labels. Kanai and Fujisawa (2024) argue that a universal theory should determine whether any fully described dynamical system is conscious by reference to intrinsic, operationally specified properties. State-Spanning Workspace Theory is not yet universal in that strong sense. Terms such as representation, global availability, recurrent grounding, and self-model require formal criteria that can be applied across substrates.

The trajectory formulation offers a path forward. A candidate system can be characterized by its transition dynamics, causal bottlenecks, cross-module accessibility, intervention structure, retained variables, and self-modeling operations. The closer these constructs come to measurable intrinsic properties, the less the theory will depend on anthropomorphic interpretation. Universality should be treated as a development goal rather than assumed at the outset.

11. Implications for machine consciousness

The theory supplies a stricter standard for machine consciousness than the presence of a context window, recurrent state, or verbal self-report. A candidate artificial conscious stream would need ongoing internal states that partially overlap, influence one another causally, and remain available across multiple specialist systems. The states would need recurrent grounding in perceptual, interoceptive or homeostatic, and action-related models. They would also need sufficient differentiation to avoid becoming a static attractor.

Reflective machine consciousness would require more. The system would have to represent the source, confidence, ownership, and temporal status of its own contents. A self-model would have to persist across changing tasks and distinguish externally caused input from internally generated imagery or inference. These variables should not merely be included in a textual explanation after the fact. They should participate causally in attention, learning, planning, and action selection.

Interpretability is also relevant. A displayed workspace is meaningful only if it is a causal bottleneck rather than a decorative summary. Counterfactual edits to the workspace should produce predictable changes in subsequent processing and behavior. Hidden bypass channels should be minimized or measured. A machine would provide stronger evidence for a state-spanning workspace if its logged contents mediated the effects of perception, memory, evaluation, and planning on action.

These criteria do not settle whether such a machine would possess basal phenomenality. They do, however, distinguish an artificial stream of thought from a sequence of outputs generated without an enduring, recurrently grounded, self-conditioning internal trajectory. State-Spanning Workspace Theory therefore offers an architecture-level research program while remaining explicit about the limits of functional evidence.

12. Conclusion

State-Spanning Workspace Theory proposes that consciousness is not merely a globally integrated state. It is a globally available state that inherits content and causal organization from its predecessors while continuously revising itself. State-spanning coactivity preserves a representational core; incremental change in state-spanning coactivity transforms that core; iterative updating regulates retention and replacement; multiassociative search selects context-sensitive additions; and progressive imagery modification recurrently grounds the evolving configuration.

The theory is best understood as an account of the diachronic unity, temporal organization, and progressive evolution of conscious experience. It explains why the present is temporally thick, why new contents inherit context, why thoughts can accumulate, why a changing field can remain organized around one subject, and why abrupt reductions in overlap feel like boundaries between thoughts. Its central claim can be stated compactly: the conscious stream is a globally available, recurrently grounded, self-conditioning trajectory of partially overlapping representational states.

This claim is narrower than a complete theory of phenomenal existence, but it is stronger than a metaphor of flow. It specifies variables, mechanisms, formal measures, interventions, and falsifying predictions. The next step is to test whether causal state-spanning continuity explains aspects of consciousness that cannot be explained by recurrence, access, integration, or metarepresentation considered one state at a time.

References

Albantakis, L., Barbosa, L., Findlay, G., Grasso, M., Haun, A. M., Marshall, W., Mayner, W. G. P., Zaeemzadeh, A., Boly, M., Juel, B. E., et al. (2023). Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms. PLOS Computational Biology, 19(10), e1011465. https://doi.org/10.1371/journal.pcbi.1011465

Baars, B. J. (1988). A cognitive theory of consciousness. Cambridge University Press.

Baddeley, A. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417-423. https://doi.org/10.1016/S1364-6613(00)01538-2

Block, N. (1995). On a confusion about a function of consciousness. Behavioral and Brain Sciences, 18(2), 227-247. https://doi.org/10.1017/S0140525X00038188

Brown, R., Lau, H., & LeDoux, J. E. (2019). Understanding the higher-order approach to consciousness. Trends in Cognitive Sciences, 23(9), 754-768. https://doi.org/10.1016/j.tics.2019.06.009

Christophel, T. B., Klink, P. C., Spitzer, B., Roelfsema, P. R., & Haynes, J.-D. (2017). The distributed nature of working memory. Trends in Cognitive Sciences, 21(2), 111-124. https://doi.org/10.1016/j.tics.2016.12.007

Cogitate Consortium, Ferrante, O., Gorska-Klimowska, U., Henin, S., Hirschhorn, R., Khalaf, A., Lepauvre, A., Liu, L., Richter, D., Vidal, Y., et al. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature, 642, 133-142. https://doi.org/10.1038/s41586-025-08888-1

Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87-114. https://doi.org/10.1017/S0140525X01003922

Damasio, A. R. (1989). Time-locked multiregional retroactivation: A systems-level proposal for the neural substrates of recall and recognition. Cognition, 33(1-2), 25-62. https://doi.org/10.1016/0010-0277(89)90005-X

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

D’Esposito, M., & Postle, B. R. (2015). The cognitive neuroscience of working memory. Annual Review of Psychology, 66, 115-142. https://doi.org/10.1146/annurev-psych-010814-015031

Fleming, S. M. (2020). Awareness as inference in a higher-order state space. Neuroscience of Consciousness, 2020(1), niz020. https://doi.org/10.1093/nc/niz020

James, W. (1890). The principles of psychology. Henry Holt.

Jang, H., Mashour, G. A., Hudetz, A. G., & Huang, Z. (2024). Measuring the dynamic balance of integration and segregation underlying consciousness, anesthesia, and sleep in humans. Nature Communications, 15, 9164. https://doi.org/10.1038/s41467-024-53299-x

Kanai, R., & Fujisawa, I. (2024). Toward a universal theory of consciousness. Neuroscience of Consciousness, 2024(1), niae022. https://doi.org/10.1093/nc/niae022

Kent, L., & Wittmann, M. (2021). Time consciousness: The missing link in theories of consciousness. Neuroscience of Consciousness, 2021(2), niab011. https://doi.org/10.1093/nc/niab011

Lamme, V. A. F. (2020). Visual functions generating conscious seeing. Frontiers in Psychology, 11, 83. https://doi.org/10.3389/fpsyg.2020.00083

Lamme, V. A. F., & Roelfsema, P. R. (2000). The distinct modes of vision offered by feedforward and recurrent processing. Trends in Neurosciences, 23(11), 571-579. https://doi.org/10.1016/S0166-2236(00)01657-X

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

Mongillo, G., Barak, O., & Tsodyks, M. (2008). Synaptic theory of working memory. Science, 319(5869), 1543-1546. https://doi.org/10.1126/science.1150769

Reser, J. E. (2016). Incremental change in the set of coactive cortical assemblies enables mental continuity. Physiology & Behavior, 167, 222-237. https://doi.org/10.1016/j.physbeh.2016.09.019

Reser, J. E. (2024). A cognitive architecture for machine consciousness and artificial superintelligence: Thought is structured by the iterative updating of working memory. arXiv:2203.17255.

Rose, N. S., LaRocque, J. J., Riggall, A. C., Gosseries, O., Starrett, M. J., Meyering, E. E., & Postle, B. R. (2016). Reactivation of latent working memories with transcranial magnetic stimulation. Science, 354(6316), 1136-1139. https://doi.org/10.1126/science.aah7011

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

Singhal, I., & Srinivasan, N. (2021). Time and time again: A multi-scale hierarchical framework for time-consciousness and timing of cognition. Neuroscience of Consciousness, 2021(2), niab020. https://doi.org/10.1093/nc/niab020

Stokes, M. G. (2015). “Activity-silent” working memory in prefrontal cortex: A dynamic coding framework. Trends in Cognitive Sciences, 19(7), 394-405. https://doi.org/10.1016/j.tics.2015.05.004

Tononi, G., Boly, M., Massimini, M., & Koch, C. (2016). Integrated information theory: From consciousness to its physical substrate. Nature Reviews Neuroscience, 17, 450-461. https://doi.org/10.1038/nrn.2016.44

Voitov, I., & Mrsic-Flogel, T. D. (2022). Cortical feedback loops bind distributed representations of working memory. Nature, 608, 381-389. https://doi.org/10.1038/s41586-022-05014-3

Wolff, M. J., Jochim, J., Akyürek, E. G., & Stokes, M. G. (2017). Dynamic hidden states underlying working-memory-guided behavior. Nature Neuroscience, 20, 864-871. https://doi.org/10.1038/nn.4546

Author note

This manuscript develops and narrows the consciousness-related claims of the iterative updating and state-spanning coactivity framework into a testable theory of stream consciousness.