Jared Edward Reser, Ph.D.

Abstract

Progressive imagery modification is proposed as a recurrent mechanism through which mental imagery contributes to simulation, reasoning, planning, and the continuity of thought. The theory holds that working memory maintains a partially persistent set of higher-order associative representations that repeatedly constrains the construction of lower-order sensory and sensorimotor maps. These maps are not merely passive depictions of information already present in working memory. By integrating incomplete abstract constraints according to perceptual regularities acquired through experience, they can introduce spatial, temporal, causal, and compositional information that was previously implicit. Salient features extracted from each internally generated map return to the associative workspace, where some representations are maintained, others are removed, and new image-derived representations are added. The resulting state then constrains the next map. Consecutive imagery states are related because they share many of their determining parameters, yet they are progressively modified because each cycle can contribute new information. Later imagery may therefore depend causally on discoveries made during earlier imagery, permitting otherwise automatic operations to accumulate into multistep simulations, counterfactual branches, plans, and deductions. The framework extends earlier proposals concerning state-spanning coactivity, incremental change in state-spanning coactivity, polyassociativity, iterative updating, and multiassociative search. It also provides a functional interpretation of reciprocal signaling between persistent association-level representations and comparatively transient topographic mappings. The present article formalizes the progressive imagery modification cycle, distinguishes it from static recall and unconstrained imagery drift, identifies several measurable properties of the process, and advances predictions concerning its neural implementation. An artificial system built on the same principle would need not only to generate internal imagery but also to analyze its own generated maps and allow them to modify the internal state responsible for constructing the next map. Progressive imagery modification thus offers a candidate explanation for how a cognitive system can think by iteratively transforming and interrogating its own internally generated models.

Keywords: mental imagery, working memory, state-spanning coactivity, iterative updating, mental simulation, recurrent processing, multiassociative search, planning, consciousness, artificial intelligence

1. Introduction

Mental imagery allows a cognitive system to represent objects, events, actions, and relationships that are not currently presented to the senses. It contributes to recollection, anticipation, navigation, planning, language comprehension, creativity, and the rehearsal of possible behavior. Yet the fact that imagery occurs does not by itself explain how imagery participates in thought. A complete theory must account for how an internal image is constructed, how one image develops into another, how newly generated imagery can disclose information that was not explicit in the initial state, and how a sequence of such images can culminate in a prediction, decision, plan, or solution.

Many theories of mental imagery focus primarily on representational format. They ask whether imagery is depictive or propositional, whether imagined and perceived content recruit overlapping neural substrates, or how sensory information is retained in working memory. These questions are important, but they leave a temporal and computational problem unresolved. A mental image is seldom an isolated endpoint. During internally generated thought, imagery can transform continuously, preserving some properties while changing others, and each transformation can affect what the mind represents next.

The concept of progressive imagery modification was developed to address this problem. An early formulation described reciprocal transformations between a working-memory updating system and an imagery-generation system (Reser, 2013). The subsequent theory of incremental change in state-spanning coactivity proposed that sustained representations in association cortex constrain successive topographic mappings in sensory and motor systems. Because some higher-order representations remain active while others enter and leave the active set, consecutive mappings share subject matter while also introducing modifications. This reciprocal, recursively organized exchange was termed progressive imagery modification (Reser, 2016). 

Incremental change in state spanning cortical.pdf

The 2016 account placed progressive imagery modification within a broader model of state-spanning coactivity, or SSC, and incremental change in state-spanning coactivity, or icSSC. SSC refers to a group of cortical representations that remains coactive across successive brain states. icSSC refers to the gradual change in the membership of that group, as some representations persist, others deactivate, and new representations enter. This overlapping organization permits successive states to share content, become recursively interrelated, and exhibit progressive, algorithmic, thematic, and narrative properties (Reser, 2016). 

Incremental change in state spanning cortical.pdf

A later cognitive architecture recast these dynamics in terms of iterative updating and multiassociative search. Iterative updating occurs when the contents of working memory undergo partial replacement, so that some representations are added, others are removed, and still others remain active. The coactive contents jointly spread activation through long-term memory, selecting the most contextually appropriate representation to update the next state. Each state is therefore both the product of the preceding search and the starting point for the next search (Reser, 2022/2024). 

Thought iterative updating.pdf

The present article isolates progressive imagery modification from this broader architecture and develops it into a standalone theory of imagination and mental simulation. The earlier formulations supply the neural and cognitive foundation. The present extension specifies the processing cycle more explicitly, interprets imagery as an active computational operation, distinguishes continuity from progress, describes how imagery can uncover latent relational information, and derives a set of empirical and engineering implications.

2. Conceptual Foundations

2.1 State-spanning coactivity and the persistence of mental context

Any account of progressive imagery must explain why successive images remain related. If all active representations were replaced at once, each new image would be generated from an unrelated set of constraints. The result would be a succession of disconnected states rather than a coherent mental simulation.

SSC provides the required persistence. During a given interval, several representations remain simultaneously active across more than one brain state. These representations may encode the enduring subject, setting, goal, relation, or problem being considered. When membership in the active set changes incrementally, the surviving representations provide a frame of reference to which the newly activated representations can relate.

Suppose representations B, C, D, and E are active at one moment, and C, D, E, and F are active at the next. C, D, and E span both states and preserve the context within which F is interpreted. The second state is not merely later than the first. It is a revision of the first, constructed partly from the same active neural and psychological material.

This principle applies at multiple timescales. Sustained firing may maintain highly prioritized contents within the focus of attention over seconds, while short-term synaptic changes and cortical priming preserve a broader residue of recent activity over longer intervals. The former supports immediate continuity between imagery frames, while the latter permits suspended imagery threads to be resumed and earlier material to continue biasing the unfolding sequence.

2.2 Iterative updating

Iterative updating describes the transition rule governing these active representations. Instead of completely clearing working memory between processing steps, the system retains a subset of the previous state and combines it with one or more additions.

The result can be written schematically as:

W_t = \{B,C,D,E\}

W_{t+1} = \{C,D,E,F\}

W_{t+2} = \{C,E,F,G\}

Each state preserves information from the state before it, but none is identical to its predecessor. The process is iterative because the same operation is repeatedly applied to the product of the previous operation. It is recursive in the broader functional sense because the outputs of earlier cycles return as inputs to later cycles.

The proportion of content replaced at each step can vary. A low rate of updating preserves more contextual constraints and promotes tightly coupled, sustained processing. A high rate of updating introduces more novelty and responsiveness but may weaken continuity. The appropriate balance depends upon the task. Focused reasoning may require a relatively stable set of constraints, whereas exploratory or creative thought may benefit from more rapid turnover.

2.3 Multiassociative search

The next update is not selected independently of the retained contents. The active representations pool their excitatory and inhibitory effects, searching long-term associative memory for representations that best fit the present combination.

The 2016 article described this process as polyassociativity. The later formulation uses multiassociative search, emphasizing that all coactive and cospreading items contribute jointly to the selection of the next item. A representation that is only weakly associated with any individual item may nevertheless be strongly associated with their conjunction. This allows a novel combination of search constraints to converge on a contextually appropriate addition. 

Thought iterative updating.pdf

Multiassociative search and iterative updating perform complementary roles. Iterative updating preserves and modifies the active context, while multiassociative search determines which representation should be introduced into that context. Progressive imagery modification adds another stage: internally generated sensory and sensorimotor maps participate in producing the information from which the next update is selected.

2.4 Hierarchical sensory and associative representations

The cerebral cortex contains a hierarchy extending from comparatively concrete, modality-specific representations to increasingly abstract, invariant, and multimodal ones. Early sensory networks encode metric and topographic structure. Higher association networks encode objects, people, places, intentions, categories, rules, relationships, and other postcategorical constructs.

During perception, ascending sensory activity provides much of the driving input, while descending associative expectations modulate its interpretation. During imagination, the direction of influence can be partially reversed. Higher-order representations provide the primary specifications, and sensory systems use their learned organization to construct an internally generated map consistent with those specifications.

This does not imply that association cortex contains no imagery. Higher-order representations also embody structure derived from experience, but their organization is more abstract and less directly tied to a single sensory coordinate system. Progressive imagery modification concerns the recurrent interaction between these levels, rather than assigning imagery exclusively to one cortical region.

3. Defining Progressive Imagery Modification

Progressive imagery modification can be defined as follows:

Progressive imagery modification is a recurrent process in which a partially persistent set of higher-order associative representations repeatedly constrains the construction of lower-order sensory or sensorimotor maps, while information extracted from each generated map partially updates the associative state that will construct the next map.

Four properties are central to this definition. First, some higher-order constraints must persist across successive cycles. Second, the generated map must integrate those constraints into a structured configuration. Third, the map must be capable of introducing or exposing information not explicitly represented in the state that initiated it. Fourth, information extracted from the map must causally contribute to a subsequent state.

The process can therefore be summarized as:

\text{persistent associative state}
\rightarrow
\text{topographic construction}
\rightarrow
\text{feature extraction}
\rightarrow
\text{partial associative update}
\rightarrow
\text{revised topographic construction}

This cycle distinguishes progressive imagery modification from static image retrieval. Recalling a familiar image once may activate a sensory representation, but no progressive sequence follows unless the generated representation modifies the conditions responsible for constructing another representation.

The theory also differs from unconstrained imagery drift. A succession of images may be associative without being progressive. Progress requires that later states depend upon information introduced during earlier cycles. The sequence must accumulate, transform, test, or otherwise carry forward processing products.

The word progressive does not imply that the sequence necessarily improves or approaches truth. Imagery can drift toward false conclusions, perseverate, confabulate, or amplify a misleading assumption. Progressiveness refers to cumulative and path-dependent development, not guaranteed accuracy.

4. The Processing Cycle

4.1 Maintenance of the higher-order state

A PIM cycle begins with a set of active higher-order representations. These representations may include externally derived percepts, internally retrieved concepts, goals, emotional priorities, motor intentions, and recent intermediate results.

Some items are newly activated, while others have persisted from earlier cycles. Their coexistence creates a structured context. A representation for glass, for example, will contribute differently when combined with table, edge, and hand than when combined with cabinet, shelf, and wash.

Persistent activity is therefore not merely memory storage. It determines the relational field within which each representation operates. By remaining active, an item continues to shape the interpretation of new information and the probability distribution over possible updates.

4.2 Top-down construction of a map

The active associative state sends divergent activation toward modality-specific systems. For explanatory convenience, this transformation can be called rendering, although no literal screen or central observer is implied.

If the active state includes glass, table, edge, and hand, the visual system is constrained to construct a configuration in which these elements coexist. The resulting map must commit to spatial relationships that the abstract concepts alone leave unspecified. The glass must occupy a location, the edge must have an orientation, and the hand must be positioned relative to both.

The same principle applies to other modalities. Auditory networks can construct a temporally organized acoustic sequence, motor networks can construct a trajectory of bodily movement, and somatosensory networks can model expected contact, effort, or discomfort.

4.3 Generative completion

The sensory system receives incomplete specifications. It must fill in the missing structure using regularities encoded through previous perception and learning.

This completion process is a major source of PIM’s computational power. Sensory networks have learned how objects occupy space, how surfaces occlude one another, how bodies move, how sounds unfold, and how physical interactions tend to occur. When asked to combine several abstract constraints, they can supply probable relations and details that were not separately represented in the initiating state.

The map is therefore more than a transcription. It is a structured completion of an underspecified problem.

4.4 Bottom-up extraction

Once an internal map has been constructed, ascending pathways can analyze it much as they analyze an externally driven percept. Features, conjunctions, relations, and affordances present in the map can activate corresponding higher-order representations.

This avoids the homunculus problem that arises when mental imagery is treated as an internal display watched by another cognitive agent. No inner observer is required. The map acts directly upon the network, and its structure changes which assemblies and ensembles become active.

The 2016 formulation described a rapid feedforward sweep in which the topographic bindings of a generated map are disintegrated into salient higher-order features. Those features are then combined with representations that remained active from previous cycles, creating the specifications for another round of imagery generation. 

Incremental change in state spanning cortical.pdf

4.5 Partial updating

The newly extracted features compete for entry into the active associative state. Some previous representations continue to receive sufficient activation and are maintained. Others lose relevance or become inhibited. One or more map-derived representations become active.

The state:

\{\text{glass, table, edge, hand}\}

may therefore become:

\{\text{glass, edge, hand, push}\}

The new item push was not necessarily present in the initial state. It emerged because the constructed map positioned the hand in a way that implied contact and movement.

4.6 Recurrence

The revised state is sent down the hierarchy again. The next map may depict the glass moving beyond the edge, which introduces falling. A later map may introduce impact, breaking, or injury.

The endpoint was not directly retrieved from the starting concepts. It was reached through intermediate imagery states that disclosed the relations needed to advance the simulation.

This is the fundamental PIM sequence:

\text{glass + table + edge + hand}

\downarrow

\text{image of contact}

\downarrow

\text{push + glass + edge + movement}

\downarrow

\text{image of falling}

\downarrow

\text{impact + breaking}

Each map is both a product and a probe. It expresses the current model while testing what that model entails.

5. Imagery as Computation

5.1 Imagery is not merely illustrative

A common intuition treats imagery as a picture accompanying an already completed thought. PIM assigns it a stronger role. Imagery can participate in determining what the thought becomes.

The later iterative-updating architecture states this explicitly. Internally generated maps may introduce features or objects incidental to the image itself, and these image-derived additions can enter working memory. Logical and relational information contained in visual or acoustic imagery can therefore inform reasoning (Reser, 2022/2024). 

Thought iterative updating.pdf

This interpretation is especially important when abstract representations underdetermine their joint consequences. A thinker may understand each component separately while remaining unable to determine what follows from their combination. Rendering the components into a common map forces the system to resolve spatial, temporal, and compositional relations.

5.2 Alternating compression and expansion

The present article interprets the PIM cycle as an alternation between compressed and expanded representational formats.

Higher-order concepts are comparatively compressed. The concept glass preserves invariant information across many possible glasses while omitting most details about position, orientation, illumination, and current use. A sensory map expands that concept into a particular configuration.

The downward transformation can therefore be represented as:

\text{abstract constraints}
\rightarrow
\text{concrete configuration}

The upward transformation performs a complementary operation:

\text{concrete configuration}
\rightarrow
\text{salient concepts and relations}

PIM repeatedly alternates between these formats:

\text{compress}
\rightarrow
\text{expand}
\rightarrow
\text{inspect}
\rightarrow
\text{recompress}
\rightarrow
\text{revise}
\rightarrow
\text{expand again}

This alternation allows a cognitive system to uncover implications that remain latent within compressed representations. Association cortex may represent box, opening, rotate, and shelf. A spatial map can reveal whether a particular orientation permits the box to pass through the opening. The map-derived result can then be recompressed into fits or does not fit.

5.3 From implicit structure to explicit content

Sensory networks contain large amounts of knowledge that are not ordinarily available as explicit propositions. Visual networks embody regularities concerning geometry, occlusion, object permanence, and bodily movement. Auditory networks embody regularities concerning rhythm, phonology, source identity, and temporal order. Motor networks embody regularities concerning reachability, balance, effort, and action sequences.

PIM can convert some of this implicit knowledge into explicit working-memory content. The system provides a set of abstract specifications, allows a specialized network to instantiate them, and then extracts the relations that appear in the resulting map.

In this sense, progressive imagery modification is an implicit-to-explicit conversion mechanism. It permits knowledge encoded in the structure and weights of lower-order networks to become a conscious or reportable element of higher-order thought.

5.4 Novel convergence through imagery

The 2016 paper illustrates this property with the representations pink, rabbit, and drum. When these higher-order features are combined within visual and auditory systems, the resulting map may resemble the Energizer Bunny. This composite can activate the representation for battery, even though none of the original concepts independently specified batteries. 

Incremental change in state spanning cortical.pdf

The inference can be represented as:

\text{pink + rabbit + drum}
\rightarrow
\text{perceptually completed composite}
\rightarrow
\text{Energizer Bunny}
\rightarrow
\text{battery}

The imagery system has reconciled several partial constraints and exposed a higher-order identity implicit in their conjunction. Such events may contribute to recognition, insight, analogy, and creative recombination.

6. What Makes an Imagery Sequence Progressive?

6.1 Continuity

Continuity occurs when consecutive imagery states share causal constraints. Image I_{t+1} resembles image I_t because many of the active representations responsible for I_t remain active during the construction of I_{t+1}.

This property gives mental imagery its scene-like and video-like character. The later iterative-updating paper describes consecutive maps as capable of exhibiting video-like continuity because successive images use many of the same working-memory items as constraining parameters. 

Thought iterative updating.pdf

6.2 Novelty

A perfectly preserved state would produce repetition rather than progress. At least one representation, relation, or weighting must change.

Novelty can enter from several sources. A generated map may introduce an incidental feature, multiassociative search may retrieve a new concept, external sensory input may alter the model, or goal and reward systems may change which content receives priority.

6.3 Accumulation

A newly discovered feature must remain available long enough to influence later processing. If every image-derived result disappears immediately, the sequence cannot compound.

Accumulation allows image₃ to depend on information generated during image₁ and image₂. This is the imagery-specific form of the broader iterative compounding process described in the working-memory architecture. Simple processing products become components of more complex states, making results attainable that no single association could produce independently. 

Thought iterative updating.pdf

6.4 Path dependence

A genuinely progressive sequence is path-dependent. Changing an intermediate image or changing which feature is extracted from it can redirect the subsequent trajectory.

Suppose an imagined glass is represented as plastic rather than glass during an intermediate cycle. The predicted endpoint may change from shattering to bouncing. The final state depends not only on the starting constraints but also on the representational decisions made during the intervening transformations.

6.5 Convergence or productive transformation

Some PIM sequences converge toward a goal, prediction, decision, or stable interpretation. Others remain exploratory and transform the problem without reaching a single endpoint.

Both can be progressive. A sequence may reduce uncertainty, reveal an incompatibility, generate alternatives, or reframe the original question. The defining property is cumulative transformation, not the presence of a predetermined answer.

7. Progressive Imagery Modification as Mental Simulation

7.1 Predictive simulation

Prediction requires a system to represent the present conditions and infer how they are likely to develop. PIM supplies a mechanism for doing so through internally generated state transitions.

The later working-memory article offers the example of imagining a wilting plant with dry soil. This state activates water, then a watering can, then tilting, pouring, and eventually stopping. Each update modifies the scenario and supplies the conditions for the next imagined event. 

Thought iterative updating.pdf

Such a sequence does not require a complete symbolic program stored in advance. Learned associations and sensorimotor regularities can jointly determine each local transition. The sequence nevertheless acquires algorithmic structure because every intermediate state constrains what can occur next.

7.2 Counterfactual simulation

Counterfactual reasoning requires the mind to preserve much of a model while changing one or more assumptions. PIM is naturally suited to this operation because its active state is only partially updated.

A person can preserve room, table, glass, and edge while replacing adult with child, empty glass with full glass, or stationary hand with moving hand. The imagery system then reconstructs the scenario under the altered constraint and exposes a different set of consequences.

Counterfactual reasoning therefore involves controlled variation within a persistent representational frame. The cognitive system asks what changes when one parameter changes while holding the remainder sufficiently constant.

7.3 Branching

An imagery sequence need not proceed along only one path. An earlier intermediate state can be reinstated and modified differently, producing an alternative branch.

The iterative-updating framework depicts this possibility as returning to the midpoint of an earlier sequence and solving the problem in another way. The second trajectory begins with a reinstated subset of the earlier state but diverges when different updates are introduced. 

Thought iterative updating.pdf

Applied to imagery, this operation allows the thinker to compare possible futures. One branch may depict opening a door, another waiting, and a third taking an alternate route. Their anticipated consequences can then be evaluated against the same enduring goal.

7.4 Suspension, resumption, and merging

Complex problems can exceed the capacity of the focus of attention. One imagery thread may therefore be suspended in a broader short-term store while another subproblem is processed.

The later architecture proposes that separate iterative threads can produce partial solutions that are later merged. Selected contents from two subsolutions are coactivated and used together to generate a final solution. 

Thought iterative updating.pdf

Within PIM, one sequence might explore the spatial arrangement of a device while another explores the sequence of actions needed to operate it. Their relevant products can then be combined into a new state that supports a more complete simulation.

8. Progressive Imagery Modification and Deliberative Thought

8.1 Iterated automatic processing

Dual-process theories distinguish rapid, automatic processing from slow, controlled deliberation. PIM suggests that these may differ partly in their temporal organization rather than requiring entirely separate computational machinery.

A rapid associative or perceptual operation produces a local result. If task-relevant representations persist, that result is incorporated into another operation. The outputs of successive automatic processes can therefore support, constrain, and correct one another.

The 2016 paper proposed that System 2 cognition may emerge when System 1-like processing operates repeatedly under sustained contextual constraints. A difficult task recruits ensembles that remain active across successive imagery-generation cycles, allowing intermediate images to provide scaffolding for later deductions and expectations. 

This yields a concise interpretation:

Deliberative thought consists partly of rapid associative and perceptual operations whose intermediate products are stabilized and recursively reused.

8.2 Algorithmic sequences

Many cognitive tasks require a prescribed order of intermediate operations. Arithmetic, route planning, tool use, sentence production, and multistep problem solving depend on states that successively recruit the information required for the next operation.

PIM can implement such algorithms without requiring the entire procedure to be simultaneously active. Each imagery state represents the current status of the procedure. The retained contents preserve the problem, while the newly generated content specifies the next operation or intermediate result.

A later state may therefore contain information unavailable to any earlier state. The final output is a compound product of the sequence.

8.3 Error correction and alternative search

The newest update is not always useful. An image-derived inference may be incompatible with the goal or may lead the simulation into an implausible state.

In such cases, the candidate representation can be inhibited while the remaining contents continue searching for another update. Repeated exclusion of unhelpful candidates narrows the search space and permits the system to explore alternatives. This imagery-mediated search resembles deliberation because the mind can generate a possibility, inspect its consequences, reject it, and try again.

8.4 Cognitive compilation

Repeatedly traversing the same imagery sequence can alter long-term associative structure. The starting conditions and final result may become so strongly associated that the intermediate sequence is no longer required.

The later iterative-updating article describes how repeated reconciliation of an initial state with a derived solution can make the intermediate steps implicit. When the initial state is encountered again, multiassociative search may retrieve the endpoint directly. 

Thought iterative updating.pdf

This suggests a mechanism by which explicit deliberation becomes intuition. PIM initially constructs a result through multiple imagery cycles. Learning then compiles the sequence into a more direct association, permitting rapid future performance while preserving the possibility of reconstructing the longer chain when necessary.

9. A Modality-General Theory

The term imagery often evokes visual pictures, but the proposed mechanism is broader. A topographic or map-like internal representation can occur in visual, auditory, somatosensory, proprioceptive, motor, and linguistic systems.

In auditory PIM, a persistent set of higher-order representations can constrain a sequence of sounds. Features of the generated acoustic pattern may then alter the higher-order state, changing the next sound, word, rhythm, or melodic phrase.

In motor PIM, goals and object representations can constrain internal action maps. Predicted proprioceptive and sensory consequences feed back into the workspace, permitting the movement to be revised before execution.

In linguistic PIM, concepts, syntactic expectations, and communicative goals can constrain the construction of an utterance. The partially generated sentence then changes which words and relations are most likely to follow. The same reciprocal dynamics may contribute to inner speech.

The earlier papers explicitly extend progressive modification beyond vision. They propose that analogous processes can participate in language production, internal monologue, motor sequencing, preparatory states, and planning. 

These modality-specific processes can also interact. A visual map can introduce a motor affordance, a motor simulation can predict a visual outcome, and inner speech can change the goal constraining both. Progressive imagery modification is therefore best understood as a distributed multimodal process coordinated by a partially persistent associative workspace.

10. Regulation of Progressive Imagery Modification

10.1 Rate of updating

The rate of iterative updating is a major control variable. When few representations are replaced during each cycle, imagery remains tightly coupled to its recent past. This favors sustained analysis, detailed simulation, and the preservation of intermediate results.

When a larger proportion is replaced, the sequence becomes more responsive and exploratory. Greater turnover may facilitate spontaneous associations and novel combinations, but it also increases the risk that essential constraints will be lost.

PIM therefore operates within a tradeoff between stability and flexibility. Too little stability produces fragmentation. Too much stability produces rigidity or perseveration. Effective cognition requires a context-sensitive balance.

10.2 Working-memory capacity

The number of representations that can remain active also shapes the imagery sequence. A larger active set allows more constraints to be jointly rendered and may support more specific, context-sensitive maps.

However, increasing capacity does not guarantee improvement. An overfilled state may contain incompatible or irrelevant constraints. Selection and prioritization remain necessary so that the generated map reflects a coherent problem rather than an indiscriminate accumulation of content.

10.3 Motivation, novelty, and dopamine

The 2016 account links sustained firing and contextual maintenance to dopaminergic modulation. Novel, rewarding, punishing, or surprising events can prolong the activity of representations judged relevant to the situation.

Within PIM, this provides a plausible mechanism for deepening a simulation. An important problem causes its defining features to remain active across more imagery cycles. The system continues to render and interrogate the same general scenario rather than allowing it to dissolve rapidly into unrelated thought.

10.4 Schemas and learned scripts

A previously learned schema can enter the associative state and supply a sequence of expected relations. The schema does not determine every detail, but it constrains how generated maps are interpreted and which updates become probable.

This allows prior knowledge to organize PIM without reducing the process to rigid replay. A restaurant schema, for example, may provide expectations about entering, ordering, eating, and paying, while the imagery sequence fills in context-specific people, objects, conversations, and deviations.

11. Mental Continuity and Conscious Experience

Progressive imagery modification offers an account of why internally generated content can feel continuous even though its neural components are constantly changing. Consecutive images share enduring higher-order causes, while each introduces a limited transformation.

The experienced stream can therefore resemble a moving scene rather than a sequence of isolated snapshots. The continuity lies neither in the complete preservation of one image nor in the activity of a single representation. It lies in the overlapping succession of distributed states.

This relationship also explains why abrupt attentional shifts can interrupt a train of thought. When most of the active associative constraints are replaced at once, the next map is generated from a substantially different state. The former imagery thread may persist weakly in short-term memory, but immediate phenomenal continuity is reduced.

PIM may contribute to the continuity and elaboration of conscious content, but the theory does not identify progressive imagery modification with consciousness as a whole. The 2016 formulation explicitly acknowledges that icSSC and mental continuity resemble consciousness in important respects without being identical to it. Other mechanisms and conditions are required for a complete theory of subjective experience. 

Incremental change in state spanning cortical.pdf

The process may also operate with varying degrees of phenomenal access. Some cycles may produce vivid visual or auditory experience, while others may remain weak, schematic, or inaccessible to report. The essential functional criterion is causal recirculation, not subjective vividness alone.

12. Formal Model

Let W_t denote the higher-order working-memory state at time t. It is a graded, distributed pattern rather than a literal list of discrete symbols, although item notation can be used as an abstraction.

Let I_t^m denote the internally generated map in modality m. A modality-specific generative function G_m constructs that map from the current working-memory state:

I_t^m = G_m(W_t, X_t^m)

Here X_t^m represents any concurrent external input. During perception, external input may dominate and top-down activity may be modulatory. During imagination, external input may be absent or attenuated, allowing W_t to provide the primary constraints.

The generated maps are analyzed by an encoding function E:

Z_t = E(I_t^1, I_t^2, \ldots, I_t^M)

Z_t contains candidate features, relations, predictions, and affordances extracted from the multimodal imagery state.

A partial updating function U then creates the next working-memory state:

W_{t+1} = U(W_t, Z_t, Q_t)

Q_t represents goals, reward signals, task requirements, and other control variables. The update function retains selected components of W_t, suppresses others, and adds selected components of Z_t.

The full cycle is:

W_t
\rightarrow
I_t
\rightarrow
Z_t
\rightarrow
W_{t+1}
\rightarrow
I_{t+1}

A sequence qualifies as progressive imagery modification when the following conditions are satisfied. W_t and W_{t+1} must share information, they must also differ, information extracted from I_t must contribute causally to W_{t+1}, and at least one later state must depend upon an intermediate image-derived update.

Several measurable properties follow from this formulation.

PIM continuity is the representational similarity between successive higher-order states or successive imagery states. It can be estimated using overlap coefficients, cosine similarity, representational similarity analysis, or other appropriate measures.

PIM novelty is the amount of information introduced into W_{t+1} that was not already explicit in W_t. The strongest evidence of novelty would be a feature first detectable in an internally generated sensory map and only later detectable in higher-order association activity.

PIM depth is the number of causally linked imagery cycles completed before an action, response, solution, or attentional reset.

PIM branching is the number and extent of alternative trajectories generated from a reinstated state.

PIM convergence is the degree to which successive states approach a stable interpretation, prediction, or goal condition.

These variables separate imagery vividness from imagery function. A sequence could be low in reported vividness yet high in continuity, depth, and task relevance.

13. Empirical Predictions

13.1 Persistent associative codes should span changing imagery frames

During a multistep imagery task, representations corresponding to the enduring subject, setting, or goal should remain decodable across several consecutive states. Sensory and sensorimotor patterns should change more rapidly as individual configurations are successively constructed.

The degree of persistence in association areas should predict the similarity and coherence of successive imagery reports. A larger shared associative state should generally produce more closely related maps.

13.2 Image-derived information should appear first in modality-specific networks

The strongest test of the theory concerns the direction of information flow. A relation or feature that is not explicit in the initial instructions should sometimes become detectable in a generated sensory map before it becomes detectable in the higher-order state governing the next cycle.

For example, participants could be given several abstract spatial constraints whose consequence becomes apparent only when they are jointly visualized. Time-resolved neural measures could test whether the critical relation emerges first in visual-spatial activity and subsequently appears in frontoparietal or association-level representations.

13.3 Intermediate imagery should have causal effects on later conclusions

If intermediate maps perform computation, disrupting them should alter later reasoning. Interference delivered during a critical imagery step should change the endpoint more than equivalent interference delivered after the relation has already been encoded into the higher-order state.

This prediction could be tested with temporally targeted transcranial magnetic stimulation, visual masking, concurrent spatial tasks, or modality-specific interference. A selective disruption of intermediate imagery would support the claim that the sequence is computationally necessary rather than merely epiphenomenal.

13.4 Working-memory overlap should predict continuity, while moderate novelty should predict progress

Very low overlap between successive states should produce fragmented imagery and poor multistep performance. Near-complete overlap should produce repetitive imagery with little advancement.

The most productive sequences should combine substantial continuity with nonzero novelty. The optimal ratio may vary by task, with precise planning favoring greater stability and creative exploration favoring somewhat greater turnover.

13.5 Reinstating an earlier state should produce measurable branching

Participants could first simulate a sequence toward one outcome, then return to an identified intermediate state and alter one assumption. Neural activity should initially reinstate part of the earlier pattern and then diverge as the alternative trajectory unfolds.

The degree of reinstatement should predict how effectively the participant preserves the original context. The degree of subsequent divergence should predict the distinctiveness of the counterfactual outcome.

13.6 Repeated PIM should support cognitive compilation

A task that initially requires several imagery transformations should become faster and less dependent on intermediate imagery after repeated practice. The starting condition should gradually acquire a more direct association with the final result.

Neural activity should correspondingly shift from an extended sequence of intermediate states toward more rapid endpoint recruitment. Reintroducing an unusual or conflicting condition should restore the longer sequence because the compiled shortcut no longer suffices.

13.7 Neuromodulatory engagement should increase PIM depth

Tasks involving novelty, error, anticipated reward, or important consequences should prolong the maintenance of task-relevant representations. This should increase the number of tightly coupled imagery cycles before attention shifts.

The same manipulation may improve performance on problems requiring sustained simulation while impairing tasks that require rapid disengagement and flexible switching. The effect should therefore depend on whether persistence is adaptive in the current context.

14. Implications for Artificial Intelligence

An artificial system would not instantiate progressive imagery modification merely by generating an image from a prompt. The generated image must become part of the system’s own causal processing loop.

A PIM-capable architecture would maintain a persistent multimodal workspace containing active internal representations. These representations would constrain generative visual, auditory, linguistic, and sensorimotor modules. The outputs of those modules would then be re-encoded, evaluated, and permitted to alter the workspace before another generation cycle.

The architecture would therefore require at least five interacting capacities. It would need persistent higher-order state, modality-specific generative models, encoders capable of analyzing internally generated outputs, a partial state-updating mechanism, and a goal or value system that regulates persistence, selection, branching, and termination.

The essential cycle would be:

\text{internal state}
\rightarrow
\text{generated world model}
\rightarrow
\text{self-perception of that model}
\rightarrow
\text{state revision}
\rightarrow
\text{new world model}

The later cognitive architecture already proposes that artificial sensory maps should be generated in synchrony with iterative changes in association-level representations. Consecutive maps would then form synthetic imagination, allowing the system to see, hear, and model hypothetical situations internally (Reser, 2022/2024). 

Thought iterative updating.pdf

The crucial distinction is between output generation and self-informing generation. In an ordinary output system, an image is produced for an external observer. In PIM, the machine itself extracts consequences from the image and incorporates them into the next internal state.

Such a system could use imagery to investigate spatial arrangements, predict physical consequences, rehearse actions, compare alternatives, and detect incompatibilities among abstract constraints. It could also suspend and resume simulation threads, branch from earlier states, and merge partial results from separate modalities or subproblems.

The architecture may offer interpretability benefits if internally generated maps and state transitions can be recorded. The resulting images would not provide a complete translation of distributed representations, but they could reveal part of the sequence through which the system arrived at a plan or conclusion. The later article similarly proposes that generated visual and auditory maps could be saved and inspected as a partial record of artificial inner processing. 

Thought iterative updating.pdf

Progressive imagery modification also offers a possible bridge between subsymbolic and symbolic cognition. Distributed neural systems would construct map-like representations, while persistent higher-order items would provide stable, reusable variables. Repeated transformations between these formats could permit continuous neural computation to support sequential, compositional, and algorithmic thought.

15. Relationship to the Central Executive

Traditional working-memory models often assign the manipulation, coordination, and updating of imagery to a central executive. This label identifies a set of functions without specifying how they arise from distributed neural processing.

PIM offers a more mechanistic alternative. No single controller needs to construct the image, inspect it, decide what it means, and issue the next command. Specialized systems perform these functions through reciprocal signaling, competitive and cooperative activation, persistent context, and partial updating.

The apparent executive operation emerges from the cycle. Association networks supply constraints, sensory networks integrate them, ascending pathways extract relations, multiassociative search selects updates, and motivational systems regulate persistence. The later iterative-updating article similarly argues that executive functions may emerge from collective interactions among specialized subsystems rather than from a separate central executive mechanism. 

Thought iterative updating.pdf

This interpretation does not eliminate executive control. It decomposes it into interacting processes whose combined activity produces organized, goal-sensitive progression.

16. Limitations and Boundary Conditions

The theory remains qualitative and requires direct empirical testing. The neural units called representations, assemblies, ensembles, items, and maps are useful abstractions, but their exact biological realization is unresolved.

Sustained firing provides one plausible basis for the persistence required by PIM, yet it may not be the only one. Activity-silent synaptic states, dynamic population codes, oscillatory coordination, attractor dynamics, and other mechanisms may preserve information across cycles. The theory requires functional persistence and partial state overlap, not commitment to a single cellular implementation.

The distinction between high-order associative representations and low-order topographic maps is also simplified. Cortical processing contains many intermediate levels, recurrent pathways, and cross-modal interactions. PIM is likely to involve multiple nested loops rather than a single alternation between two sharply separated systems.

The theory does not claim that all thought depends on vivid sensory imagery. Some reasoning may rely heavily on linguistic, motoric, schematic, or weakly phenomenal representations. These forms can still participate in progressive modification if their outputs feed back into the evolving state.

PIM is also not guaranteed to be rational. Generated maps reflect learned priors and can introduce stereotyped, incomplete, or false structure. A coherent sequence may still be inaccurate if the underlying associations are poorly calibrated.

Finally, progressive imagery modification is not proposed as a sufficient condition for consciousness. It may help explain the continuity, elaboration, and self-referential development of conscious content, while leaving unresolved why some neural states possess phenomenal character.

17. Discussion

Progressive imagery modification provides a unified account of several capacities often treated separately. Mental imagery, working-memory updating, associative retrieval, simulation, planning, and deliberation can all be understood as aspects of one recurrent process.

The key departure from static accounts is the treatment of imagery as a causal participant in thought. Higher-order representations do not simply command a finished image, and sensory systems do not merely display one. Each level transforms the information supplied by the other.

The associative state compresses experience into invariant concepts and task-relevant constraints. The imagery system expands those constraints into a concrete configuration. Ascending analysis then recompresses the configuration into salient relations and concepts. Partial updating preserves the most relevant components while introducing the newly discovered information.

This repeated expansion and recompression makes imagery an instrument of inference. It enables a system to discover what a set of abstract conditions jointly implies when instantiated in a structured representational medium.

The process also clarifies the relation between continuity and change. Continuity alone would preserve a static state, while change alone would produce fragmentation. Progressive cognition requires both. A stable subset of representations carries the problem forward, while selective replacement introduces the information needed to transform it.

This combination may underlie the experienced flow of imagination. A person does not ordinarily construct every mental scene from nothing. Each scene inherits a set of concerns, objects, goals, and relationships from what came before. The newly constructed scene then reveals what should be considered next.

The proposal also helps explain how slow deliberation can emerge from rapid operations. Individual associative and sensory processes may remain automatic and local. Their products become intelligent in combination because persistent context permits them to accumulate, correct, and elaborate upon one another.

With repetition, the products of PIM can be consolidated into more direct associations. The result is a developmental and learning continuum from effortful simulation to intuitive recognition. A conclusion that initially required several transformations may later be reached immediately, although the longer sequence remains available when conditions change.

The earlier articles supplied the principal components of this theory. The 2013 formulation emphasized reciprocal transformations between working memory and imagery. The 2016 article introduced SSC, icSSC, polyassociativity, and progressive imagery modification as mechanisms of mental continuity. The later artificial-intelligence papers placed the same dynamics within an architecture of iterative updating, multiassociative search, progressive modification, and synthetic imagination (Reser, 2013, 2016, 2022, 2022/2024). The present theory draws these elements together and treats the reciprocal imagery loop as an independent explanatory mechanism. The bibliographic continuity of these proposals is documented in the later manuscript’s references to the 2013, 2016, and 2022 works. 

Thought iterative updating.pdf

18. Conclusion

Progressive imagery modification is a recurrent process in which partially persistent higher-order representations generate sensory and sensorimotor maps, and the emergent structure of those maps modifies the higher-order state responsible for generating the next one. The resulting imagery sequence preserves context while accumulating novelty.

This organization allows a cognitive system to do more than picture what it already knows. It can render abstract constraints into concrete relations, inspect their implications, revise its active model, and repeat the process. Through this cycle, implicit knowledge embedded in sensory and motor networks can become explicit content in working memory.

Later images can depend on discoveries made during earlier images. The sequence can therefore simulate consequences, branch into counterfactual alternatives, merge subsolutions, implement learned procedures, and progress toward conclusions unavailable to any single processing step.

The same architecture offers a possible account of deliberative thought. Rapid automatic operations become components of an extended reasoning process when their products are stabilized and recursively reused. Mental continuity supplies the temporal scaffold on which complex cognition is assembled.

For artificial intelligence, the principle requires a system capable of generating internal models and perceiving the implications of its own constructions. An image must function as an internal computational state rather than merely as an external product.

Progressive imagery modification thus provides a candidate mechanism by which minds and machines can think through the iterative transformation of internally generated worlds. It connects neural persistence, working-memory updating, hierarchical imagery, associative search, mental simulation, and cognitive continuity within a single recurrent architecture.

References

Baddeley, A. D. (2000). The episodic buffer: A new component of working memory? Trends in Cognitive Sciences, 4(11), 417–423.

Baddeley, A. D. (2007). Working memory, thought and action. Oxford University Press.

Baars, B. J., & Franklin, S. (2003). How conscious experience and working memory interact. Trends in Cognitive Sciences, 7, 166–172.

Carpenter, G. A., & Grossberg, S. (2003). Adaptive resonance theory. In M. A. Arbib (Ed.), The handbook of brain theory and neural networks (2nd ed., pp. 87–90). MIT Press.

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.

D’Esposito, M., & Postle, B. R. (2015). The cognitive neuroscience of working memory. Annual Review of Psychology, 66, 115–142.

Damasio, A. R. (1989). Time-locked multiregional retroactivation: A systems-level proposal for the neural substrates of recall and recognition. Cognition, 33, 25–62.

Fuster, J. M. (2009). Cortex and memory: Emergence of a new paradigm. Journal of Cognitive Neuroscience, 21(11), 2047–2072.

Goldman-Rakic, P. S. (1995). Cellular basis of working memory. Neuron, 14(3), 477–485.

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

Lewis-Peacock, J. A., Drysdale, A. T., Oberauer, K., & Postle, B. R. (2012). Neural evidence for a distinction between short-term memory and the focus of attention. Journal of Cognitive Neuroscience, 24(1), 61–79.

Mellet, E., Petit, L., Mazoyer, B., Denis, M., & Tzourio, N. (1998). Reopening the mental imagery debate: Lessons from functional anatomy. NeuroImage, 8(2), 129–139.

Meyer, K., & Damasio, A. (2009). Convergence and divergence in a neural architecture for recognition and memory. Trends in Neurosciences, 32(7), 376–382.

Miller, E. K., & Cohen, J. D. (2001). An integrative theory of prefrontal cortex function. Annual Review of Neuroscience, 24, 167–202.

Miyashita, Y. (2005). Cognitive memory: Cellular and network machineries and their top-down control. Science, 306, 435–440.

Panichello, M. F., & Buschman, T. J. (2021). Shared mechanisms underlie the control of working memory and attention. Nature, 592, 601–605.

Reser, J. E. (2013, July). The neurological process responsible for mental continuity: Reciprocating transformations between a working memory updating function and an imagery generation system. Association for the Scientific Study of Consciousness Conference, San Diego, California.

Reser, J. E. (2016). Incremental change in the set of coactive cortical assemblies enables mental continuity. Physiology & Behavior, 167, 222–237.

Reser, J. E. (2022). Artificial intelligence software structured to simulate human working memory, mental imagery, and mental continuity. arXiv:2204.05138.

Reser, J. E. (2022/2024). A cognitive architecture for machine consciousness and artificial superintelligence: Updating working memory iteratively. arXiv:2203.17255.

Stokes, M. G. (2015). Activity-silent working memory in prefrontal cortex: A dynamic coding framework. Trends in Cognitive Sciences, 19(7), 394–405.

Zanto, T. P., Rubens, M. T., Thangavel, A., & Gazzaley, A. (2011). Causal role of the prefrontal cortex in top-down modulation of visual processing and working memory. Nature Neuroscience, 14, 656–661.