Making Vision — Version 2

Abstract

Vision is often described as though its primary task were to turn photons into recognized objects.

That is too narrow.

An organism does not encounter the world as an empty visual field waiting to be labelled. It already possesses persistent knowledge, spatial maps, object and material knowledge, expectations, learned behaviours, inherited information, social information, current needs, physiological state and action machinery.

Vision enters that already-functioning system and supplies current information to it.

But vision is not merely an API into a database either.

A genuinely unfamiliar environment can still be navigated. A novel surface can still be matched. Moving neighbours can still be tracked. Free space can still be extracted. Apparent depth can alter action. Partial objects can generate useful candidates. A degraded visual representation can remain sufficient for some actions while becoming insufficient for reading or fine discrimination.

Current visual structure can therefore be useful before the organism knows what the thing is.

At the same time, the organism cannot deeply process everything it receives. Processing has costs. What becomes worth additional processing depends on current sensory information, persistent knowledge, expectations, physiological and motivational state, available energy, current needs, uncertainty, risk, learned relevance and the statistical structure of the environment.

Making Vision V2 therefore treats biological vision as a selective allocation system operating on current sensory structure and continuously recalibrating that allocation to the organism's current sensory field and internal state.

The central problem is not:

How does vision recognize everything?

It is:

How does an organism allocate limited processing across incomplete, changing, internally and externally generated information while continuously adapting that allocation to the environment?

V2 retains the central V1 insight:

Current visual computation + persistent knowledge + selective promotion + state-dependent allocation.

It expands it by separating recognition from representation, candidate generation from object identity, promotion from generic attention, learning from storage, current sensory information from persistent spatial knowledge, apparent geometry from physical geometry, and external acquisition from internal projection.

It also adds:

• dimensional impoverishment;

• degraded visual information and task dissociations;

• temporal construction;

• cue compression and substitution;

• apparent-depth/action calibration;

• environmental stimulus-regime adaptation;

• recalibration after environmental change;

• behavioural commitment as a readout;

• map and action-rule staleness;

• energy and physiological state as allocation variables;

• stronger treatment of cross-modal correction;

• more explicit research tests.

1. The organism does not start empty

Vision does not create the organism's world knowledge from scratch.

Before an organism encounters a particular visual situation, it may already know:

• objects;

• materials;

• components;

• properties;

• characteristic behaviours;

• affordances;

• dangers;

• safe locations;

• routes;

• landmarks;

• social information;

• inherited responses;

• learned procedures;

• recurring environmental relationships.

This knowledge can come from several sources.

Inherited information

Some organisms arrive with executable behavioural information already available.

The useful architectural distinction is:

know + executable set

rather than:

everything must first be learned through individual experience.

Individual learning

Repeated encounters can turn:

unfamiliar thing

into:

known object/material + properties + behaviours + consequences.

Social learning

One organism can acquire information through another.

A behaviour performed by one individual can therefore become information available to others.

This means that the first individual encountering a new object can change the information landscape for every later individual.

Distributed information

Persistent information need not reside entirely inside one individual.

It can exist across:

• individuals;

• groups;

• cultures;

• environmental structures;

• physical artifacts;

• recurring social practices.

The supporting research for V1 found direct evidence that developmental history can alter object interaction and imitation, including differences between enculturated and mother-reared primates. It also found ecological differences in novelty response between wild and captive animals. These support experience-dependent object knowledge and the cost of exploration, but they do not by themselves establish the complete Making Vision architecture.

2. Object knowledge is not a visual catalogue

An object representation is not simply:

picture → name.

An object can accumulate information about:

• appearance;

• material;

• texture;

• smell;

• taste;

• components;

• affordances;

• danger;

• safety;

• characteristic movement;

• characteristic interaction;

• methods;

• consequences;

• familiarity;

• location;

• social significance.

An object therefore has:

verbs as well as nouns.

A bird does not merely have wings.

It has characteristic ways of:

extending, flying, landing, turning, fleeing.

A predator does not merely know prey shape.

It may know:

prey movement, escape behaviour, preferred routes, likely reactions.

This becomes especially important when current sensory information disappears temporarily.

A predator following prey behind vegetation does not necessarily need to invent a new trajectory model from scratch. Existing knowledge about what that prey tends to do can contribute to what it expects next.

Affordance theory provides a strong independent correspondence here. Gibson's ecological approach explicitly treats objects in terms of what they afford an organism, rather than requiring useful perception to terminate first in an abstract object label.

Modern object-action research likewise treats object knowledge as containing action and manipulation information rather than only visual identity.

3. Object knowledge is individual

Two members of the same species do not necessarily possess the same object knowledge.

The Punch/wiper example remains useful.

For one monkey:

wiper → familiar material/object → manipulable → acceptable.

For another:

wiper → unfamiliar strange thing → uncertainty → caution.

The physical object is identical.

The persistent knowledge attached to it is not.

Thus:

same species + different developmental experience → different persistent object knowledge → different future promotion and behaviour.

Object familiarity is therefore not merely an objective property of the object.

It is partly a property of the relationship between object and organism.

This is why “the object is familiar” is incomplete.

The more useful statement is:

the organism has a sufficiently cheap and useful representation of this object.

4. Spatial knowledge is not a photograph

An organism can maintain persistent spatial knowledge without maintaining a detailed visual recording of every moment.

A spatial representation can contain:

• locations;

• layout;

• routes;

• landmarks;

• stable relationships;

• dangerous places;

• recurring destinations.

Persistent spatial knowledge tells the organism:

where things generally are.

Current sensory information tells it:

what is there now.

These are different.

A chair being moved does not necessarily require rewriting the entire room representation.

A new painting may eventually become part of familiar spatial knowledge.

A tiny temporary object may never become part of it.

A dangerous change may be incorporated much more rapidly.

Cognitive-map research independently supports map-like relational spatial representations and flexible navigation rather than requiring a photographic record of the environment.

5. Local spatial structure

For immediate action, the organism does not need its entire spatial knowledge simultaneously.

A relevant local portion can become operationally available.

V1 called this a:

maplet / local inset.

The underlying concept remains useful, but V2 treats maplet as Making Vision terminology rather than an established neuroscientific unit.

The local spatial structure combines:

• persistent spatial relationships;

• current sensory information;

• current body position;

• immediate action requirements.

It can support:

where to step
where to reach
where the door is
where the obstacle is
where the next route segment lies.

The persistent spatial representation supplies a skeleton.

Current sensory acquisition supplies what has changed.

6. The current sensory field

V2 retires the V1 term board.

This is deliberate.

“Board” has a different architectural meaning in the Memory Prism framework and cannot safely carry both meanings once Making Vision also incorporates energy, physiological state, hormonal state and resource allocation.

V2 therefore uses:

current sensory field

to mean:

the currently accessible external situation as available to the organism through its sensory systems.

The current sensory field is not the physical universe.

It is not persistent memory.

It is not an internal world model.

It is not the organism's energetic state.

It is the currently accessible external situation from which sensory systems can acquire information.

A bee and a human can occupy the same physical location while having different accessible sensory fields because their sensory systems admit different information.

The current sensory field can contain:

• surfaces;

• objects;

• people;

• pathways;

• movement;

• light;

• shadows;

• openings;

• obstacles;

• temporary opportunities;

• temporal changes.

Its accessible information is constrained by the organism's sensory apparatus before higher-level allocation and interpretation occur.

This retains the V1 species-relative point while removing the terminology collision.

7. Current sensory acquisition is incomplete

The current sensory field is never acquired without limits.

Visual access is constrained by:

• field of view;

• spatial resolution;

• sampling;

• blur;

• motion blur;

• occlusion;

• contrast;

• signal-to-noise ratio;

• illumination;

• colour availability;

• depth-cue availability;

• temporal sampling;

• peripheral versus central information;

• sensor noise.

These are not exceptional failures.

They are normal operating conditions.

A visual architecture that required complete information before acting would be unable to explain ordinary behaviour in clutter, motion, poor lighting, occlusion or degraded input.

Therefore:

incomplete sensory information is normal input.

8. Vision has direct access to current structure

Vision is not merely:

current photons → database lookup.

A genuinely unfamiliar environment can still provide immediately useful information.

A human entering a room full of unfamiliar structures can still extract:

• boundaries;

• free space;

• distances;

• relative positions;

• obstacles;

• movement;

• pathways.

The person does not need to know what every object is.

Likewise, an animal can respond to current spatial relationships among moving objects without possessing durable knowledge of every element.

Therefore vision has at least two complementary sources of useful information:

Current sensory computation

What is structurally present in the current input?

Persistent knowledge

What do I already know that helps interpret or act on it?

Persistent knowledge can dramatically amplify current perception.

It is not a prerequisite for useful current sensory computation.

9. Direct visual structure

Current visual input can provide:

• boundaries;

• contrast;

• colour distribution;

• texture;

• spatial frequency;

• orientation;

• relative position;

• motion;

• free space;

• surface continuity;

• object relationships;

• occlusion boundaries;

• negative space;

• shape/proportion relations;

• component structure.

Some of these can be useful without identifying an object.

This is particularly important for action.

The system can use:

current visual structure → downstream action

without requiring:

object recognition → persistent storage → retrieval → action.

This is consistent with the affordance/direct-perception literature, while Making Vision remains narrower than Gibson's complete theoretical position.

10. Structural extraction is not necessarily serial

V1 sometimes allowed an overly neat picture:

structure → candidate → object → action.

V2 does not require that sequence.

Current structure and prior knowledge can interact.

Object and scene processing can interact.

Scene context can influence object interpretation, while objects can influence scene interpretation. Contemporary reviews describe object and scene processing as partly parallel and bidirectionally interacting rather than as a simple serial pipeline.

Therefore V2 uses:

interacting operations

rather than:

one mandatory processing sequence.

11. Candidates are hypotheses, not objects

Partial contours, components, movement, context and relations can generate candidates.

A candidate is a processing hypothesis.

It may be:

• strong;

• weak;

• ambiguous;

• familiar-looking;

• novel;

• partially visible;

• behaviourally relevant;

• ultimately rejected.

Candidate generation can occur before stable object recognition.

Candidate rejection is itself a mechanism.

The system can generate:

“something may be there”

without committing to:

“I know what that thing is.”

The V2 inventory explicitly separates candidate generation, candidate confidence, candidate selection and candidate rejection.

12. Recognition is not necessarily image retrieval

A familiar encounter can become cheap because existing representation provides sufficient correspondence.

But the representation need not be a photograph.

It can contain:

• diagnostic shape;

• proportions;

• components;

• material;

• behaviour;

• context;

• relations;

• affordances;

• consequences.

A strong match can permit cheap recognition.

A partial match can trigger recognition plus verification.

A poor match can trigger promotion and deeper inspection.

No useful match can trigger construction of a new representation.

Therefore:

recognition can use stored structure without being retrieval of a stored picture.

Recognition and identification remain separate.

An organism can recognise:

“familiar”

without being able to:

name, classify or fully identify.

13. Dimensional impoverishment

V2 adds a direct question:

How much dimensional information does perception actually require?

Consider:

real 3D object → photograph → silhouette → cutout.

A rich object supplies:

• width;

• height;

• depth;

• volume;

• physical occlusion;

• curvature;

• texture;

• material cues.

A planar representation can preserve:

• contour;

• width;

• height;

• some component relations;

• colour;

• some texture.

It loses or alters:

• physical depth;

• volume;

• physical occlusion;

• surface curvature;

• some illumination relationships.

The important question is not simply:

“Can the organism recognise the picture?”

It is:

Which operations remain possible after dimensional information has been removed?

The V2 inventory identifies the cutout/silhouette/photo/real-object sequence as an explicit test family.

14. Planar information may be operationally sufficient

Dimensional reduction does not imply total information loss.

Contour can survive.

Proportion can survive.

Component relationships can survive.

Diagnostic markings can survive.

Some affordances can remain inferable.

Therefore:

loss of full dimensional information does not imply loss of all useful object information.

A system may operate directly on planar information without constructing a complete three-dimensional object.

Whether it does so for a particular task is an empirical question.

15. Degraded vision can dissociate tasks

Blur and reduced resolution should not be treated as a single scalar loss of “vision.”

Different operations require different information.

A familiar object may remain actionable under degraded visual input while:

• reading;

• symbol discrimination;

• fine identification;

• subtle texture discrimination

fail.

Conversely, an unfamiliar object may require fine visual structure that a familiar object does not.

V2 therefore asks:

Which operations survive which forms of information loss?

Relevant manipulations include:

• blur;

• contrast reduction;

• spatial resolution reduction;

• occlusion;

• illumination changes.

Relevant dependent variables include:

• navigation;

• approach;

• grasping;

• object recognition;

• fine identification;

• reading;

• symbolic discrimination.

The point is not to establish “vision = acuity.”

The point is to identify task-specific information thresholds.

Research on object recognition also shows that visual competence develops under conditions of initially blurry vision, demonstrating that visual processing can be functional before adult-level fine resolution is available.

16. Familiarity can make degraded information cheaper

A familiar representation can supply information that a novel representation cannot.

This means that under blur or partial visibility:

current sensory evidence + persistent knowledge

may be sufficient where:

current sensory evidence alone

would not be.

This is another reason recognition cannot be reduced to image matching.

The system can exploit partial correspondence.

It can also fail when the remaining structure is insufficient for the particular operation being attempted.

17. Temporal construction

Vision is not only spatial.

Objects move.

Objects disappear behind occluders.

Objects change state.

New structures appear.

The system therefore needs some means of relating information across time.

V2 includes:

• motion extraction;

• biological motion;

• object persistence;

• changing object states;

• temporal identity through occlusion;

• frame-to-frame correspondence;

• motion versus independent appearance;

• temporal prediction;

• temporal surprise;

• dynamic information.

A moving object does not arrive as one perfect representation.

It arrives as changing evidence.

18. Temporal identity

Current visibility and continued identity are different.

An object can become:

partially visible → invisible → visible again.

The system may treat these observations as:

one continuing entity

rather than:

three independent events.

This requires temporal correspondence.

It also creates opportunities for error.

The system can maintain the wrong identity.

It can lose identity.

It can update identity.

Temporal construction is therefore not simply “motion detection.”

19. Motion can promote

A stable object can remain background.

The same object moving unexpectedly can become a candidate event.

Promotion can therefore be triggered by:

temporal change

rather than spatial novelty.

This matters because much of an environment is stable while relatively little changes.

Change can therefore be an efficient allocation signal.

20. Temporal surprise

A system can have expectations about:

• where something will move;

• when something normally moves;

• what movement belongs to an object;

• what state follows another state.

Unexpected temporal behaviour can therefore produce perturbation even when appearance remains unchanged.

This gives another version of the V1 “behaviour is part of object knowledge” principle.

21. Lazy processing

The organism cannot deeply process everything.

There can be broad, comparatively cheap monitoring alongside more expensive selective processing.

V1 called this:

lazy scan.

V2 keeps the concept but removes a possible misunderstanding.

Lazy processing does not imply:

• one literal global scan;

• unattended information being absent;

• a strict serial coarse-to-fine algorithm;

• a universal attentional mechanism.

It means:

processing expenditure is uneven.

Some regions, features, events and objects can remain cheap while others receive additional computation.

22. Regional allocation

Allocation can vary across the sensory field.

Processing can depend on:

• familiarity;

• anomaly;

• uncertainty;

• current goal;

• risk;

• information value;

• task;

• learned relevance.

The V2 inventory explicitly adds allocation under clutter, uncertainty, familiar-cue failure and post-learning changes.

Contemporary work on visual selection increasingly treats visual prioritisation as shaped by prior knowledge, task demands and uncertainty rather than by stimulus salience alone.

23. Managed expectations

The organism maintains expectations about its current sensory field.

Examples:

a dining table has certain things on it

a doorway normally permits passage

a familiar switchboard contains familiar controls

vehicles on a road behave within ordinary patterns.

Expected information can therefore become cheap.

A familiar chair in its familiar location does not repeatedly require:

“WHAT IS THAT?”

The system already has a representation.

But expectations are not guaranteed truth.

They can produce failures.

24. Perturbation

A perturbation is information that escapes cheap background treatment and becomes potentially worth further processing.

It can be:

• unexpected movement;

• unexpected location;

• missing expected object;

• unusual object;

• unusual behaviour;

• unusual relation;

• relevant sensory event;

• failed familiar cue;

• internal sensory anomaly.

But:

not every perturbation is promoted.

A change can be noticed without becoming important.

A candidate can be inspected without being retained.

A sensory anomaly can be dismissed.

A tiny environmental change can remain operationally irrelevant.

Perturbation is therefore a possible allocation trigger, not a guaranteed allocation outcome.

25. Promotion

The critical transition remains:

background → candidate → promoted candidate/event

Promotion means:

this now warrants additional processing.

It does not necessarily mean:

“I know exactly what this is.”

Promotion can precede:

• stable classification;

• naming;

• complete identification;

• persistent storage.

Promotion is therefore an allocation operation.

It allocates additional processing, and potentially behavioural resources, to something that was previously receiving cheaper treatment.

26. Promotion is not identical to attention

Attention is an established scientific term with multiple meanings and mechanisms.

Making Vision uses promotion for a more specific architectural function.

Promotion can involve attention.

But it is not defined as attention.

Promotion can depend on:

• current state;

• physiological condition;

• energy availability;

• need;

• uncertainty;

• novelty;

• expected information value;

• prediction;

• anomaly;

• learned relevance;

• potential consequence.

Object-based attention research does provide independent evidence that attentional allocation can be organised around objects rather than only locations.

Making Vision's stronger claim—that this should be treated as a specific promotion operation embedded in a larger allocation architecture—is the synthesis.

27. A familiar object can be promoted

Novelty is not required.

A familiar object can become important because:

• its state changed;

• it moved unexpectedly;

• it behaved unexpectedly;

• its normal cue failed;

• its relation to another object changed;

• it acquired current consequence.

A traffic light is a particularly clean example.

Its identity may be extremely cheap.

Its current state can nevertheless become highly important.

Thus:

familiar object → state change → promotion → action

is a valid promotion pathway.

The promoted information need not be:

“traffic light.”

It can be:

“current state has changed.”

28. CAPSLOCK as a promotion observation

CAPSLOCK belongs here as a cross-domain observation.

A stream can contain ordinary information and then contain a visually marked element that causes the reader to allocate additional processing to it.

CAPSLOCK can therefore function as an explicit promoter in an artificial information stream.

The important functional question is:

How does an element escape cheap background treatment?

CAPSLOCK is not evidence that biological vision contains a literal CAPSLOCK mechanism.

It is an artificial analogue that makes the allocation problem visible.

29. Promotion is state-dependent

Novelty alone does not determine promotion.

The same stimulus can produce different processing depending on:

• physiological state;

• energetic condition;

• hormonal state;

• current need;

• risk;

• prior knowledge;

• recent experience;

• task.

Thus V1's:

current state × existing knowledge × current sensory field × current needs/constraints

becomes more explicit in V2:

current sensory information × persistent knowledge × internal state × environmental statistics × current consequences

where internal state can include energy availability and physiological/hormonal signals.

The exact biological implementation is not assumed.

The architectural claim is that the organism's internal condition can alter the value assigned to current sensory information.

30. Energy is part of the allocation problem

Processing costs resources.

Exploration costs resources.

Movement costs resources.

Memory maintenance costs resources.

Risk can cost survival.

Therefore an allocation architecture cannot treat all potentially useful information as equally worth processing.

The relevant question is:

What is worth spending resources on now?

This does not mean that every visual computation is consciously or explicitly budgeted.

It means that biological processing operates under energetic and physiological constraints.

V2 therefore treats energy as a constraint on the allocation system rather than as a property of the current sensory field itself.

31. Hormonal and physiological state can alter promotion

V2 adds a specific route by which internal state can influence visual allocation:

physiological state → altered valuation/sensitivity → altered promotion.

There is empirical evidence for state-dependent modulation of visual processing.

For example, hunger has been shown to alter attentional capture by food-related visual stimuli, with hungry participants showing greater capture in one established paradigm.

More recent work also finds that hunger/satiety can alter visual attention to food cues, although the direction and conditions of the effect are not uniform across paradigms.

This is important precisely because V2 should not turn “hormones promote objects” into a universal rule.

The defensible claim is:

physiological state can alter the allocation value of sensory information.

The specific hormone, pathway and task must be established separately.

32. Exploration is an expenditure

Exploration is not free.

Investigation can require:

• energy;

• time;

• attention;

• exposure;

• motor effort;

• risk;

• opportunity cost.

Therefore exploration behaviour reflects not merely:

“what is interesting?”

but:

“what is worth spending resources on right now?”

The supporting V1 research found ecological evidence consistent with this interpretation, including differences in novelty responses between wild and captive animals.

33. Novelty is not danger

A novel object does not automatically become:

dangerous.

Nor does novelty automatically produce:

exploration.

Nor does it automatically produce:

deep processing.

The response depends on the organism's existing knowledge, current state, ecological conditions and likely consequences.

The novel-object literature is heterogeneous rather than supporting a single universal novelty response.

Therefore:

novelty is one allocation variable, not an allocation command.

34. Familiar sensory field versus novel sensory field

These are different perturbation regimes.

An organism can encounter:

Familiar sensory field + novel object

known room, strange object.

Novel sensory field + familiar object

known object, unfamiliar location.

Novel sensory field + novel objects

much of the current situation uncertain.

The relative load carried by:

• persistent knowledge;

• direct sensory structure;

• exploration;

• expectations;

• candidate generation

can differ between these conditions.

A novel sensory field places greater load on direct current sensory structure because persistent spatial knowledge is less useful.

But the organism still carries object knowledge, danger knowledge, inherited defaults and learned action routines into the novel field.

35. Assessment before commitment

A candidate can be worth investigating without immediately being worth acting upon.

An organism can:

• approach;

• hesitate;

• inspect;

• probe;

• retreat;

• redirect;

• commit.

This makes behavioural geometry useful as a readout.

Potential measures include:

• approach distance;

• hesitation;

• trajectory;

• velocity;

• braking;

• takeoff angle;

• paw extension;

• head orientation;

• gaze;

• corrective movement.

The V2 inventory explicitly identifies behavioural geometry as a potentially stronger dependent variable than emotional labels in some tests.

36. The first explorer changes the information environment

When a new object enters a social environment:

new object → individual exploration → observed behaviour → social information → later exploration.

Therefore the first encounter is not interchangeable with later encounters.

Later animals may respond to:

object + information supplied by another animal.

Social-learning research independently supports the idea that observing others can alter later behaviour, attention and learned value.

Thus:

video zero matters.

If the first monkey touches, enters, climbs or plays with a new object before others encounter it, that first interaction is evidence about individual first-contact behaviour.

Later interactions are contaminated—in the scientifically useful sense—by social information.

37. Persistence is selective

Promotion does not automatically create persistent knowledge.

An organism can:

notice → inspect → act → forget.

Persistence has costs.

Useful persistent information can require:

• storage;

• connectivity;

• retrieval;

• updating;

• maintenance.

The object knowledge system should therefore not be imagined as infinitely expanding.

It is a:

managed persistent resource.

38. Persistence and learning are different

Learning is not identical to storage.

Experience can change future processing without creating a durable explicit object representation.

Conversely, something can be retained without changing every aspect of future processing.

V2 therefore distinguishes:

learning = change in future processing

from:

persistence = information retained over time.

A meaningful encounter can alter:

• candidate generation;

• recognition cost;

• promotion threshold;

• diagnosticity of a cue;

• action selection.

39. Cue compression

Repeated experience can turn expensive classification into cheap cue use.

Useful cues can include:

• colour;

• shape;

• logo;

• package geometry;

• language;

• symbols;

• combinations of these.

A supermarket environment is an especially useful example.

Initially:

inspect package → read → compare → classify.

After learning:

colour + shape + logo → category.

Eventually:

symbol alone → category.

The organism has effectively acquired a local compression dictionary.

The processing required for a familiar category can therefore become dramatically cheaper.

The V2 inventory identifies cue compression as a candidate mechanism linking recognition, promotion and learning.

40. Cue substitution

Compression can fail.

A familiar brand may appear in an unfamiliar country.

Its usual:

• language;

• logo;

• colour;

• package arrangement

may no longer work.

The system can substitute another route:

shape → comparison → text → verification.

This makes previously hidden perceptual work visible.

Thus:

learning does not merely add knowledge; it changes which sensory cues are sufficient.

41. Learning changes what counts as surprising

Repeated anomalies can become ordinary.

Repeated ordinary events can cease to receive expensive processing.

A cue that was once highly diagnostic can become redundant.

A previously ignored cue can become important after learning.

Therefore:

surprise is not a fixed property of a stimulus.

It depends partly on what the organism has learned about its environment.

This is one reason predictive processing is relevant to V2, but Making Vision does not require adopting predictive processing as its complete explanatory framework.

Contemporary work explicitly treats visual prioritisation as shaped by expectations, uncertainty, learning and task demands rather than stimulus properties alone.

42. Environmental statistics alter processing economy

The sensory system is not calibrated only to individual stimuli.

It is exposed to statistical regularities.

An environment can be:

• visually rich;

• visually monotonous;

• predictable;

• variable;

• cluttered;

• sparse;

• rapidly changing;

• slowly changing.

Repeated predictable information can become cheap.

Rare events can become important.

The system can therefore adapt not only to what is present but to:

how information is distributed in the environment.

Research on sensory adaptation increasingly describes adaptation in terms of complex environmental statistics rather than isolated stimulus properties.

43. Environmental regime change produces recalibration

V2's preferred formulation is:

change in environmental stimulus regime → recalibration period

rather than:

deprivation → overload.

A transition can occur between:

• enriched ↔ monotonous;

• predictable ↔ variable;

• familiar ↔ novel;

• one visual ecology ↔ another;

• prolonged absence ↔ return;

• one working environment ↔ another.

The system arrives carrying an allocation policy calibrated to the previous regime.

The new environment initially violates that calibration.

Processing can therefore become temporarily inefficient.

The organism must recalibrate:

• sensitivity;

• expectations;

• cue weighting;

• promotion;

• exploration;

• shortcut selection.

Sensory-adaptation research directly supports the general principle that sensory systems continuously recalibrate according to current and recent stimulus environments.

44. Recalibration is not simply “more stimulation”

The important variable is not raw stimulus quantity.

A visually rich environment can contain huge amounts of repetitive information.

A sparse environment can contain rare but highly consequential changes.

Therefore the relevant variable is closer to:

the statistical structure of sensory information and the allocation policy calibrated to it.

This is why enriched ↔ monotonous transitions matter.

The question is not merely:

how much stimulation?

It is:

what distribution of information has the system learned to expect?

45. Environmental editing

Humans can deliberately alter the current sensory field to exploit the same machinery.

Suppose something must be soaked overnight.

Instead of relying entirely on:

“remember to soak this later”

a person can place the object beside the water bottle.

The future sequence becomes:

predictable route → familiar landmark → unexpected object → promotion → task recovered.

The object becomes a:

temporary task landmark.

Its identity need not be important.

The important feature may simply be:

something visibly interrupts an expected pathway.

The person has altered the environment so their future self encounters a useful perturbation.

46. Environmental editing can substitute for memory

If a person has to remember:

“I must do X later”

they can instead arrange:

“Later I will inevitably encounter something that reminds me.”

This is cognitive offloading.

Distributed-cognition and extended-mind research provide established independent precedents for information being distributed across people, tools and environmental structures.

Making Vision's specific contribution is to connect this to future sensory allocation:

environmental edit → future sensory perturbation → promotion → action.

47. The wet-floor example

A wet floor provides a minimal case.

The immediate problem is not:

dry floor.

It is:

don't step there while it is slippery.

A conspicuous wet patch can function as a temporary landmark:

environmental change → visible perturbation → promotion → altered foot placement.

If evaporation can solve the underlying problem, the person may not need to spend additional effort mopping immediately.

The intervention can instead be:

change the sensory field enough to prevent the dangerous action.

This illustrates a broader principle:

organisms do not have to solve every environmental problem directly if they can cheaply alter their interaction with it.

48. Camouflage: anti-promotion

Camouflage provides an unusually useful counterexample to promotion.

The camouflaged organism is not necessarily trying to make the observer identify it as another species.

Often the desired outcome is simpler:

remain non-actionable.

The organism attempts to remain within:

• background;

• surface;

• environmental texture;

• expected structure.

In functional terms:

don't become the candidate worth additional processing.

Camouflage therefore provides a useful hypothesis about the other side of the promotion boundary.

But V2 must not overstate it.

Camouflage strategies differ.

Background matching, disruptive coloration, masquerade and motion camouflage do not necessarily defeat exactly the same stage of an observer's processing.

A large meta-analysis found that camouflage increased predator search time overall and reduced attack rate, with different strategies producing different effects.

Thus the defensible MV claim is:

some camouflage strategies reduce the probability or reliability of becoming an actionable perceptual candidate.

49. Camouflage is conditional

Camouflage-capable organisms do not necessarily maintain maximal camouflage continuously.

Deployment depends on:

• exposure;

• predator presence;

• environmental background;

• lighting;

• movement;

• current state;

• cost.

The capability can exist continuously while expensive deployment remains conditional.

This is structurally similar to promotion itself:

capability available ≠ capability continuously deployed.

50. Behaviour is part of object knowledge

Objects are not static visual identities.

They have expected behaviours.

Therefore:

familiar object + unexpected behaviour

can generate perturbation without visual novelty.

Examples:

familiar animal → unexpected movement

familiar bird → unexpected vocalisation

familiar object → wrong interaction

familiar switch → unexpected outcome.

The object and expected behaviour have come apart.

This can be more informative than appearance change.

The V2 inventory explicitly preserves unexpected behaviour as a candidate-generation/promotion mechanism.

51. Cross-modal disagreement

Vision does not operate alone.

It interacts with:

• proprioception;

• vestibular information;

• touch;

• audition;

• motor feedback.

Different sensory systems can disagree.

A glass door can produce:

visual open space

followed by:

physical collision.

The resulting contradiction can become a highly informative event:

this apparent open region is physically occupied.

This does not require vision to be treated as an absolute authority.

Instead:

cross-modal disagreement can trigger correction.

52. Degraded vision can recruit other senses

When visual information is degraded, other sensory systems may carry more of the operational load.

Relevant systems include:

• touch;

• proprioception;

• vestibular information;

• audition.

The organism does not necessarily compensate by making vision itself “better.”

It can change the allocation of work across modalities.

This is another reason V2 treats vision as part of a broader sensory-action system rather than an isolated module.

53. Motor shortcuts and stale representations

Well-practiced actions can operate from procedural knowledge.

Examples:

• household routines;

• familiar routes;

• repetitive actions;

• practiced motor sequences.

This is efficient when the environment remains stable.

But:

learned motor expectation + changed current sensory field

can produce errors.

The system may possess the new information and still act according to an old shortcut.

The problem is therefore not necessarily visual failure.

It can be:

old expectation + new situation + insufficient reinspection.

54. Map staleness

Persistent spatial representations inevitably become stale.

Objects move.

Controls fail.

People develop workarounds.

Routes change.

Old and new representations can coexist.

A familiar switch arrangement can change.

A workaround can become habitual.

Later the original arrangement can return.

The organism may retain both action patterns and select according to current context.

Therefore stale spatial knowledge is not necessarily catastrophic.

The architecture only needs mechanisms by which:

action failure → reinspection → updating

can occur.

55. Internal sensory events

Not every sensory event originates in an external object.

Examples include:

• phosphenes;

• illusions;

• hallucination-like percepts;

• sensory noise;

• internally generated spatial events.

The architecture does not need to prevent these from entering processing.

Instead:

classification is not guaranteed truth.

An internally generated event can become:

“something is happening.”

The organism can then:

• investigate;

• compare;

• revise;

• dismiss.

This preserves the distinction between:

sensory-like content

and:

externally verified environmental state.

56. Internal projection is broader than vision

Dreaming should not become a special theory of vision.

The broader observation is:

the brain can generate internally projected sensory/spatial content without current external sensory input.

Echolocation demonstrates a different pathway:

nonvisual sensory input → spatial representation.

Dreaming demonstrates another:

stored/internal material → internally generated scenes and events.

Therefore internal projection is broader than external visual acquisition.

Making Vision only needs to acknowledge:

current external visual input is not the only possible source of visual-like content.

57. Sleep is not sensory death

Sleep is not equivalent to total information-processing shutdown.

Sensory information can still be processed without producing awakening.

Sleep also participates in memory consolidation and reactivation. Research has documented replay of waking neural activity during sleep and continuing investigation of how sensory processing and memory reactivation interact.

Therefore:

offline from ordinary interaction ≠ offline from information processing.

V2 should avoid the stronger V1 phrase “deep repair” as a general explanation.

The useful claim is narrower:

sleep changes the operating state while permitting substantial internal processing and memory-related activity.

58. Vision does not own spatial representation

Vision supplies unusually rich current spatial information.

It does not own spatial knowledge.

Spatial information can be acquired through:

• audition;

• touch;

• proprioception;

• vestibular information;

• learned action;

• social information.

Vision's distinctive contribution is the quantity and structure of current visual information available at a distance.

Therefore:

vision supplies spatial evidence; spatial representation is not exclusively visual.

59. Apparent depth is not physical depth

Visual depth cues can imply geometry that differs from physical geometry.

Relevant cues include:

• perspective;

• shading;

• texture;

• internal pattern;

• shadow;

• stereoscopic disparity.

V2 therefore separates:

visual/apparent geometry

from:

physical geometry.

This creates an unusually clean experimental manipulation:

hold physical geometry constant → alter only visual depth cues → measure action.

60. Perceptual geometry can alter motor geometry

The relevant dependent variable need not be verbal report.

It can be:

• trajectory;

• velocity;

• takeoff angle;

• vertical impulse;

• paw extension;

• head orientation;

• gaze;

• landing velocity;

• braking;

• corrective movement.

The question is:

Does represented geometry alter motor output?

Research on familiar objects also provides a useful example of long-term visual knowledge influencing grasp parameterisation: familiarity with everyday objects altered the relationship between physical size and grip scaling.

This supports the broader MV principle that persistent visual/object knowledge can enter action control.

61. The apparent-depth test

The strongest control is:

same physical apparatus + altered visual cues implying depth.

Possible conditions:

shallow-looking

deep-looking

bottomless-looking.

Measure motor geometry rather than relying only on emotional reaction.

If physical geometry remains constant but action changes systematically with visual geometry, the visual representation is demonstrably participating in action.

The V2 inventory identifies this exact manipulation and dependent-variable family.

62. Learning loop

The system is recursive.

Existing knowledge guides current interpretation.

Current encounters generate new structure.

New structure can be retained.

Retained structure changes:

• future candidate generation;

• future recognition cost;

• future promotion;

• cue diagnosticity;

• action shortcuts.

Therefore:

learning changes what the organism will notice next.

This is one of the most important consequences of Making Vision.

The system is not merely processing the environment.

It is continually changing the processing system that will encounter the environment next.

63. Research test: dimensional representation

Manipulation

real object → photograph → silhouette → cutout

Measure

• recognition;

• identification;

• depth inference;

• correspondence;

• action.

Questions

• Which properties survive?

• Which become ambiguous?

• Is implied depth reconstructed?

• Can planar information support action directly?

• How much dimensional information is actually necessary?

64. Research test: cue failure

Manipulation

Remove or replace a familiar package cue.

Measure

Whether processing shifts from:

logo/colour → category

toward:

shape → comparison → text → verification.

Question

Does learned cue compression reduce processing cost, and does cue failure expose the previously compressed work?

65. Research test: temporal identity

Manipulation

One object moves through multiple scenes, including partial occlusion.

Measure

Whether one continuing identity is maintained.

Question

How much temporal information is required to maintain continuity?

66. Research test: behavioural perturbation

Manipulation

Keep appearance familiar.

Change behaviour.

Measure

• promotion;

• inspection;

• gaze;

• action.

Question

Can unexpected behaviour promote a familiar object without appearance novelty?

67. Research test: traffic-light promotion

Manipulation

Use a familiar traffic-light object while changing its state.

Measure

• state detection;

• gaze;

• processing latency;

• action initiation.

Question

Does promotion follow state change rather than object novelty?

This is a particularly clean way of separating:

object recognition

from:

current object-state processing.

68. Research test: degraded visual information

Manipulation

Vary:

• blur;

• contrast;

• spatial resolution;

• occlusion;

• illumination.

Measure separately:

• navigation;

• approach;

• grasping;

• object recognition;

• fine identification;

• reading;

• symbol discrimination.

Question

Which operations survive which information losses?

The desired output is not one visual-performance score.

It is a pattern of dissociations.

69. Research test: environmental remapping

Manipulation

Rearrange a familiar sensory field.

Measure:

• old shortcut errors;

• reinspection;

• hesitation;

• correction;

• new route learning.

Question

How quickly does current sensory information override stale spatial and motor knowledge?

70. Research test: apparent depth → action

Manipulation

Hold physical geometry constant.

Alter visual depth cues.

Measure:

• trajectory;

• velocity;

• takeoff;

• braking;

• body extension;

• gaze;

• corrective movement.

Question

Does represented visual geometry alter motor geometry?

71. Research test: environmental regime shift

Manipulation

Transition between:

• enriched ↔ monotonous;

• predictable ↔ variable;

• familiar ↔ novel;

• one visual ecology ↔ another.

Measure:

• detection;

• exploration;

• promotion;

• anomaly response;

• shortcut use;

• task performance;

• recalibration time.

Question

How long does allocation remain mismatched after the environment changes its statistical regime?

This directly tests:

stimulus-regime change → recalibration period.

72. Research test: internal event

Manipulation

Introduce controlled internally generated sensory-like events.

Measure:

• investigation;

• revision;

• dismissal;

• behavioural consequence.

Question

How does the system distinguish internally generated sensory-like information from external environmental events?

73. Research test: cross-modal conflict

Manipulation

Create controlled disagreement between:

• vision;

• vestibular information;

• proprioception;

• touch;

• audition.

Measure:

• corrective behaviour;

• reinspection;

• recalibration;

• sensory weighting.

Question

When does disagreement trigger revision?

74. Research test: sparse sensory field

Manipulation

Remove expected contextual cues while preserving diagnostic structure.

Measure:

What can still be recognised or acted upon?

Question

Which information is supplied by current sensory structure and which by persistent knowledge?

75. Research test: novelty allocation

Manipulation

Present novel objects differing in:

• risk;

• information value;

• familiarity;

• ecological relevance;

• current task relevance.

Measure:

• approach;

• investigation;

• gaze;

• processing time;

• avoidance;

• persistence.

Question

Does novelty alone determine allocation?

MV predicts:

no.

76. Research test: promotion under physiological state

Manipulation

Alter a relevant physiological/motivational state while holding visual stimuli constant.

Measure:

• detection;

• fixation;

• exploration;

• processing latency;

• object choice;

• action.

Question

Does internal state alter which sensory information receives additional processing?

Existing hunger research demonstrates that physiological state can alter visual attention to food cues, but the direction and mechanism depend on the experimental condition.

The V2 hypothesis is therefore:

internal state changes allocation value

rather than:

one hormone universally promotes one class of objects.

77. Important distinctions

V2 explicitly preserves the following separations.

Do not collapse Keep separate

recognition / identification familiarity is not naming

representation / retrieval a representation can be constructed

candidate / object a hypothesis is not the entity itself

promotion / attention promotion is MV's allocation operation

learning / memory changing future processing is not identical to storage

spatial knowledge / current sensory field persistent spatial structure is not current input

visual depth / physical depth apparent geometry can differ from physical geometry

internal projection / external acquisition visual-like content can have different origins

object identity / object state familiar objects can generate new events

novelty / promotion novelty is one allocation factor

perturbation / promotion not every change deserves expensive processing

current sensory field / internal state external situation and organism state are distinct

energy constraint / sensory information energy modifies allocation rather than defining the field

environmental change / map update the environment can change before persistent knowledge does

action / perception action can reveal the representation being used without reducing perception to action

biological promotion / FV focus artificial representation manipulation is not biological perceptual promotion

78. The full working architecture

                    CURRENT SENSORY FIELD

                              │

                              ▼

                   SPECIES-SPECIFIC ACCESS

                              │

                              ▼

                      CURRENT INPUT

                              │

                ┌──────────────┴──────────────┐

                │                                │

                ▼                                 ▼

      CURRENT SENSORY                PERSISTENT KNOWLEDGE

       COMPUTATION                             │

              │                         ┌───────┼────────┐

              │                         │                  │                     │

        structure               objects          maps      expectations

        motion                  methods   routes     cues

        surfaces               affordances      learned relations

        spatial relations     danger            social information

               │                    │

                    └───────────┬─────────┘

                           │

                            ▼

                 CANDIDATE GENERATION

                            │

                            ▼

               ALLOCATION / PROMOTION

                              ▲

                              │

       ┌─────────────────┼─────────────────┐

       │                       │                     │

   INTERNAL STATE     ENVIRONMENTAL     EXPECTATION /

   energy             STATISTICS        ANOMALY

   hormones              │              uncertainty

   needs                     │

   arousal               │

        └─────────────────┼─────────────────┘

                             │

                   ┌──────────┴──────────┐

                     │                        │

         not promoted            promoted

               │                     │

           background           deeper processing

                                     │

                     ┌───────────────┼───────────────┐

                     │                     │                     │

                   object         component           state

                         │               │               │

                  method        affordance       relation

                    │               │               │

                         └───────────────┼───────────────┘

                                     │

                            cross-modal / motor

                                       │

                                       ▼

                                  ACTION

                                       │

                                ┌──────────┴──────────┐

                          │                       │

                    field changes         learning /persistence

                                                │

                                                ▼

                                   future allocation changes

This is not a literal neural wiring diagram.

It is an architectural description of functional relationships.

79. The central loop

The core loop is not:

see → identify → remember.

It is closer to:

current sensory input → economical processing → candidate/change → promotion when warranted → deeper processing → action and/or persistence

while:

persistent knowledge + internal state + environmental statistics + current consequences

continuously influence what is worth processing.

At the same time:

current sensory input → useful structure → immediate action

can operate without requiring object identification or database retrieval first.

80. Familiar sensory fields

A familiar environment can feel visually rich without requiring detailed processing of everything.

Persistent spatial knowledge supplies:

structure.

Local spatial knowledge supplies:

relevant organization.

Current vision supplies:

present detail.

Expectations reduce the need to deeply inspect unchanged information.

Then:

object moved
object missing
unexpected movement
unexpected obstacle
familiar cue failed

can produce perturbation and possible promotion.

The efficient compromise is:

remember the structure, sample the present, investigate the exceptions.

81. Novel sensory fields

A genuinely unfamiliar sensory field reduces the usefulness of persistent spatial knowledge.

The organism therefore relies more heavily on:

direct current sensory structure

while still carrying:

• object knowledge;

• danger knowledge;

• inherited defaults;

• material familiarity;

• learned action routines;

• current state.

What it approaches first is therefore informative.

It may:

• ignore;

• avoid;

• inspect;

• touch;

• approach;

• manipulate.

The first voluntary investigation is a behavioural form of object promotion.

The chosen object can reveal something about:

• safety;

• interest;

• usefulness;

• familiarity;

• information value;

• energy/risk budget.

82. Novelty, risk and information value

Novelty is not danger.

Danger is not novelty.

Information value is not novelty.

A novel object can be:

safe + useful + worth investigating.

Another can be:

novel + risky + not worth approaching.

Another can be:

novel + low consequence + ignorable.

The allocation system therefore needs more than a novelty detector.

It needs some interaction among:

current state + knowledge + current sensory structure + anticipated consequence + processing cost.

The exact biological implementation remains an empirical question.

83. Persistence is an energetic architectural choice

Persistent knowledge is not free.

Durable information can require:

• storage;

• connectivity;

• retrieval;

• maintenance;

• updating;

• interference management.

Therefore persistence should be selective.

Location and landmark structure may persist.

Transient decorative detail may not.

A dramatic event may persist after one exposure.

A trivial object may disappear after repeated exposure.

Persistence is therefore not simply:

number of times seen.

It is closer to:

what the organism has reason and capacity to retain.

84. Vision and persistent knowledge are reciprocal

Persistent knowledge helps interpret current sensory information.

Current sensory encounters update persistent knowledge.

The loop is:

existing knowledge → interpretation/promotion

current encounter → learning

learning → modified future processing.

The system therefore changes itself through experience.

An animal that once investigated an object may later treat it as ordinary.

A child who learns a tool's affordances no longer needs to discover them from scratch.

A predator that learns a prey pattern can detect it more efficiently.

The organism is continually changing the machinery that determines what it will process next.

85. Vision is not “world creation”

The organism already has:

• memory;

• maps;

• objects;

• methods;

• expectations;

• constraints;

• social knowledge;

• inherited information.

Current vision supplies current information to that system.

But vision also extracts structure that has never been stored before.

Therefore neither extreme is sufficient.

Too little

Vision is just a camera feeding a database.

This fails on genuinely novel visual structure.

Too much

Vision reconstructs the entire world independently every moment.

This fails to explain the enormous efficiency of:

• persistent spatial knowledge;

• expectations;

• object knowledge;

• procedural behaviour;

• learned cue compression.

The useful architecture lies between them:

current sensory computation + persistent knowledge + selective promotion + internal-state-dependent allocation + continual recalibration.

86. What V2 adds to V1

V1 established the core architecture.

V2 closes the following loopholes.

V1: “board”

Retired.

V2:

current sensory field

This avoids collision with Memory Prism's board.

V1: lazy scan

Retained, but no longer interpreted as one literal serial scan.

V1: promotion

Retained, but explicitly separated from generic attention, novelty and recognition.

V1: object

Separated from candidate and from current object state.

V1: recognition

Separated from identification and image retrieval.

V1: persistent knowledge

Expanded to include cue compression and altered future processing.

V1: novelty

Expanded into novelty × state × risk × information value.

V1: current state

Expanded to explicitly include energetic and physiological constraints.

V1: board/environment adaptation

Expanded into:

stimulus-regime change → recalibration period.

V1: current visual structure

Expanded into dimensional, degraded, temporal and cross-modal cases.

V1: action

Expanded into motor geometry as an empirical readout of represented spatial structure.

V1: camouflage

Strengthened with detection/search evidence while reducing the claim that all camouflage literally means “avoid promotion.”

V1: social learning

Retained and made more explicit as an alteration of the information available to later explorers.

87. The V2 research status

Making Vision V2 is not presented as an existing published theory.

Its components have different evidential statuses.

Strong independent correspondences

The V1 research review identified strong matches to:

• affordance theory;

• direct perception;

• cognitive maps;

• Umwelt;

• change blindness;

• predictive processing;

• object-file theory;

• state-dependent exploration;

• distributed cognition;

• camouflage;

• sleep-dependent consolidation.

V2 research adds substantial independent evidence around

• object/scene interaction;

• object-based allocation;

• temporal object correspondence;

• visual adaptation;

• environmental-statistics adaptation;

• degraded and developing vision;

• object-action integration;

• camouflage effects;

• physiological-state modulation of visual attention;

• sleep replay and internal processing.

For example, contemporary object/scene research supports partly parallel processing and bidirectional scene-object interaction rather than a simple serial recognition pipeline.

Visual-adaptation research supports continual recalibration to current and recent stimulus environments, including changes in sensitivity and what captures attention.

Research on environmental statistics argues that sensory adaptation can reflect complex ecological regularities rather than only isolated stimulus features.

Camouflage has measurable effects on predator search and attack, supporting the functional importance of reducing detectability even though the exact mechanism differs by camouflage strategy.

Physiological state can modulate visual allocation: hunger, for example, has been shown to alter attentional responses to food cues, although results depend on the manipulation and task.

Making Vision synthesis

The following remain MV architectural constructs rather than established named biological mechanisms:

• current sensory field as the replacement for V1's board;

• promotion as a distinct allocation operation;

• maplet/local inset;

• the background → candidate → promotion funnel;

• the full selective-allocation architecture;

• energy/state/knowledge/field as interacting allocation variables;

• environmental editing as a deliberate future-promotion mechanism;

• the unified learning → changed promotion economy loop.

This distinction is intentional.

The supporting V1 research explicitly concluded that the integrated Making Vision architecture was a novel synthesis across multiple literatures rather than an existing published framework.

88. V2 research test families

The current V2 research programme therefore includes:

1. 2D/3D representation

2. degraded visual information

3. cue failure and substitution

4. temporal identity

5. unexpected behaviour

6. traffic-light state promotion

7. promotion under physiological state

8. novelty allocation

9. apparent depth → action

10. environmental remapping

11. environmental regime shift

12. cross-modal conflict

13. internal sensory events

14. sparse sensory-field completion

15. socially altered first-contact exploration

16. camouflage/detection

17. learning-induced promotion changes

These are not all equally mature.

Some have strong existing literatures.

Some have clear component evidence but require the MV combination to be tested.

Some are specifically Making Vision predictions.

89. The strongest V2 experimental principle

The most useful V2 experiments manipulate one source of information while holding another constant.

Examples:

Apparent depth

physical geometry constant
visual geometry changed

Cue compression

object constant
learned cue removed

Environmental remapping

organism constant
spatial arrangement changed

Behavioural perturbation

appearance constant
behaviour changed

Degraded vision

task constant
information resolution changed

Environmental recalibration

sensory system initially adapted
statistical regime changed

These designs allow the system to be interrogated rather than merely observed.

90. The deepest V2 claim

Being physically present is not the same as becoming perceptually important.

A chair can occupy the current sensory field without becoming an object of active processing.

A camouflaged animal can be physically present without becoming an actionable candidate.

A spilled patch of water can become important because it interrupts an expected route.

A familiar object can become a task reminder because a person deliberately places it where future sensory processing will promote it.

A traffic light can be completely familiar while its current state becomes highly important.

A CAPSLOCK-marked element can become a deliberately promoted unit in an artificial information stream.

A novel object can remain ignored because its information value is low or its exploration cost is high.

The organism therefore does not simply process everything equally.

It continuously allocates processing according to the interaction between:

current sensory information + persistent knowledge + internal state + environmental statistics + anticipated consequence.

91. The simplest statement of Making Vision V2

A biological visual system does not need to deeply understand everything it receives.

It needs to:

1. Acquire current sensory structure from the current sensory field.

2. Operate under incomplete and noisy sensory access.

3. Use existing knowledge where available.

4. Extract useful structure without requiring stable object identification.

5. Generate candidates from partial, spatial and temporal evidence.

6. Maintain expectations about recurring environmental structure.

7. Allocate processing unevenly.

8. Promote some candidates, states, regions or events for additional processing.

9. Use internal state, energy, physiological condition, needs and consequences to alter allocation.

10. Use direct sensory structure for action even without full classification.

11. Use cross-modal and motor information to test current representations.

12. Retain useful information selectively.

13. Allow learning to change future recognition, cue use and promotion thresholds.

14. Adapt processing economy to the statistical regime of the environment.

15. Recalibrate after changes in that regime.

16. Allow the environment itself to become part of the information-management loop.

17. Continue operating without requiring complete reconstruction of the world.

The result is neither a camera nor a database.

It is a:

selective, state-dependent, continuously recalibrating interface between current sensory structure and a persistent organismic system.

92. V2 boundary

Making Vision does not attempt to explain all cognition.

It does not attempt to fully explain:

• consciousness;

• all dreams;

• all motor control;

• all decision theory;

• pathology;

• every sensory illusion;

• every evolutionary adaptation;

• social learning as a complete field;

• distributed intelligence as a complete field;

• all forms of memory;

• every hormonal mechanism;

• every attentional mechanism.

Those systems intersect with vision.

They are not all the problem.

The V2 question is narrower:

How does an organism allocate limited processing across incomplete, changing, internally and externally generated sensory information when it already possesses persistent knowledge, spatial structure, expectations, internal state and action machinery, and when its processing economy itself changes with experience and environmental regime?

The current answer is:

By combining direct current sensory computation with persistent knowledge and spatial structure, generating and selectively promoting candidates, allocating processing according to expectation, perturbation, need, state, energy, consequence and learned relevance, using action and cross-modal information to test representations, and continuously recalibrating that allocation to the statistical structure of the organism's environment.

Making Vision version 1 

References

§10, 87 — Object/scene interaction

• Peelen, M.V. et al. (2023). Predictive processing of scenes and objects. Nature Reviews Psychology.

• Oliva, A., & Torralba, A. (2007). The role of context in object recognition. Trends in Cognitive Sciences, 11, 520–527.

• Bar, M. (2004). Visual objects in context. Nature Reviews Neuroscience, 5, 617–629.

§2 — Affordances / object knowledge as verbs

• Gibson, J.J. (1979). The Ecological Approach to Visual Perception.

§4 — Spatial knowledge as map, not photograph

• Tolman, E.C. (1948). Cognitive maps in rats and men. Psychological Review.

• O'Keefe, J., & Nadel, L. (1978). The Hippocampus as a Cognitive Map.

§6 — Species-relative accessible field (Umwelt)

• von Uexküll, J. (1934). A Foray into the Worlds of Animals and Humans.

§11, §24 — Candidates, perturbation, change blindness

• Rensink, R.A., O'Regan, J.K., & Clark, J.J. (1997). To see or not to see: The need for attention to perceive changes in scenes. Psychological Science.

§12, §38 — Object-file theory (recognition without full identification)

• Kahneman, D., Treisman, A., & Gibbs, B.J. (1992). The reviewing of object files: object-specific integration of information. Cognitive Psychology.

§15, §16 — Newly sighted / blurry-vision-first development (verified this session)

• Sinha, P., Chatterjee, G., Gandhi, T., & Kalia, A. (2013). Restoring Vision through "Project Prakash": The Opportunities for Merging Science and Service. PLOS Biology.

• Held, R., Ostrovsky, Y., de Gelder, B., Gandhi, T., Ganesh, S., Mathur, U., & Sinha, P. (2011). The newly sighted fail to match seen with felt. Nature Neuroscience, 14.

§31, §76 — Physiological state and visual allocation (hunger) (verified this session — note the correction)

• Piech, R.M., Pastorino, M.T., & Zald, D.H. (2010). All I saw was the cake. Hunger effects on attentional capture by visual food cues. Appetite.

• Note: two 2023 direct replications of this effect found no significant result, and a separate 2024 study found the reverse direction. Cite the original for the hypothesis; cite the replications as the honest caveat.

§41 — Predictive processing (used narrowly, not adopted wholesale)

• Friston, K. (2010). The free-energy principle: a unified brain theory? Nature Reviews Neuroscience.

§46 — Distributed cognition / environmental editing

• Clark, A., & Chalmers, D. (1998). The extended mind. Analysis, 58(1), 7–19.

§48 — Camouflage effectiveness (verified this session)

• de Alcantara Viana, J.V. et al. (2022). Predator responses to prey camouflage strategies: a meta-analysis. Proceedings of the Royal Society B.

§57 — Sleep replay / consolidation (verified this session)

• Wilson, M.A., & McNaughton, B.L. (1994). Reactivation of hippocampal ensemble memories during sleep. Science, 265, 676–679.

• Louie, K., & Wilson, M.A. (2001). Temporally structured replay of awake hippocampal ensemble activity during REM sleep. Neuron, 29, 145–156.

§75, §81 — Novelty allocation in children, caregiver presence (verified this session)

• Ainsworth, M.D.S., & Bell, S.M. (1970). Attachment, exploration, and separation: Illustrated by the behavior of one-year-olds in a strange situation. Child Development.

• Bowlby, J. (1988). A Secure Base: Parent-Child Attachment and Healthy Human Development.

§60 — Grip scaling from object knowledge

• Marked in the working document as intuitively grounded (banana vs. iron rod, cat judging a jump, diver entering water) rather than literature-dependent — left uncited by request.