Относится к сборнику статей теори МВАП https://t.me/thinking_cycles
The classical view of the stepwise formation of a hierarchy of perceptual primitives in the neocortex follows a model of sequential maturation of neocortical structures (fornit.ru/460). In this model, the earliest such structure extracts elementary "atoms" of perception—such as points in the visual field, individual auditory sounds, etc. For visual perception, the simplest combinations include lines of varying orientation, thickness, and color; arcs and circles; dots of varying intensity. From all these diverse lines emerges a generalized image—Line; from circles—a generalized Circle. Generalized images like Square, Stripe, Surface, etc., also emerge. The simplest mechanism for such generalization is an “OR”-based convergence: each specific line, when activated, triggers a detector for the generalized image Line.
Based on these first-level primitives, as subsequent layers of detectors mature, more complex combinations arise and similarly form generalized categories. This process continues until highly complex, holistic recognizers emerge—such as those for “grandmother’s face” or “automobile.”
The Perceptual Tree Is Not the Neocortical Hierarchy of Perceptual Primitives
It is impossible to construct a tree of increasingly complex images (fornit.ru/66797) solely from the neocortical hierarchy of perceptual primitives, where the root would contain all first-level detectors, each branching into second-level detectors, and so on up to the most complex holistic images. This is because multiple distinct generalized images can appear simultaneously within a perceptual field.
Instead, the system must immediately isolate the single most behaviorally relevant final image from the entire perceptual scene—and only then does the entire ancestral branch of features that contributed to it become defined. This is precisely the primary function of the Orienting Reflex (OR) (fornit.ru/44471): among all currently active perceptual elements, it selects those with the highest significance and novelty, initiating their sustained retention for conscious processing.
Thus, the following work must be performed by pre-conscious attention mechanisms:
Development of the Orienting Reflex System
For each sensory modality, this selection occurs at the level of image detectors—not in the relatively newer neocortex, but in ancient sensory-processing structures that serve as precursors and foundations for the orienting reflex system (fornit.ru/68305).
For example, in frogs, the tectum receives visual input and constructs a kind of “salience map” of the surrounding world. Each point on the tectum’s rounded surface corresponds to a location in the space around the animal. The right tectum contains a precise map of the left eye’s visual field, and vice versa. When a small black dot moves chaotically near the frog, visual signals travel via the optic nerve to the tectum, which then triggers motor commands—resulting in the frog’s tongue striking with astonishing accuracy to catch a fly.
In salamanders, electrical stimulation of the tectum elicits complex, coordinated behaviors: the animal turns, opens its mouth, extends its tongue, and makes grasping motions with its long, thin fingers—as if capturing prey. Regardless of which spatial region the stimulated tectal neurons encode, the animal will reach toward that exact location.
In owls, the visual map in the tectum is integrated with an auditory map. During hunting, the owl can target prey either by sight or—when hunting at night—by sound alone.
All vertebrates use the tectum in roughly the same way, though many species have additional specializations. This brain region gathers sensory information, selects the most salient event in the environment, and physically orients the animal’s sensory organs toward it.
Evolution has hardwired mechanisms that prioritize certain stimuli: large objects occluding the background are prioritized, as are moving occlusions across the visual field.
For each fundamental behavioral mode (feeding, exploratory, sexual, defensive, etc.), the relevant stimulus is selected according to the adaptive purpose of that behavioral state. Thus, the significance of a detected object is tied to the homeostatic relevance of the currently active behavioral program. The appearance of such an object constitutes novelty relative to background perception, which lacks significant content. The concept of novelty evolves alongside awareness mechanisms, which are designed to process stimuli containing unfamiliar components under current conditions—ensuring that unexpected consequences or inappropriate reactions are not missed (fornit.ru/5214).
In summary: the most ancient mechanisms select what is most important in perception, while neocortical image recognizers form connections with concurrently active tectal elements, enabling pre-conscious selection of a final image (fornit.ru/70785) along with its entire hierarchical lineage of constituent features.
Thus, the perceptual tree is not defined by its initial primitives (of which many may be activated in parallel across different regions of the visual field), but rather by reflexive focusing—first by ancient subcortical structures, then voluntarily guided by conscious evaluation. The final image activated in tertiary (associative) cortex across all sensory modalities—and confirmed by the tectum—becomes the most relevant percept, along with its full ancestral branch. This image triggers hippocampal feedback (output routed back to input via structures linked to active behavioral programs), creating a self-sustaining loop of activity that supports processing by the frontal cortex’s single-channel priority attention system. As a result, the image’s significance is refined within the “understanding model” (fornit.ru/69260)—either confirming its priority or leading to its suppression, allowing the next most relevant image to take precedence.
The perceptual tree thus emerges as a structure optimized for efficient recognition and selection of the most relevant content. It is not a hierarchy of all activated elements across the entire sensory field; instead, relevance is first established by ancient pre-conscious attention mechanisms, and later adjusted based on conscious evaluation.
The Ancient Mechanism of Relevance Selection
The core of this process is the method of selecting the most relevant image from the entire perceptual field. This method relies on hereditary neural connections between combinations of features that are characteristic of behaviorally relevant stimuli under specific conditions and within specific behavioral contexts. At this level, there is no strict hierarchy (though detection mechanisms may evolve by building upon existing ones). Unlike the strictly logical, sequential organization of neocortical primitives, this resembles “patchwork programming”—layer upon layer of evolutionary fixes.
Consequently, the foundation of perceptual images is not geometric primitives like circles, lines, or dots—but rather “blobs” or proto-images: coarse, holistic representations characteristic of external-world objects. Thus, attention to “grandmother’s face” is not triggered by detecting a circle (for an eye), but by a characteristic silhouette—regardless of its exact form—recognized by its most general facial features. If we suddenly see “grandmother” in an inkblot, it is this ancient pre-selection mechanism that has been activated.
A “blob”—in the case of a running cat—is the most generalized, averaged contour of the body, recognizable by its key features as a small, fast-moving creature resembling a cat. Similarly, all initial perceptual images form the base nodes of the perceptual tree, representing the broadest category; subsequent nodes refine this categorization. It might not even be a generalized contour of a “running animal,” but something even more abstract—e.g., an elongated shape moving rapidly. This depends on the species-specific hereditary set of ancient features, and even individual variations within a species.
Therefore, the hierarchy in the perceptual tree originates not from the bottom up (primitives → whole), but from the top down—starting with the most general, yet behaviorally relevant image.
For example:
In a predator (e.g., an owl), the “blob” might be “thermal spot + rustling sound.”
In a herbivore: “sudden change in horizon silhouette.”
In humans: “face-like configuration”—even in clouds or paint stains.
This implies that when designing an artificial living system, one need not replicate the biological complexity of the orienting reflex and perceptual tree directly. Instead, one can engineer an optimally programmable structure that fulfills the same adaptive functionality—precisely what was achieved in the Beast adaptive individuality system.
Co-activation of Ancient Hereditary and Neocortical Systems
The overall picture is as follows: although numerous image detectors across all sensory modalities are activated in the neocortex at any moment, the orienting reflex selects only one final image—the most relevant among them. Ancient structures do not drive neocortical activation; they merely select the most important image for sustained processing. This selection is implemented via a hippocampal switch that routes the image’s output back to its input and connects it to the frontal lobes (per A. Ivanitsky).
Clarification: The perceptual tree is not constructed top-down, but emerges from the neocortical hierarchy of perceptual primitives in conjunction with ancient feature detectors. The resulting tree is thus an inter-structural formation, shaped during ontogenesis by Hebbian learning: “neurons that fire together, wire together.”
Ancient structures (tectum / superior colliculi), processing the same sensory input, extract the most significant holistic pattern—the “blob.” This pattern corresponds to an existing high-level neocortical image (if one exists). Synaptic connections are strengthened between co-active tectal and neocortical neurons across all hierarchical levels involved in constructing that image.
This explains why we instantly recognize “grandmother” without first analyzing her nose, then eyes, then hairstyle—the entire hierarchical branch is already linked to the tectal “blob” and fires as a unified whole.
This is the key distinction of the proposed model from both classical hierarchical models (Hubel–Wiesel) and purely global approaches (Navon).
The New Quality of Novelty
If the highest neocortical level remains inactive—because the OR-selected image components are so novel that they fail to activate the associative cortex—this signals high novelty (fornit.ru/68852). Unlike older forms of novelty detection (e.g., mere sensory change), this semantic novelty influences the more recently evolved OR mechanisms that develop during ontogenesis.
Thus, novelty shifts from being purely sensory (“something changed”) to semantic (“this doesn’t fit my categories”). This allows the system to focus attention not just on the brightest stimulus, but on the most informative one—initiating learning by holding the unrecognized image in the hippocampus for consolidation and enabling future adaptation through the formation of new recognizers.
Modern neuroscience confirms powerful ascending (subcortical → cortical) and descending (cortical → subcortical) pathways. For instance, the pulvinar and superior colliculus modulate cortical activity via the thalamus. Hence, ancient structures exert not only selective but also regulatory influence.
Supporting Research Evidence
The principle that holistic perception precedes detailed analysis—the “global-before-local” phenomenon—is well-established in psychophysiology and cognitive science:
“The superior colliculi participate in reflexive orienting to sudden visual and auditory stimuli… they direct head and eye movements toward novelty even without cortical control.”
This directly confirms that ancient structures detect objects as unified wholes, without decomposing them into primitives. Each such object is treated as a unique entity (analogous to an ID in software or a specific neuronal output channel in biology).
Selection of One Image Among Many: “Narrow Attention” and the Role of the Hippocampus
The mechanism of selecting a single image from many active representations—i.e., “narrow attention”—and the hippocampus’s role in this process were described by A.M. Ivanitsky:
Thus, the mechanism of singular image selection is directly supported by Ivanitsky’s work.
Prior to this article, there was no sufficiently integrated and concrete account of how the orienting reflex develops functionally and how it interconnects with the ontogenetic formation of the neocortical image hierarchy. Although the Beast system already attempted to implement categorical structuring within the perceptual tree, this theoretical clarification now enables more accurate design of subsequent artificial life systems—such as Isida—bringing them closer to biologically grounded logic.