Learning
another
direction.

Abstract

Human beings developed in a world represented through three spatial dimensions, and higher-dimensional geometry is consequently treated as something that can be manipulated symbolically but not inhabited intuitively. Existing evidence weakens the assumption that this limit is fixed. People have learned path integration in computer-generated four-dimensional mazes; made visual judgments involving a fourth coordinate; and developed non-verbal comprehension through interactive hypercube environments. Adjacent work shows that adults can incorporate persistent artificial direction signals, extend spatial perception through sensory substitution, and recover binocular depth perception through training.

We propose a stronger experiment. A native OpenXR system assigns objects an independently manipulable coordinate w and translates that coordinate into controlled relations between the two eyes, temporal history, chromatic structure, transparency, or localized coherence. Participants do not merely observe projections. They reach, rotate, sort, collide, navigate, and predict within a world whose lawful consequences depend on w. An adaptive model searches the space of encodings while behavioral transfer, comfort, rendering integrity, and phenomenology constrain retention.

The central question is whether a stable sensorimotor law can make an additional world variable function as a spatial-like primitive. We distinguish cue discrimination, task learning, cross-object generalization, geometrical competence, cue-independent transfer, automaticity, and subjective perceptual recoding. The programme does not assume that humans will literally see Euclidean four-space. It asks whether virtual reality can become an instrument for constructing new forms of mathematical intuition.

1. From representation to inhabitation

Most higher-dimensional visualization compresses a 4D object into a 3D projection, then compresses that projection again onto a flat display. The observer learns to interpret the result. This can support understanding, but the fourth coordinate remains something the image refers to rather than something action directly encounters.

ELSEWHERE changes the experimental object. The world itself contains w. Two objects can coincide in visible coordinates yet remain separated in the artificial coordinate. A surface can be unreachable until the user moves through w. A rotation through an XW plane can transform what is connected, occluded, or collidable. The body therefore receives consequences that cannot be explained by ordinary depth alone.

PERCEPTUAL INVITATION

You reach into two translucent forms occupying the same visible location. Your hand passes through one and stops against the other. You perform a movement with no physical counterpart. Resistance changes hands. After enough encounters, you no longer translate the cue. You simply reach correctly.

2. The empirical opening

Aflalo and Graziano trained participants in a virtual maze with four spatial dimensions. Path-integration performance showed that participants learned enough of its geometry to point back toward an occluded origin. Ambinder and colleagues found that judgments of lengths and angles incorporated information from both the 3D projection and the fourth dimension; their analysis favored visual imagery over algebraic calculation, while emphasizing that the intuition was primitive and short-lived. Collins, Regenbrecht, and Langlotz later examined interactive hypercube learning and reported a distinct contribution from embodied interaction alongside theoretical expertise.

These results do not establish a new sense. They establish that human spatial operations are less rigidly bound to familiar three-dimensional frameworks than a simple evolutionary argument would suggest.

Sensory augmentation supplies the second opening. With the feelSpace belt, magnetic north is continuously mapped to vibrotactile stimulation around the waist. Across long training, the signal can change from an isolated vibration into spatial information that alters navigation strategies and subjective orientation. Controlled work has also found improvements in cardinal and survey knowledge. Visual-to-auditory substitution has made normally inaccessible rear space behaviorally localizable. These studies show that a stable action–sensation law can become part of spatial competence.

Adult binocular vision supplies the third opening. Perceptual-learning studies have recovered partial stereopsis in adults with abnormal binocular histories, and VR protocols combining cue scaffolding, dichoptic presentation, large disparities, and perception-for-action have produced transfer beyond the trained game. The binocular system is specialized, but not necessarily immovable.

Action matters specifically. In a VR “bug-squashing” programme, participants learned to use stereoscopic information while reaching toward and striking virtual targets; improvement transferred to stereoacuity measures outside the game. The relevant lesson is not that any VR exercise repairs vision, but that perception can be recalibrated through consequences that couple seeing to doing.

A parallel hyperspace research programme has explicitly proposed computer-generated higher-dimensional environments as instruments for studying perception, cognition, action, learning, and brain plasticity. ELSEWHERE turns that broad possibility into a narrower experimental contract: one independently controlled coordinate, several bounded encodings, and transfer tests capable of distinguishing a learned display trick from an acquired geometrical relation.

3. The proposed perceptual compiler

Let the world state be F(x,y,z,w,t). A modality is not w itself. It is a parameterized translation from local values and gradients of w into sensory differences and action consequences. Candidate families include chromatic hyperslicing, binocular phase, transparency mismatch, localized shimmer coherence, temporal history, texture-band separation, and weak hybrids.

The search process must include the person:

1. ProposeA model generates a bounded encoding and predicts likely conflicts with fusion, depth, flicker, and performance.
2. RegressThe renderer is verified independently: frame deadline, no crash, no GL error, exact modality parameters.
3. ExposeThe cue ramps from neutral to weak signal. One modality remains fixed for the complete trial.
4. ActThe participant sorts, reaches, navigates, rotates, avoids, or predicts according to w.
5. MeasureAccuracy, latency, trajectory, confidence, transfer, comfort, thermals, and rendering evidence are recorded.
6. Retain or killOnly encodings that survive behavioral and comfort gates enter longer training.

The model can detect image differences. It cannot infer human viability from those differences alone. Human learning is the objective function that the rendering search cannot simulate away.

4. A ladder of claims

Subjective reports matter most at the final levels, but they cannot substitute for behavioral transfer. Conversely, behavioral success alone does not tell us what the participant experiences. The research object must preserve both without collapsing one into the other.

Research distinguishing cue-dependent from cue-invariant visual learning makes Level 5 especially important. Success with one shimmer, color, or phase code may remain tied to that code. Faster acquisition or immediate transfer when the carrier changes but the underlying w-relations remain constant would be stronger evidence that the participant learned the latent relation rather than its first display.

5. Why wonder belongs inside the apparatus

Awe is often treated as a publicity effect added after scientific work. Here it is part of the generative environment and a measurable variable. The prospect of inhabiting a geometry that cannot occur in physical space may increase curiosity, persistence, and the willingness to accommodate unfamiliar relations. It may also increase suggestibility and false confidence. Both possibilities belong in the protocol.

POSSIBILITY FIELD

Imagine a mathematical culture in which two researchers enter the same executable geometry. One performs a transformation. The other immediately says: “No—you folded through the wrong side.” Experience arrives first; vocabulary follows; proof remains the final compression.

Falsification governs retention. It need not govern conception. The paper therefore uses four visibly distinct registers: evidential record, possibility field, perceptual invitation, and killing test. Wonder is not smuggled in as evidence, and evidence is not forced to pretend that the possible world is emotionally irrelevant.

6. Mathematical consequences

Mathematicians already use spatial intuition to generate conjectures, detect impossible constructions, and compress chains of reasoning. In higher-dimensional work, that support is usually indirect: notation, slices, analogies, and projections. If a trainable interface supplied even a limited embodied intuition, it could change what happens before proof.

Candidate applications include anticipating rotations through XW, YW, and ZW planes; recognizing connectivity outside the current slice; predicting cross-sections; navigating configuration and phase spaces; interpreting high-dimensional optimization landscapes; inspecting molecular conformational spaces; and exploring latent representations without reducing every relation to a chart.

The strongest outcome would not be “seeing a tesseract.” It would be acquiring reliable expectations about transformations that currently impose a heavy symbolic load. The intuition could be wrong, just as ordinary visual intuition is wrong. Its value would be measured by novel-problem performance, not by intensity of experience.

7. Experimental constitution

Every candidate modality is versioned and hash-bound. Renderer regression and human trial are separate protocols. A neutral baseline and immediate abort are always available. Cue amplitude is ramped, not surprised. One modality remains fixed within a trial. Eye strain, headache, nausea, fusion difficulty, and thermal state are recorded beside behavioral results. Unrelated device settings and applications remain untouched.

The first campaign should attach one scalar w to otherwise identical objects. It should compare several weak modality families, identify learnable and comfortable regions, and test delayed retention. Only after above-chance transfer should the project attempt full four-dimensional navigation or tesseract manipulation.

BOUNDARY

A beautiful shader is not a perceptual result. A compelling first-person report is not geometrical transfer. A classifier that decodes w from two images says almost nothing about whether a person can inhabit the mapping. Each layer must earn its own claim.

8. What success would mean

If the strongest claims survive, the dimensionality of human experience would appear less fixed than usually assumed. Receptors would remain biological channels, but experienced structure would be partly determined by stable relations among sensation, movement, and consequence. Virtual reality would become more than a simulator of possible scenes. It would become an instrument for constructing lawful worlds and training nervous systems to inhabit their regularities.

The extra modality need not represent literal Euclidean four-space forever. It could encode magnetic fields, fluid topology, uncertainty, network connectivity, molecular configuration, medical volumes, or another agent’s predicted attention. Instead of translating these into colors and legends, a person might learn them as navigable extensions of a world.

The research question is therefore larger than one headset or one encoding:

Are the apparent dimensions of human thought fixed properties of the brain—or learned consequences of the lawful worlds in which a brain is allowed to act?