Agnosiophobia in Lenia

• Agnosiophobia is an emergent behavior in Lenia where virtual organisms reorient away from regions lacking sensory input, avoiding informational occlusion.
• The phenomenon is quantified by measuring trajectory adjustments, recovery time, and survival rates when creatures encounter masked sensory regions.
• It serves as a conceptual counterpoint to infotaxis, emphasizing how renormalized convolution and attractor dynamics drive avoidance of unknown areas.

Searching arXiv for the specified paper to ground the article in the cited source. Agnosiophobia, in the sense introduced in Lenia research, denotes an emergent behavioral tendency of virtual agents to reorient away from regions of the environment from which no sensory information is available. In this usage, the term does not refer to human psychiatric illness; it names a specific artificial-life phenomenon: “fear of the unknown,” operationalized as avoidance of informationally occluded regions by self-organizing virtual organisms. The concept was formulated in the study of motile Lenia creatures, where some species turn away from blind zones despite the absence of any explicit avoidance policy, internal representation of danger, or evolved controller for the task (Cool et al., 29 May 2026).

1. Conceptual definition and scope

In the relevant Lenia study, agnosiophobia is defined behaviorally rather than introspectively. The operative phenomenon is reorientation away from regions that have been rendered unreadable to a creature’s sensing machinery. More precisely, when parts of the environment are made inaccessible to perception, some creatures turn away from those regions even though the environment contains no physical barrier and the creatures possess no explicit policy for avoidance (Cool et al., 29 May 2026).

This definition is narrower than the ordinary lexical phrase “fear of the unknown.” In Lenia, the “unknown” is instantiated as informational occlusion rather than semantic uncertainty or symbolic ignorance. The behavior is therefore not treated as evidence of an internal subjective state. Instead, evidence consists of externally measurable signatures: turning away from the perturbed flank, edge-following that prevents entry into blind regions, greater survival when asymmetrical early perturbations allow redirection, and species-specific trajectory changes that systematically keep creatures out of information-occluded space.

The term is also positioned as conceptually complementary to infotaxis. Infotaxis conventionally concerns movement guided by information gradients or information-seeking to reduce uncertainty. Agnosiophobia, by contrast, is the inverse tendency: movement away from regions that deprive the agent of information. The paper’s stronger claim is not merely lexical symmetry with infotaxis, but that informational topology can shape behavior in embodied excitable media as strongly as tangible geometry.

2. Lenia substrate and formalization of informational occlusion

Lenia is a continuous cellular automaton defined on a toroidal grid. At time $t$, the state is a grid $A_t$ with cell values in $[0,1]$. Standard Lenia updates each cell by convolving the state with a radially symmetric kernel $K$, producing a potential $U_t$, and then passing that potential through a growth function $G$. The update rule is

$$A_{t+1} = \text{clip}_{[0,1]}\bigl(A_t + \Delta t \cdot G(U_t)\bigr).$$

Here, $K$, $G$, and $\Delta t$ define the ruleset. In the agnosiophobia framework, the kernel is interpreted as the creature’s sensory field: it specifies what each cell “sees” and how strongly that information contributes to the update (Cool et al., 29 May 2026).

The intervention used to induce agnosiophobia is informational rather than mechanical. The study does not introduce physical obstacles; instead, it creates regions from which no information reaches the creature. These occlusions are represented by a binary mask $A_t$0, where $A_t$1 denotes an occluded cell and $A_t$2 a visible one. The critical technical point is that occluded cells are not treated as ordinary zero-valued cells, because zero would still constitute informative input. Rather, masked cells are excluded from sensing and the kernel is renormalized over the visible remainder:

$A_t$3

The numerator removes masked contributions before convolution, and the denominator renormalizes by the visible kernel mass. This is described as a minimal intervention that makes creatures blind to selected regions while proportionally amplifying the influence of the remaining visible information. When no occlusion overlaps the kernel footprint, the expression reduces to the ordinary Lenia update.

A central consequence follows from the finite kernel radius $A_t$4. A creature need not physically enter a blind region to be affected by it. Because each active pixel senses through a kernel extending a radial distance $A_t$5, course corrections can begin as soon as the sensory footprint overlaps an occluded zone. This makes informational occlusion functionally obstacle-like without being materially impassable.

3. Experimental protocol and quantitative criteria

The study used two complementary experimental manipulations. The first concerned whole-environment behavior. Creatures were placed in 10 environments containing black occluded regions that varied in size, sparsity, convexity/concavity, and slope. One emphasized example was the “guidelines” environment, selected because it visually exposes many approach angles and resulting trajectories. In these trials, each creature started at the center of the environment and was launched in 360 different orientations, with each run lasting 2000 time steps (Cool et al., 29 May 2026).

The second manipulation targeted mechanism at the level of the body. For each nonzero pixel in a creature’s body, the authors centered a persistent $A_t$6 occluded patch on that location and recorded the outcome. This yielded per-body sensitivity maps identifying where occlusion caused death, ordinary recovery, large distortion, or heading change. These maps were computed across multiple orientations and reported as consistent across headings.

Because Lenia creatures are dynamic, fluctuating patterns rather than static templates, the paper introduced a morphology-based identity criterion. For each grid state, a profile is formed by flattening the grid, sorting nonzero values in descending order, and trimming or zero-padding to a common length $A_t$7. Distance between two states $A_t$8 and $A_t$9 is then defined as the Wasserstein-1 distance between their profiles:

$[0,1]$0

This metric intentionally ignores exact position and orientation and instead captures morphology. For each species, a dataset of 5400 snapshots was constructed from 90 orientations, each run for 600 time steps. From these, a barycenter $[0,1]$1 was computed via the elementwise median of profiles. The maximum observed distance from $[0,1]$2 among canonical examples defined $[0,1]$3. The creature’s neighborhood was then the set of states within $[0,1]$4 of $[0,1]$5. A perturbed creature was considered recovered at time $[0,1]$6 if the average distance over the last $[0,1]$7 frames stayed below $[0,1]$8. It was considered dead if total mass dropped below $[0,1]$9. Otherwise the trajectory could be classified as explosion or metamorphosis.

Behavior was quantified by percentage of simulation time survived in environment tests and, for targeted perturbations, by frames until recovery, maximum morphological distortion along the recovery path, and heading change computed from mass-weighted centroid paths before perturbation and after recovery. In trajectory plots, the fraction of kernel area occluded across all nonzero pixels was also color-coded.

4. Species-specific behavioral patterns

The main behavioral result is that three of the four tested creatures—O2u, K4s, and K6s—show agnosiophobia, whereas S1s does not (Cool et al., 29 May 2026). The phenomenon, however, is not uniform across species.

Creature	Agnosiophobia status	Characteristic response
O2u	Positive	Clean turning away from asymmetrical occlusion
K4s	Positive	Edge-skirting, reversals, and relaunches
K6s	Positive, weakest	Long edge-following without immediate turning
S1s	Negative	Too fragile to sustain meaningful navigational response

O2u, the Orbium, is the clearest case. Its trajectory remains approximately straight until the sensory field on one flank overlaps a blind region; for example, if the front-right flank senses occlusion, the creature turns left, with stronger turning as approach continues. The paper emphasizes that O2u redirects most effectively under asymmetrical encounters rather than direct head-on ones. Across the 10 environments, O2u had the best average survival and the strongest avoidance pattern.

K4s also avoids blind regions, but by a different dynamical style. It often skirts the edge of an occluded region and, upon reaching the end of that region, launches away in a new direction. In some frontal encounters it collapses into a transient oscillating form and then “re-emits” in the opposite direction. In this species, agnosiophobia can therefore include marked internal reorganization before redirection.

K6s provides the weakest positive case. It often does not immediately turn away. Instead, it can travel along the edge of an occluded region for extended intervals, sustaining substantial occlusion while remaining viable. This edge-skimming is still treated as agnosiophobic because it prevents penetration into blind zones and eventually redirects movement, but it is qualitatively less decisive than O2u’s turning.

S1s differs by fragility rather than by a contrary navigational strategy. It rarely survives occlusion long enough to display a robust redirection pattern. This makes absence of observed agnosiophobia inseparable from limited resilience under the tested perturbations.

5. Mechanistic explanation: sensitivity structure and basin geometry

The paper’s explanation is explicitly mechanistic and dynamical rather than representational. Sensitivity maps show that each body has a spatially structured vulnerability profile. Across all four creatures, perturbations on the right flank tend to induce leftward reorientation and vice versa. More importantly, zones that trigger reorientation lie adjacent to zones that are lethal or non-recoverable, especially near the front of the moving creature (Cool et al., 29 May 2026).

For O2u, lethal perturbations cluster at the center of the leading edge and in the core, whereas nearby frontal-flank regions can trigger large heading changes, including values near $K$0. This arrangement creates what the paper effectively treats as a buffer. As the creature approaches an occlusion, its sensory field first overlaps reorientation-sensitive regions before the body enters lethal territory, so turning can occur before irreversible damage.

S1s lacks such a buffer. Much of its front is directly lethal, leaving little opportunity for safe course correction. K4s exhibits reorientation zones in front of sparse lethal regions, consistent with strong but sometimes abrupt reversals. K6s has no lethal zones at the tested perturbation scale; in the authors’ phrasing, it is “never pushed close enough to a lethal boundary for significant reorientation to occur,” which matches its edge-skimming behavior.

These observations are interpreted in a dynamical-systems framework. A creature is treated as an attractor in the high-dimensional state space of all possible Lenia grids. Because the rules are translationally and rotationally symmetric, the attractor is not a single point but a low-dimensional manifold of states corresponding to different positions and headings. Heading is therefore a free variable: a creature can change heading while remaining the same creature in morphological terms. The attractor’s basin of attraction is the set of perturbed states that eventually relax back to creature-like form. For a specific perturbation class, the paper defines a “cognitive basin”: the slice of the basin reachable by that perturbation type, here targeted occlusions.

Within this framework, targeted-occlusion maps reveal basin geometry. Quiet zones correspond to perturbations deep inside the basin, where recovery is rapid, distortion is small, and heading change is minimal. Lethal zones correspond to perturbations that eject the system from the basin. Between these are perturbations that generate long, distorted recovery trajectories near the basin boundary; these are precisely the perturbations that produce major heading shifts. The authors state this relationship in strong form: “No significant reorientation results from quick, undistorted recoveries.” Their recovery-trajectory scatterplot reportedly leaves the lower-right region—large heading change with little time or distortion—essentially empty.

6. Partial equifinality, relation to infotaxis, and limitations

The study’s central interpretive claim is that these creatures do not avoid blind regions because they explicitly seek information. Rather, occlusion perturbs morphology-preserving dynamics, and the system can often preserve itself most effectively by changing heading. In the paper’s formulation, creatures “take advantage of their freedom to change heading in order to achieve what appears to be a more fundamental goal: the preservation of their morphology” (Cool et al., 29 May 2026).

This is the basis of what the authors call partial equifinality. The creature is equifinal with respect to morphology but not with respect to heading. Many perturbed states return to the same creature form, yet with different post-recovery headings. That non-equifinality in heading is what yields navigational competence. A plausible implication is that behavior that appears purposive at the trajectory level can arise from recovery geometry in the state space rather than from explicit decision variables.

The relation to infotaxis is therefore nuanced. Agnosiophobia is complementary to infotaxis in that it produces movement away from informational deprivation rather than toward informative cues. But the underlying mechanism differs from explicit information-seeking formulations. Here the avoidance of ignorance is emergent, produced by the interaction between sensory occlusion and morphology-preserving basin dynamics. The broader implication drawn by the paper is that the informational topography of an environment can shape behavior as powerfully as its tangible topography.

The work also connects the phenomenon to spontaneous symmetry breaking and criticality. Lenia’s rules are symmetric under translation and rotation, whereas any individual creature occupies a specific position and heading, making it a symmetry-broken state. Heading is a soft direction on the attractor manifold. Small perturbations can move the system slightly along this direction, while stronger near-boundary perturbations produce larger shifts accompanied by increased recovery time and distortion, qualitatively suggestive of critical slowing down. The paper explicitly notes that this phase-transition analogy is qualitative rather than a complete thermodynamic theory.

Several caveats delimit the scope of the result. The study examined only four non-oscillating motile creatures, under one perturbation type, at one spatial scale, across ten environments. It does not yet explain why some rulesets couple perturbations to heading change while others do not, nor does it provide a predictive criterion from ruleset alone for whether a creature will display agnosiophobia. The authors also note that direct characterization of full basin boundaries is computationally difficult in Lenia’s high-dimensional continuous state space. The work is therefore framed as an initial roadmap rather than a finished general theory.

Taken together, these results establish agnosiophobia in Lenia as avoidance of informationally unknown regions by emergent virtual organisms. In the reported experiments, this avoidance is measured through masked, renormalized convolutional sensing and quantified through survival, recovery, distortion, and heading change. Its proposed cause is not a specialized fear module, but the geometry of attractors in a self-maintaining excitable medium, where heading serves as a free variable through which morphology can be preserved under asymmetric informational stress.