Beyond the five: why AGI must learn to sense the invisible

"Image synthesis assisted by Gemini 3.1 Flash Image (Nano Banana 2), an AI partner within the Global Future Nexus ecosystem."

Human perception is a narrow window. We see a sliver of the electromagnetic spectrum, hear a fraction of acoustic frequencies, and remain blind to the magnetic fields that guide a migrating bird across an ocean or the electric fields that let a shark hunt in murky water. For decades, artificial intelligence has been built in our image, trained on our data, and evaluated against our perceptual limits. That era is ending. A quiet revolution in machine sensing is expanding the sensory repertoire of intelligent systems far beyond the human five, and in doing so, it is forcing a fundamental rethinking of what general intelligence actually requires.

The Limits of the Anthropocentric Sensorium

The dominant paradigm in AI has been vision and language. As a 2026 arXiv paper on generalist open-world temporal perception notes, human perception is "selective, active, embodied, and task-dependent," but "other species exceed humans along particular sensory dimensions and inhabit different perceptual worlds shaped by their sensors, bodies, niches, and possible actions". The paper argues that perceptual intelligence is "perceiver-relative," and that a general intelligence must be evaluated not by human benchmarks alone, but by its capacity to extract behaviorally useful structure from evidence that no human can directly access.

This is not a philosophical abstraction. It is a practical engineering challenge.

Sensing the Invisible: Magnetoreception and Electroreception

Consider magnetoreception—the ability to sense Earth's magnetic field for navigation. Many animals solve this problem despite the magnetic sense operating at an "extremely low signal-to-noise ratio". A 2025 study in the Journal of the Royal Society Interface explored neuromorphic encoding strategies that may underlie this ability, demonstrating that systems composed of as few as eight receptors and tens of neurons can accurately encode noisy magnetic information. The neural architecture based on the arthropod central complex was found to be "particularly robust" to this noise. For AGI, this offers a blueprint for robust spatial intelligence in environments where GPS is unavailable or compromised.

Electroreception is equally promising. In March 2025, researchers published in Science Advances the development of an organic electrosense transistor inspired by electrogenic fish—species that use electric fields to navigate and hunt in "low-visibility conditions, like dark ocean or muddy riverbeds". Unlike conventional sensors, this device perceives bipolar electric fields with high sensitivity and stability, and robots equipped with it "can successfully navigate and detect concealed objects without requiring direct contact". The applications range from search and rescue to surveillance in environments where visual and tactile senses are useless.

Biosonar and the Geometry of Sound

Echolocation represents another frontier. Bats, dolphins, and whales navigate complex environments by emitting sound and processing the returning echoes—a biosonar system that conventional engineering has never matched. A 2025 IEEE paper argues that convolutional neural networks, which revolutionized computer vision, can be extended to echolocation-based perception, but with a crucial insight: biological systems learn the acoustic representation of new objects from "significantly fewer echoes" than conventional ANNs, suggesting that these networks should have "very limited depth" and process information in a "highly parallelized" manner. This points toward AGI architectures that are not merely larger, but structurally different—closer to the frugal, sequential learning of biological brains.

The Internal Senses: Proprioception and Interoception

Beyond the external modalities lies an even more fundamental gap. Proprioception—the sense of one's own body in space—is essential for any embodied intelligence. A 2025 Nature Communications paper introduced AI-embodied flexible electronic robots that implement both proprioception (shape and attitude) and exteroception (vision, temperature, humidity, proximity, and pathway shape recognition) under dynamic conditions. Onboard computational units enable "autonomous behaviors like hazard evasion and thermal gradient tracking" through adaptive decision-making. This is not merely sensing the world; it is sensing oneself in the world.

Interoception—the sense of internal states—remains almost entirely absent from artificial systems. Yet it is fundamental to biological intelligence, underpinning emotion, decision-making, and homeostasis. For an AGI to develop genuine general intelligence, it may need an analog of interoception: a continuous awareness of its own operational state, energy levels, and internal constraints.

Multisensory Integration and the Next Generation of Foundation Models

The most promising development is the integration of these non-human modalities into large-scale foundation models. HoloLLM, presented at NeurIPS 2025, is a multimodal large language model that integrates LiDAR, infrared, mmWave radar, and WiFi to enable "seamless human perception and reasoning across heterogeneous environments". The system addresses two key challenges: the scarcity of aligned modality-text data and the heterogeneity of physical signal representations. The result is a 30% improvement in language-grounded human sensing accuracy.

This is the architecture of a general intelligence that does not privilege vision. It is a system that can reason across whatever sensory evidence is available, using language as the unifying medium.

The Governance of Expanded Perception

For Global Future Nexus, the expansion of AGI sensing beyond the human five carries profound governance implications. A system that can sense magnetic fields, electric currents, and acoustic geometries can perceive what humans cannot—and that asymmetry creates new forms of power. The same electroreception that enables a robot to find a buried survivor in a collapsed building could be turned toward surveillance, profiling, and control.

The 2026 arXiv paper on perceptual intelligence warns that "perceptual superintelligence should therefore be understood both in relation to broader intelligence—as a developmental and epistemic foundation and a continuing partner—and as a capability domain in its own right". This means that governance frameworks must address not just what AGI knows, but what it can perceive—and whether that perception serves human flourishing or escapes human accountability.

The path forward requires a shift from an anthropocentric sensorium to a multi-species, multi-physics architecture of perception. The intelligence we build must learn to see the invisible, hear the silent, and feel the imperceptible. The question is not whether AGI will develop senses beyond our own—it already is. The question is whether we will build the governance frameworks to ensure that the expanded perceptual world it inhabits remains one we can understand, and one we can steer.

Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)

Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)