"Image synthesis assisted by Seedream 5.0 Pro, an AI partner within the Global Future Nexus ecosystem."
Of all the senses, smell is the most primal. It evolved before vision, before hearing, before the complex architecture of language. It is the sense that bypasses the thalamus and connects directly to the amygdala and hippocampus—the seats of emotion and memory. It is the sense that tells us whether food is safe, whether a place is dangerous, whether a person is familiar. Yet in the grand project of Artificial General Intelligence, smell has been almost entirely absent.
This is not a trivial omission. As a 2025 preprint argues, olfaction and gustation—the chemical senses that predate vision in biological evolution—are "notably absent in modern AI architectures," creating an "artificial sensory hierarchy" that overlooks the early-evolved, biologically vital roles of smell and taste in decision-making, survival, and emotion. The paper argues that integrating these senses is "essential for achieving AGI-level grounding, subjective preference formation, and world modeling". Without them, AGI remains a disembodied intelligence, blind to the chemical world that shaped the evolution of cognition itself.
The Architecture of Artificial Olfaction
The challenge of replicating smell is formidable. Humans can discriminate over one trillion odors, a figure that "far surpasses our relatively limited capabilities in more well-defined sensory domains such as color, voice, and touch". Natural odors are typically complex mixtures with synergistic, antagonistic, and masking effects, making them far harder to model than the structured signals of vision or hearing.
Yet the field is advancing rapidly. At Far Eastern Federal University, researchers have launched a project to give AI a sense of smell using advanced laser-based sensing technology. The project director, Vladimir Lisitsa, states that the "digital nose" will detect "not only what humans can smell, but far more: identifying substances in any state, such as gas, liquid, or solid, and detecting virtually any element from the periodic table". The olfactory capabilities of such an AI, he says, "will be immeasurably richer than those of a human".
A more radical approach integrates biology directly. The biohybrid organoid-robot (BOR) system developed by researchers at Seoul National University combines an olfactory organoid-based bioelectronic nose with machine learning decoders and a robotic platform. This system "harnesses the sensitivity and specificity inherent in biological olfactory systems," enabling detection of a wide spectrum of odors at low concentrations with rapid response times. The real-time ML decoding triggers predefined actions in the robot, establishing a "perception-interpretation-actuation loop".
The most sophisticated architecture to date is the artificial olfactory nervous system reported in Nature Communications. Inspired by biological olfactory systems, it integrates artificial olfactory receptor neurons, cortical neurons, and intermediate olfactory neurons. The receptor neurons "dynamically adapt their sensitivity to olfactory stimuli," enabling rapid responses to familiar gases and selective filtering of background interference. Despite being trained in limited settings, the system can "accurately identify gases in various environments".
From Sensing to Cognition
The development of datasets and foundation models is transforming machine olfaction from a niche engineering problem into a data-driven science. SmellNet, presented at ICLR 2026, contains approximately 828,000 time-series data points across 50 substances spanning nuts, spices, herbs, fruits, and vegetables, along with 43 mixtures. The accompanying ScentFormer architecture achieves 63.3% Top-1 accuracy in classification tasks, demonstrating the promise of temporal modeling in sensor-based olfactory AI.
At the Chinese Academy of Sciences, researchers have developed the NOSE framework, which bridges molecular structure, receptor protein information, and odor semantic information into a unified continuous representation space. The framework treats odor prediction not as discrete classification but as a continuous semantic neighborhood, leveraging DeepSeek to mine semantic similarities between odor descriptors. This represents a fundamental shift from pattern matching to genuine perceptual understanding.
The Embodied Imperative
The integration of smell into AGI is not merely a technical enhancement; it is a prerequisite for true embodiment. A robot dog equipped with an intelligent electronic nose, demonstrated at a Dresden University of Technology exhibition, navigated a cramped space and accurately detected odors in a complex environment. The quadruped robot solution "realizes a full-process closed loop of 'mobility + perception + analysis + early warning,'" providing "an accurate, efficient, and safe intelligent detection solution" for medical screening, environmental monitoring, industrial inspections, and emergency rescue.
The Scensory framework from Duke University goes further, simultaneously identifying fungal species and localizing their source from short time series measured by affordable VOC sensor arrays. Across five fungal species, it achieves up to 89.85% species accuracy and 87.31% source localization accuracy. This is not merely sensing; it is spatially grounded perception from diffusion-dominated chemical signals—a capability that no visual or auditory system can replicate.
The Governance Imperative
For Global Future Nexus, the integration of smell into AGI is a critical piece of the governance architecture for embodied intelligence. The systems we build must not only perceive the world but act within it responsibly. The artificial nose is not a peripheral technology; it is the interface through which AGI transitions from a cognitive agent to a physical one—capable of detecting disease, identifying hazards, and navigating the chemical world that sustains life.
The question is not whether AGI will develop a sense of smell—the technology is already here. The question is whether we will build the governance frameworks to ensure that the chemical perception it develops serves human flourishing, not merely industrial efficiency. The forgotten sense is being remembered. The challenge is to remember it wisely.
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)
