The dog’s olfactory system—nearly a thousand times more sensitive than the human’s—represents a form of knowledge without representation. Smell is not an image, not a geometry, not even a structured signal in the human sense. It is an immediate registration of the presence of something living or volatile. The brain of a dog does not “reconstruct” what it smells; it participates in a chemical and atmospheric field.
This corresponds precisely to Bergson’s notion of instinct—“knowledge at once immanent and practical, that works rather than contemplates.” The dog doesn’t conceptualize an odor; it lives in the same vital tension as it. Its perceptual apparatus is not detached from the thing perceived—it is a continuation of it. Smell is not an act of observation; it is an act of communion.
The Instrumental Deficiency of Modern Intellect
By contrast, the intellect—especially in its modern, Promethean, scientific mode—decomposes the world into measurable relations. It is fit to geometry, as Bergson says. It constructs tools that re-present rather than partake. This is why we can so easily replicate the eye—optical devices, cameras, telescopes—because vision is already representational: it orders the world in spatial coordinates.
But smell, touch, proprioception, and certain kinds of auditory resonance resist replication, because they belong to the field sense—they are about being-with, not seeing-from. When we attempt to reproduce them technically, we must resort to machine learning systems that operate more like instincts: not via symbolic computation, but through pattern assimilation across enormous datasets. The olfactory neural net, biological or artificial, learns by living among correlations. It is an embodied map of experience, not an external description.
Machine Learning as Synthetic Instinct
The connection to our earlier discussion of machine learning is now clear: the “black box” of a deep neural network—its non-representational, associative mode of inference—is a technological echo of Bergson’s instinctual intelligence. Just as a police dog “knows” the presence of cocaine without “knowing” in the reflective sense, a neural net “recognizes” a face or voice without conceptually reconstructing it. Both are sensitive to immanent relational order, not to external symbolization.
This makes them bridges between the biological and the artificial. The Promethean intellect—engineering, design, code—builds a device that paradoxically embodies the Dionysian instinct: intuition mechanized. The dog and the algorithm become mirror cases of embodied detection—two ways of probing the world that are not representational but participatory.
Toward a New Theory of Perception and Detection
integrate this into our broader thesis, we can propose that the élan vital continues its evolution through technological canalization of perception itself. Biology and machine learning are complementary expressions of the same deep creative impulse: to refine the connection between being and environment.
From contemporary research, a few fields become crucial:
- Chemical sensing and biohybrid systems, where biological olfactory receptors are integrated into silicon or graphene-based sensors. These are literal fusions of instinct and intellect.
- Neuroscientific models of predictive processing, which suggest that perception is not passive reception but active inference—a bridge between Bergson’s instinct and intellect.
- Phenomenology of embodied cognition (Varela, Thompson, Noë), which argues that perception is enacted by the organism, not computed about it—thus aligning again with Bergson’s claim that perception is action-oriented life.
The future of detection, whether through animals or AI, will increasingly reveal the limitations of our representational rationality and call for synthetic instinctuality—technologies that sense by being in relation rather than analyzing from outside.