Olfaction and The Artificial Unconscious

tangled-earbudsInstructions: Find a tangle of knots. It should not be too hard – earbud wires seem to have a thing for breaking-down into the tangles of entropic disorder. Next, untangle the knots – but not with such direct intent. Do not find one end of the wire and needle it in reverse, or spot a cluster and pull at it. Instead, take that messy knot of wires, and stick your fingers into it, all of them, right into the middle of it. Now begin to move your hands in and out, while wiggling your fingers. As if the knot of wires is a blob of matter and you are massaging it, let your fingers dance with the knot, grabbing, pulling, looping, crossing. It is not efficient, and it does not take advantage of that human capacity for self-directed behavior, but it works. If you let your fingers do the thinking, the knot will eventually untie itself. And you will have seen the unconscious mind in action. This is the act of odor object recognition. It is not a record, linear, of every chemical present. The signal-to-noise ratio for chemosensation is too complex. It is a whirling chaos of switching and regrouping that comes up with the answer in a way that makes absolutely no logical sense whatsoever. In fact, it is here that the seeds of logic are sown, and so, only after the answer has been found, the fruit.

1151px-Olfactory_system.svgOlfaction is an entirely “unconscious” process. Try as we may, we cannot exercise our olfactive powers at will. We cannot “imagine” smells; they must be exogenously stimulated. We cannot not smell, lest we not breathe. The more we try to smell something, the less we smell of it (because of its high attenuation rate). To smell is a cognitive asymmetry, taking place on one side of the brain only. Olfaction performs its Herculean feat of pattern recognition and limbic response completely separate from the logic parts of the brain. If the crosstalk between hemispheres is the substrate of our sense of self, then olfaction does not take orders from us; it is the reverse.

In comparison to the human neo-cortex, olfaction is considered simple, a primitive affair. The smelling brain is overlooked in the light of such neurological virtuosities as abstract thought, or self-awareness. However, there is something to be said for an apparatus that can do so much with so little. At one point in the development of the organism, this was the height of sophistication. It was at this point where the story of cognition began, for the olfactory system is a brain unto itself.

As it turns out, we know less about this nose-brain than the neo-cortex. The epithelium, the sensory surface upon which aroma compounds are initially received, exhibits no recognizable patterns in relation to specific odors. At the next interchange, the activity observed in the olfactory bulb is almost entirely shrouded in mystery. Let this be a kind of relative indicator: retinal and cochlear prostheses have already been developed, a feat of reverse engineering of the nervous system that has decoded the visual and auditory receptors of the brain. The translation of smells from the tip of the brain’s tendrils in the nose to the percept of a smell in the mind of the subject is incompletely understood. Prosthetic noses will be in speculative fiction plotlines long before national headlines.

Screen Shot 2015-04-22 at 12.05.30 PMBeyond this, the later stages of olfactory perception include the limbic system, which is related to the movement of the body, emotion, and spatialization (among other things). More is known about these modules than the initial olfactory interchanges. However, smell is so deeply embedded into the circuitry of the limbic system that the mystery of olfaction reveals a shortcoming in our understanding. We will never fully understand how a mind creates a self, or even how it moves a body through space, without deciphering the hidden network layers of the olfactory bulb.

Perhaps it should stay hidden. Nonetheless, the odor somehow makes its way to the language centers of the brain, now searching for an appropriate response to communicate the identity of the odor. It is only at this point where olfaction enters the left hemisphere, and explanation of its behavior can come forth (explanations themselves being such a left-brain phenomenon). In eliciting a verbal response to a presented odor, the olfactory system performs a kind of Hamming distance operation of both the episodic and semantic memory data sets, comparatively. The sets are narrowed down until a final word is generated. This word we call Knowledge.

TripartiteBrain_Joe Scordo

A group of children placed together, but with no exposure to language, will eventually create their own. The structure is there. No instructions are required for creating a language, only the capacity for verbal communication and a whole lot of babble. Deep Learning neural networks are a second-generation artificial intelligence which use “hidden layers” of bottom-up information flows. Where top-down instructions say “If you see this, do that, unless it is like this, in which case do that instead”, the bottom-up script says “What do you see?” The program is not trained; it is given a structure by which it can learn by itself. Unsupervised learning, it is formally called in the field of artificial intelligence. Using these “hidden” structures, the functional neural network will tell you what features to consider. The network will organize itself, or rather the data organizes itself, using the available network.

Artificial intelligence is a matter of information-metabolism. Either you give the system nothing in terms of rules, but everything in terms of information, or you give specific rules that work only with specific forms of information. If you are willing to give access to infinite information, it can generate its own rules, for every instance no matter what (in an ideal system). Your work goes into affording information, not to “figuring it out”. The ‘deep learning artificial intelligence’ is a machine, a fulcrum that multiplies force, not by way of fossil fuel, but using a different kind of fossil – our cultural artifacts, be they odysseys of antiquity or pictures of cats.

Brain “Odor Map” or Activity Pattern for Isoamyl Acetate – 4500ppm/V see http://senselab.med.yale.edu/

The olfactory system does not produce cultural artifacts so much as individual ones. All the same, the data sets are big. Trillions upon trillions of molecules, all linked to body-state and autobiographical data that simply cannot be measured, matched against a set of words that can take virtually any form the language can provide and must be subject to the generative properties of language in that a new word can be used that did not previously exist in the lexicon. These are the data sets involved in olfaction. They are big, and they are yet manageable, operable, and capable of producing knowledge about the world around us. The nose knows something we seem to be only now figuring out: with enough data and the right substrate, knowledge creates itself.

Furthermore, the rules that govern this processing do not come pre-determined in the human organism. Smells are learned, not given as fact. Olfactory perception generates its own rules based on the data sets given to it. There exists no such thing as a universal response to any particular odor. At times it may seem so – Menthol is Cool – but this is trigeminally determined. Burning flesh is Bad, but this is culturally- determined. In any case where it seems that an odor response is pre-installed, it will turn out that either the boundaries of the system are ignored (as in the trigeminal discrepancies), the power of cognitive override is underestimated (as in suggestively-induced odor hallucinations), or the experimental group is not large enough (to include both cultures above and below the water table where burning bodies is more practical than burying them, for example). In fact, there are many reasons why a smell may seem to have a quality universal to all humans, but because olfaction is so incompletely understood, both by professionals and the general public alike, this universality is always an illusion, painted by our lack of knowledge on the subject.

Olfaction utilizes unsupervised learning on neural networks of extremely massive and complex data sets to encode and subsequently recognize odor objects. Nevermind reason – to assign such a high degree of intelligence to such a primitive sensory system had to await the development of its artificial analog.






Allen Barkkume is a full-time art teacher and first-time author of Hidden Scents: The Language of Smell in the Age of Approximation.

Hidden Scents http://www.lulu.com/spotlight/hiddenscents