The idea hinges on a concept which has recently come to be known as olfactory white. The main idea behind it is that as additional equal-intensity odors are added to any already sufficiently diverse mixture, the perceived smell of different mixtures tends to become the same. By analogy to other sensory systems, we get the visual perception of the color white when those criteria are met for wavelengths of light, or a white noise hiss when they are met for sound. The difficulty in categorizing smells is that there is no corresponding measure of wavelength for odorants. To give one quick example of this olfactory white phenomenon, consider the ooze which drips from the freshly compacted cargo of a garbage truck as it accelerates away. The familiar garbage scent that wafts from the trail left on the street is uniformly recognizable at once despite the many tens if not hundreds of different garbage-related overtones present in the mixture.
The way that the Varsheny bros. describe the so-called “space of olfactory perception” is that despite the large number of different kinds of olfactory receptors, and much larger variety of possible smell molecules to stimulate them, this space is surprisingly low-dimensional — so low-dimensional they say, that the best way we have to classify mixtures is in terms of pleasantness. In other words, the best we can do for any mixture is to give it a simple thumbs-up or thumbs-down. The brothers developed a large database of different scents and the human psychophysical interpretations of them. They were able to construct a dictionary of compounds and their various chemical properties using the PubChem site. These features were then used to derive cancellation sets for combating offending odors.
In particular, they chose four undesirable odors — durian, onion, katsuobushi (dried bonito), and sauerkraut — which they then attempted to cancel using the same set of counter compounds. The researchers found that they could get fairly good cancellation using 22 compounds, but jumping up to 38 cancellation compounds gave much better results. Matching the perceptual representation for each of these odors (shown above) with the actual chemical constituents of each is a difficult challenge.
At this point in time there is no straightforward method of predicting the pleasantness of a mixture from the pleasantness of each of its components.
IBM’s interest in this field probably runs deeper than the desire to be expert perfumists. Olfaction is still largely unexplored terrain and would present a unique opportunity to showcase the cognitive capabilities of their Watson. Determining what makes a smell good or bad, while fundamental, is only part of the larger challenge. To make a full-electronic nose capable of fully analyzing our olfactory world, detection may still be the most difficult part. As we have seen, even sensitive mass-spectrometers are no match for many flesh and blood noses in the animal kingdom. Knowing what we actually detect and most strongly respond to will undoubtedly be important in any technology attempting to cure our olfactory ills.
