A Skull Full of Mush

 

There’s an interesting paper to be presented later this month at an Association for Computing Machinery virtual conference on the theme of Interactive Media Experiences.

The paper, by three engineers and a psychologist from the University of Liverpool, is titled “Predicting the colour associated with odours using an electronic nose.” The cross-modal associations between scent and color have been of interest for ages; my colleagues and I helped put them on a quantitative basis with a couple of papers in the late 1990s. The Liverpool group’s new twist is getting an e-nose to predict the color associations.

That’s all fine. It’s of a piece with attempts to predict a molecule’s odor character from physical parameters such as molecular weight, shape, charge, etc. The authors of the new study claim to find a 70 to 81% “machine-human similarity rating.” Whatever. I’m not that interested in the details of their e-nose or experimental protocols, but I’m fascinated by their conceptual point of view as expressed in this sentence fragment in the conclusions section:

Thus, highlighting the possibility of using e-noses to predict human olfactory perception and implying that the colour associated with odours are partly written into the molecular properties of the stimulus [5].

Reference 5 is to the color-odor study—using human noses—that I published with Sarah E. Kemp. In it, we discussed the psychological associations that emerged between the test odors and measurable parameters of color. Nowhere in that paper did we attribute those association to the “molecular properties” of the odor stimuli. It is a mistake for the Liverpool researchers to cite that paper in support of their claim.

The bigger problem is with their philosophical view, namely that an odor’s color associations “are partly written into the molecular properties of the stimulus.” This is a dumb but persistent conceptualization that’s popped up before. Here’s what I said about it in 2009 while commenting on a study by Mandairon et al. called “Humans and mice express similar olfactory preferences.”

What I can’t understand is Mandairon’s endorsement of a mathematical model that predicts odor pleasantness. The idea is that odor pleasantness is “partially dependent on the odorants’ physicohemical properties.” Of course this has to be true at some level: different molecules produce different smells because they have different structures. But Mandairon et al. go beyond tautology; the shared response of mice and men suggests

that olfactory preferences are indeed partially engraved in the structure of the odorant molecule

and

there is an initial part of the [odor] percept which is innate and engraved in the odorant structure.

Perceptions engraved on the molecule? This is simply a bizarre way to think. What else is engraved on a molecule of phenylethanol: Visual associations to red roses? The name of my florist? An olfactory memory of my dead grandmother?

Perceptions happen in the central nervous system of an organism. To talk as if odor pleasantness resides in the structural features of a molecule is animistic thinking, pure and simple.

My objection to Mandairon, et al. applies equally to Ward, et al. No matter how cool the math and the engineering, the idea that odor-linked human perceptions are “written” or “engraved” in the structural features of a molecule is rubbish and unworthy of a place in scientific discourse.

The studies discussed here are “Predicting the colour of odours using an electronic nose,” by Ryan J. Ward, Shammi Rahman, Sophie Wuerger and Alan Marshall, published in SensoryX ‘21: Workshop on Multisensory Experiences, together with IMX 2021: ACM International Conference on Interactive Media Experiences, June 21-23, 2021, and “Humans and mice express similarolfactory preferences,” by Nathalie Mandairon, Johan Poncelet, Mousafa Bensafi and Anne Didier, published in PLoS One 4:e4209, 2009.