Last year, a paper published in Science made waves with the stunning claim that the human nose can detect a whopping one trillion different odours. But if you feel like your nose can't detect a trillion smells, you may be on to something. It's possible that none of us can.
As a Caltech researcher pointed out last fall, and a new Arizona State University paper asserts today, the data collected in last's year odiferous study does not support this radical claim. Rather, the researchers' interpretation of their data -- and the massive figure they came to -- seem to be the result of flawed mathematical logic. And that's a big problem, because the one trillion odor estimate is already making its way into neuroscience textbooks, misinforming students, researchers, and the public.
"We disagree with several aspects of the 2014 study," said Rick Gerkin of Arizona State University, who, along with Jason Castro of Bates College, authored a rebuttal paper that appears today in the journal eLife. "First, the assertion that humans can discriminate between at least one trillion odours is based on a fragile mathematical framework -- one that's capable of creating nearly any result with small variations in the data or the experiment design. So the result in question could be tens of orders of magnitude -- a factor of one with dozens of zeros after it -- larger or smaller than first reported."
Which is to say, based on the Science paper's own reasoning, the correct number of odours the human nose can distinguish could be as few as ten -- or it could be many, many times more than there are galaxies in the observable universe.
How Many Smells Can A Human Smell?
The Science paper that made such a splash sought to answer a simple question: How many different smells can the average human smell? We have clearly defined boundaries for human vision (390-750 nanometre wavelengths on the electromagnetic spectrum) and hearing (20-20,000 hertz), and we also have a pretty good understanding of the resolution of these senses -- that is, how far apart two colours or frequencies have to be in order for our senses distinguish them. But smell isn't so straightforward.
Image via Liz West / Flickr
Most of the scents we encounter in nature are actually mixtures of tens to hundreds of different, odorous molecules. The "scent" of a rose, for instance is composed of over 275 distinct compounds. The question of how well we can smell, then, might be reframed as this: How different do two scents have to be, in terms of their chemical makeup, for the human nose to tell them apart?
To find out, a team of researchers from The Rockefeller University and Howard Hughes Medical Institute performed a simple experiment. They concocted mixtures of 10, 20, or 30 different fragrant compounds, drawn from a collection of 128 scents (including things like orange, spearmint, and anise). 26 volunteers were enlisted to perform a sniff test. Each volunteer was given three vials, two with identical mixtures, and one with a different mixture, and asked to identify the outlier. If the outlier couldn't be correctly recognised, the mixtures were considered indistinguishable.
Repeating this test hundreds of times with each subject, the researchers found that mixtures containing more than half of the same compounds tended to smell the same, while mixtures that were more than 50% divergent smelled different. Here's where things get fuzzy: From this 50% similarity rule, the researchers extrapolated the total number of combinations possible -- from their pool of 128 different base smells. If humans can distinguish any combinations of 10, 20, or 30 (pulled from 128) that are more than 50% different, that gives our noses at least one trillion distinct odours. By comparison, we can only distinguish a puny 2.3 -- 7.5 million colours.
Where They Went Wrong
If humans can indeed detect one trillion different smells, it would (somewhat confusingly) make the human nose far and away our most sensitive sensory organ. But the rebuttal paper published today argues that the approach used to reach the one trillion figure is mathematically unsound, because the answer you get depends steeply on parameters of the experiment:
Had the experiment enlisted ∼ 100 additional subjects similar to the original ones, the same analysis would have concluded that all possible stimuli are discriminable (i.e., that each of the more than 1029 olfactory stimuli possible in their framework are mutually discriminable). By contrast, if the same experimental data were analysed using moderately more conservative statistical criteria, it would have concluded that there are fewer than 5000 discriminable olfactory stimuli -- no larger than the folk wisdom value that the new estimate purports to replace.
Precise numbers often change when more subjects are enrolled in an experiment, but as Gerkin and Castro point out, they shouldn't change in expectation -- that is, if your estimate can swing wildly in either direction when you add more data, that's a big red flag. By re-analysing the Science paper's own data using the very same methods, but varying the parameters -- things like the number of mixture pairs, the number of subjects, and the threshold for statistical significance -- Gerkin and Castro shows that the correct number of smells the human nose can distinguish could be as few as 4,500 or as many as 10^29. Clearly, something is very wrong.
Gerkin and Castro showed that the number of smells estimated in a 2014 paper published in Science was heavily dependent on the number of participants in the experiment, and on the strictness of a statistical test that paper employed. Via Rick Gerkin and Jason Castro.
According to Gerkin and Castro, the logical flaw lies in the assumption that smell can be broken down into discriminable 'intervals' in the same manner as colour vision. To estimate how many different colours the average human can detect, we only need to know two things: The range of visible wavelengths, and the minimum distance between two colours that our eyes can discriminate. But whereas colour varies progressively along a single dimension (wavelength) the intrinsic dimensionality of smell isn't known. If the human nose can only sense smell along a single dimension (imagine a number line with "cheese" at one end and "poop" at the other, and everything in between smelling like some mixture of cheese and poop), then the true number of detectable smells may be very small indeed.
"Scientists can easily compute the number of discriminable colours because they know the organisation of colour perception," said Gerkin. "For example, think about the colour wheel we learned in elementary school or the red-green-blue colour values that make it possible to display colour on television. For smells, there is no accepted 'smell wheel' just yet. To make one, we must first discover the organisation of olfactory perception. Only then can a principled calculation be used to determine how many unique smells there really are."
Science is an inherently error-fraught business, but mistakes like this remind us of one of the most important ground rules of the process: reproducibility. Just because a finding is flashy, paradigm shattering, and (most dangerous of all) ego-stroking, doesn't make it true. For now, at least, it looks like science may be back to the drawing board on what's proving to be the most elusive of human senses.
But that only means the most fascinating discoveries are still ahead of us.