Machines can smell

Machines can smell

Odours have a power of persuasion stronger than that of words, appearances, emotions or will.” – Patrick Süskind

Smell is one of the five senses shared by an immense range of living creatures and plays an important role in how they inspect and react to their surroundings. For human beings, our sense of smell is linked to our ability to savour food and can also trigger lively thoughts. Smell allows us to understand all of the scents that occur in our daily lives, whether they are the fragrant roses, freshly baked cookies, or perfume. Despite its significance, smell has not obtained the same level of attention from researchers as having sight and hearing.

The science of olfaction lags behind many other fields. Light, for example, has been acknowledged for centuries. The research revealed that what we perceive as different colours are different wavelengths. A glance at a colour wheel and you get a simple representation of how those wavelengths compare, the longer wavelengths of the warm hues transitioning into the shorter wavelengths of the cold hues. But scent has no such guide.

If wavelengths are the basic elements of light, molecules are the building blocks of smell. When they enter our noses, the molecules interact with receptors that send signals to a part of our brain called the olfactory bulb. Small odorant molecules are the foremost building blocks of flavours and fragrances, and thus represent the simplest version of the odour prediction problem. Yet each molecule can have various odour descriptors. Vanillin, for example, has descriptors like sweet, vanilla, creamy, and chocolate, with some descriptors being more obvious than others. Scientists can observe a wavelength and know what colour it will appear as, but they can’t do the same with molecules and smell.

The researchers first developed a system for a computer to determine chemical features that activate known human odorant receptors. They then selected roughly half a million compounds for new ligands, molecules that bind to receptors, for 34 different receptors. Then they focused on whether the algorithm that could determine odorant receptor activity could also predict distinct perceptual characteristics of odorants. Wiltschko, the leader of the Google research team for the project, says that it is difficult to predict a molecule’s odour by looking at its chemical structure. If a single bond from the structure was removed or changed, we could end up with rotten eggs instead of roses.

The researchers have tutored the algorithm to recognize 10 chemical smells: acetone, acetaldehyde, carbon monoxide, ammonia, butanol, ethylene, methane, methanol, benzene, and toluene. The plan is that in the presence of an odour, the algorithm establishes the scent’s presence and identify exactly which smell it is.

Over the sequence of a few cycles of exposure to each odour, the AI was ultimately able to associate specific receptor patterns with particular smells. The AI correctly recognized eight of the smells 100% of the time, and the remaining two with 90% accuracy.

Further simulations were run in which the target odour was 80% concealed by other odours, the way it often is in the real world. In these experiments, accuracy dropped down to merely 30%, though it is still pretty impressive for such an amateur AI. For further steps, the researchers hope to build a colour wheel for smell using principle chemical detection through the algorithm.

Within the field of machine learning, the smell remains the most baffling of the senses, and researchers are excited to continue with the experiments to shed light on it through further fundamental study. Firstly, it will help companies predict how a new chemical will ultimately smell to individuals – whether it will be soft, disturbing, floral or tart, and so on. Besides, it will give food and fragrance industries the necessary data to interchange particular chemicals should they need to, without altering a product’s odour. Or let’s just say, create bizarre products like a mosquito repellent that smells like coffee. 

The opportunities for future research are numerous and involve everything from inventing new olfactory molecules that are more economical and more sustainably manufactured, to digitizing smell, or even one day giving those with a deprived sense of olfaction access to the smell of roses (and, regrettably, also rotten eggs).