What An AI Nose Knows

Make:cast #21 – Benjamin Cabé

Like many of us during Covid-19, Benjamin Cabé was baking sourdough bread but he was at home in France. He wondered how he might tell if his dough was done proofing. He began working on an artificial nose that uses physical sensors to detect gas but also uses machine learning to identify the smell. His project is on the cover of Make: Volume 77. In this conversation, we talk about how his nose works, how it knows what it does and the limits of what it knows.

Subscribe to Make:cast on Itunes or Google Podcasts. You can also find it on Spotify, Deezer, Podcast Addict, PodChaser and Spreaker. A list of previous episodes of Make:cast can be found here.

Benjamin Cabé’s project, titled “Second Sense” in Make: V77, is about the build of an artificial nose. It does three things: it uses a gas sensor to detect what’s in the air; it uses a neural network to identify the smell and then it displays a result on a Wio terminal. Oh, and it’s all embedded in a 3D printed enclosure in the shape of a nose. The build is described in detail in the magazine.

The components for the Artificial Nose; its enclosure and final assembly

Our conversation is more about how you know what an artificial nose can and can’t do. We talk about the need for an open catalog of scents (or smells, if you prefer), which might be developed by the community and would help machine-learning applications based on smell become easier to build and scents easier to identify. We also talk about how you might not need to identify what a smell is, just that it’s potentially harmful, like food burning on the stove. In short, there’s a whole world to discover right under your artificial nose.

Benjamin works at Microsoft by day and he’s says he wouldn’t have thought of himself as a maker. But the problem of identifying certain conditions by smell was a tangible, physical thing and it was something to tinker with. He is experienced in computing from his work but this application caused him to jump in and learn machine-learning. He’s still perfecting his bread recipes.