The startup teaching your computer how to smell

Generative AI has proven it can spit out a sound, an image, and written thoughts in mere seconds.

But what about smells?

Osmo — a startup founded by a former Google researcher and backed by Google Ventures — thinks that scent is the next frontier for our computers.

It’s no easy feat:

• An AI model was trained on a dataset of 5k scent molecules from fragrance catalogs to learn to recognize associations between aroma molecules and human descriptions.
• Researchers used that data to develop an odor map that could sort scents into categories, like fruity, musk, meaty, and floral, based on their molecular makeup.
• To test the machine-learning software, they fed it 400 scent molecules that were designed by scientists but didn’t exist and asked the AI model to predict what the aromas would smell like.

To get a sense of the machine’s accuracy, 15 volunteers also smelled and categorized each scent.

Researchers found the AI model’s guesses were better than the average panelist’s, most of the time.

Wait… What do we need this for?

Fair question. Osmo has some really cool potential applications, per Wired:

• The $60B global fragrance market is the most obvious. Osmo’s technology could help fragrance companies clone scents that are otherwise made using problematic processes, like overharvesting endangered plants or harming animals.
• Mosquito-borne diseases, like malaria and dengue fever, kill 440k+ people annually. To find a more effective mosquito repellent, Osmo used its testing process to identify 10+ scents out of 400 with a repellency similar to or greater than today’s gold-standard repellents.
• The company’s long-term goal is to be able to detect diseases through scent.

Ultimately, Osmo could become the “Shazam for smell,” letting users identify and save scents via their smartphones.

Now, whoever wants to get started on letting us taste those food pics on our feeds — we’re waiting.