Key Takeaways
- Attempts to incorporate smell into entertainment have existed since the early 1900s, but previous technologies failed to function effectively.
- Osmo, a startup utilizing AI to digitize smell, has developed a predictive model that demonstrates notable accuracy in identifying scents from molecular structures.
- The potential applications of this technology could revolutionize film, virtual reality, and sensory experiences, paving the way for a feasible Smell-O-Vision.
Innovative Advances in Olfactory Technology
Since the early 1900s, the entertainment industry has made various attempts to blend the experience of smells with video content. Notably, the Rivoli Theater in New York City introduced scents during the screening of The Story of Flowers in 1916, while the Rialto Theater installed a smell system in 1933. Additionally, Hans Laube introduced Scentovision at the 1939 World’s Fair. Although there have been more recent efforts, such as a Japanese initiative to create “Smell-O-Vision” for television, these technologies have not gained traction due to their inability to accurately replicate the diversity of smells.
While the fragrance industry has thrived with products designed specifically for consumers, creating a “smell printer” that can emit scents on demand has proven to be a far more complex task. This complexity arises from the necessity of having a comprehensive digital understanding of scent molecules, a goal that has recently become more attainable thanks to advancements in technology.
Osmo, a startup founded in 2023 by Alex Wiltschko, a Harvard-trained expert in olfactory neuroscience, is at the forefront of these advancements, having raised $60 million in funding. With a background in machine learning at Google, Wiltschko aims to “digitize smell,” developing capabilities that enable computers to replicate human olfactory functions.
The company employs artificial intelligence to explore the link between molecular structure and smell perception, showcasing that machines can predict scents with impressive accuracy. Osmo created a machine-learning model using graph neural networks, trained on a dataset of 5,000 known compounds labeled with descriptors like “fruity” and “floral.” This model was further tested on 400 new, structurally distinct compounds, aiming to determine how accurately it could predict their scents in comparison to human panels.
Osmo’s model performed particularly well, even in challenging scenarios, successfully predicting scents 50% of the time for structurally similar molecules that produced differing odors. Furthermore, it was able to generalize its findings to assess olfactory properties, such as odor strength, across a broader database of 500,000 potential scent molecules.
The creation of a Principal Odor Map (POM) by Osmo’s model marks a significant breakthrough in olfactory science, surpassing human experts in predicting the consensus scent for various molecules. This achievement suggests AI’s growing potential to accurately predict smells based on molecular structures, paralleling advances already made in capturing other sensory categories, such as vision.
As technology continues to evolve, the prospect of “teleporting scent” has emerged, allowing for the virtual transfer of smells across distances. Wiltschko envisions a future where local AI sensors can identify the molecular makeup of various scents, enabling Osmo’s odor map to create digital formulas for these aromas without heavy reliance on human scent experts.
This AI-driven capability could finally fulfill the long-held ambition of film and television creators to reproduce odors and scents on a large scale, potentially leading to a future where Smell-O-Vision becomes a reality. The implications of this technology extend beyond entertainment, promising transformative applications in virtual reality, food service, and travel, akin to the innovations seen in recent years with computer and machine vision.
The content above is a summary. For more details, see the source article.