I regularly keep up with artificial intelligence — a subject that fascinates me as much as it shapes my professional life. And while browsing YouTube, I came across the video Smell in the Age of AI: What Machines Know About Perfume (That We Don’t), an interview with the CTO of Osmo, a New York startup that wants to model the sense of smell using AI.
In my field, digital marketing, I already use AI to optimize content, automate tasks, and generate copy. I’m also familiar with its applications in customer relations, software development, and healthcare. But the idea of creating a perfume using AI in a field more readily associated with intuition and craftsmanship surprised me.
Osmo, the startup that wants to digitize our sense of smell
After watching the video, I dug a little deeper to understand who was behind Osmo. The company was founded by Alexander Wiltschko, a former Google Brain researcher, driven by a simple yet mind-boggling idea: to digitize the human sense of smell. Their headquarters are located on the Manhattan waterfront, but their ambition is much broader. Osmo aims to bridge the gap between chemistry, biology, and sensory perception. To achieve this, the startup focuses its efforts on two main areas. First, the development of novel and patented ingredients, called captives, capable of replacing banned or controversial molecules while creating new olfactory notes. Second, the creation of formulation tools for perfumers. These AI-assisted software programs help them manage the stability, performance, and regulatory constraints of their formulas. Osmo doesn’t claim to revolutionize perfume by erasing creators. Its aim is more subtle: to make the technology invisible, yet indispensable, s that perfumers can focus on what they do best — imagining.
How Osmo’s AI models odors from molecules
In the video, Richard Whitcomb, Osmo’s CTO and a deep learning expert formerly with Google, explains the method he uses to tackle the complexity of smell. To predict the odor of a molecule, the startup didn’t attempt to model each biological step—from volatility to binding to olfactory receptors. Instead, it opted for a radically different approach: directly asking people what it smells like, then training a model to learn from those perceptions. Osmo uses a technology particularly well-suited to molecular structures, called graph neural networks (GNNs) . A molecule can be represented as a graph with a, oms as nodes and chemical bonds as edges. Unlike other approaches, GNNs can learn relevant features without human intervention, even from limited datasets—a major advantage in an industry where data remains highly protected. But the most instructive aspect is their approach to the subjectivity of smell . Mo distinguishes between two dimensions: the scent itself (floral, woody, fruity, etc.), which can be described and taught, and preference (whether we like it or not), influenced by culture, context, and experience. This separation allows them to design perfumes that “smell” a certain way while also being tailored to different audiences. Human intuition, thereforremains very much present, but supported by much more precise predictions.
When the perfume industry shifts towards artificial intelligence
Upon discovering Osmo’s model, I wanted to know if this approach was an isolated case or revealed a broader trend. And the answer is clear: AI is already deeply integrated into the perfume industry . The contrast is striking. Where Osmo promises the creation of a personalized sample in 48 hours, the classic formulation process can take 6 to 18 months, between trials, stability tests, and regulatory validation — not to mention the natural aging of certain ingredients like iris or sandalwood. The industry giants are well aware of this. Givaudan has developed Carto, a formulation aid tool. DSM-Firmenich relies on EmotiON, an AI designed to create fragrances that influence mood and well-being. IFF and Symrise, two other leaders, have also integrated models to streamline their development processes. So this isn’t just a startup gimmick. Artificial intelligence is becoming the standard. It optimizes, it accelerates, but above all, it opens up new perspectives — for brands, perfumers, and perhaps soon, consumers themselves.
The new role of the perfumer in the age of AI
In this new era where AI is used daily, the question arises as to the role of the perfumer. According to Richard Whitcomb, perfumers remain at the heart of the creative process. AI is there to manage the technical complexity—formula stability, performance in different substrates (laundry detergent, candles, skin, etc.), and regulatory compliance—to free up time for imagination. This change in process does not mean the end of a profession, but its evolution: the perfumer becomes a “conceptual curator”, capable of exploring new ideas, understanding trends, collaborating with brands to design coherent and original olfactory experiences.
Opportunities, but also controversies
But behind the promises of efficiency and democratization, AI in perfumery also raises criticisms and concerns (just as in other sectors). First, there’s the question of craftsmanship. Some professionals, like Pierre Vouard, a professor at the Fashion Institute of Technology, fear that “hand mixing,” the intuition of the nose, and the slow alchemy of experimentation will be relegated to the background, or even erased, by ultra-fast algorithms. AI can certainly reduce costs, but at what price for the uniqueness of the creations? Furthermore, transparency is also an issue. Osmo, for example, refused to disclose its datasets or the names of the perfumers involved. In an industry built on sensory experience and emotion, this lack of transparency can create a distance between brands and consumers. Another gray area is environmental impact . Osmo claims its models consume little energy, but doesn’t measure their actual consumption and hasn’t shared any life cycle assessment tests. At a time when sustainability is becoming a purchasing criterion, this lack of data raises questions. Finally, ethical lapses are beginning to emerge. Some brands have used the faces or voices of perfumers via deepfakes, without their consent. Others sell perfumes supposedly “created by renowned noses” when they are actually based on AI generators. Industry players, notably the retailer Stéle, are denouncing a loss of trust and a blurring of the lines for the general public.
A fragrant future between art and algorithm
Through my research, I observe that artificial intelligence is now a well-established reality in the perfume industry: AI-powered perfume creation, formulation assistance, generation of new ingredients… But this evolution also raises fundamental questions: about the transparency of methods, environmental impact, the authenticity of creations, and more broadly about the place of craftsmanship and human intuition in this profession. The industry is at a turning point. Its ability to integrate AI without losing what makes perfumery so special will define the next olfactory era. The challenge will be to reconcile technological performance and artistic sensibility, to broaden access to beauty without ever erasing its soul!
