Some of the world’s most coveted perfume ingredients come from species under severe pressure. Agarwood, known as oudh, is produced when certain Aquilaria trees are damaged and generate a dark, fragrant resin. The ingredient is so prized that wild Aquilaria populations have been devastated by illegal felling. This is the kind of supply chain problem that biotechnology and artificial intelligence are now being asked to solve. The question is whether they actually can.
From a Fragment of Bark to a Reconstructed Scent
The core technical claim is striking. Osmo, a spin-off from Google Brain that has raised $130 million to develop AI capable of predicting and recreating smells, says it reconstructed the scent of a rare tree from a piece of bark described as roughly the size of a finger. A spectrometer analysed the sample, the company’s AI rebuilt the underlying chemical formula, and the scent was reproduced in a laboratory setting.
What makes this different from older approaches is not the ability to identify molecules in a scent. Machines have been doing that for a long time. The shift, as Osmo’s founder Alex Wiltschko describes it, is in what happens after the data is collected: the system can now predict how combinations of molecules will be perceived by humans. Osmo says it trained its AI on more than five million human responses to smells, which gives the model a basis for anticipating sensory outcomes rather than simply cataloguing chemical compositions.
BioHarvest Sciences is pursuing a different but related path. The company says it has developed a stable cell culture from an endangered fragrance plant native to south-east Asia, allowing the plant’s valuable compounds to be produced without cultivating or repeatedly harvesting the original organism. The plant has not been publicly named, but BioHarvest has announced an agreement anticipating production of twenty tonnes of the fragrance composition, with limited production potentially beginning in the near future.
The Conservation Claim That Doesn’t Quite Hold
Here is what most coverage of this topic tends to skip over: the companies involved are not primarily conservation organisations, and the technology does not automatically protect the species it draws from.
Wiltschko acknowledged directly that conservation is not an explicit priority for Osmo’s work. The appeal of recreating endangered plant scents is framed as a win for both the environment and commercial brands, but the two motivations are not the same thing, and they do not always point in the same direction.
Conservationists are sceptical. Mandy Aftel, a natural perfumer and founder of the Aftel Archive of Curious Scents, argues that producing a synthetic substitute does not stop demand for the original, and does not address the other pressures driving a species toward decline. Rhian Smith of the Royal Botanic Gardens, Kew, makes a similar point using the Aquilaria tree as an example: AI-assisted alternatives may reduce some harvesting pressure, but wild-sourced material remains highly prized, meaning the incentive to extract it does not disappear simply because a synthetic version exists.
There is also a technical ceiling. A flower’s scent can contain hundreds of molecules, and those molecules shift depending on which part of the plant they come from and the conditions in which it grows. Christina Agapakis, whose work involves reconstructing genes from preserved flowers to produce fragrance compounds, described the result of her own project as “a ghostly fragment of what had been lost.” The living organism, the soil, the climate, the season, the surrounding ecology: none of that is captured in a formula.
What This Moment Actually Reveals About AI and Nature
The perfume industry’s encounter with AI is a useful lens for a broader pattern. Technology is increasingly being applied to problems that are fundamentally ecological, and the results are genuinely impressive at the molecular level while remaining limited at the systems level.
AI can predict how a molecule will smell. It can reconstruct a chemical signature from a tiny sample. It can identify microbes collected from clouds, as the perfumer Future Society did, isolating thirteen microbes that produced fifty-seven scent molecules and using them as the basis for a new fragrance. These are real capabilities, and they expand what is possible in ways that were not available a decade ago.
What AI cannot do is replace the conditions that produce a scent in the first place. It cannot recreate the ecology of a forest, the relationship between a tree and the damage that triggers its resin, or the centuries of evolutionary pressure that shaped a flower’s chemistry. As Agapakis put it, a smell is not simply a formula or a dataset.
This distinction matters beyond perfumery. When AI is applied to natural systems, it tends to excel at capturing and reproducing specific outputs while leaving the underlying system untouched. That is useful. It is not the same as conservation, and treating it as such risks substituting a technological narrative for the harder work of protecting what remains.
In Short
AI is enabling genuine advances in fragrance science: reconstructing scents from tiny samples, predicting how molecules will be perceived, and producing compounds from endangered plants without repeated harvesting. These capabilities are real. What they are not, at least not yet, is a conservation strategy. Synthetic alternatives do not automatically reduce demand for wild-sourced originals, and they do not address the broader pressures on endangered species. The technology augments what perfumers and scientists can do. It does not resolve the ecological problem that made the search for alternatives necessary in the first place.
Based on reporting from The Guardian.